Hi. #next returning nil is definitely looks bad because we will always have case when nil value will not means the end: #(1 nil 2) readSteam next; next; atEnd But I understand that in many cases analysing #next for nil is much more suitable than #atEnd. I am sure that it can be always avoided but it can be not easy to do. Alternatively we can reify real EndOfStream object which will be used as result of reading operation when stream is at end. It can provide nice API like #ifEndOfStream:. which will reflect intention in much cleaner way than ifNil: checks. 2018-04-04 12:00 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>:

...

On 4 Apr 2018, at 11:38, Nicolas Cellier <nicolas.cellier.aka.nice@ gmail.com> wrote:

Hi Sven, See also discussion at https://github.com/OpenSmalltalk/opensmalltalk- vm/pull/232

Thanks Nicolas, I had already seen parts of it.

Now, I still want image level changes to be based on clear semantic definitions of the abstract stream API, we're not just talking about file or unix streams, the general Smalltalk concept.

For reading, they are IMHO,

#next #readInto:startingAt:count: #peek #atEnd #upToEnd (can be derived but still the semantics are important in relation to #atEnd)

For writing, we have

#nextPut: #next:putAll:startingAt: #flush

For both, we have

#atEnd #close #closed (new)

So, I know, #next returning nil exists, but is it universally/officially defined as such ? Where is that documented ?

Positioning, sizing are not universal, IMHO, but should be clearly defined as well.

...

2018-04-04 11:32 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>: Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Now, all kinds of changes are being done image size to work around this.

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

2018-04-04 16:12 GMT+02:00 Eliot Miranda <eliot.miranda@gmail.com>:

Hi Denis, Hi Sven,

On Apr 4, 2018, at 3:16 AM, Denis Kudriashov <dionisiydk@gmail.com> wrote:

Hi.

#next returning nil is definitely looks bad because we will always have case when nil value will not means the end:

#(1 nil 2) readSteam next; next; atEnd

But I understand that in many cases analysing #next for nil is much more suitable than #atEnd. I am sure that it can be always avoided but it can be not easy to do. Alternatively we can reify real EndOfStream object which will be used as result of reading operation when stream is at end. It can provide nice API like #ifEndOfStream:. which will reflect intention in much cleaner way than ifNil: checks.

There is a much better scheme which is to allow streams to have an end-of-file element in an inst var, "endOfStream". It is initialised with the stream itself (nil cannot be used as we want it as a value). If the end of the stream is reached and endOfStream == self an EndIfStream exception is raised. Otherwise the value of endOfStream is answered. We can't answer nil because in some applications (streams of objects) nil is a valid element to read from the steam and we need to manufacture a unique sentinel.

We made this change in VisualWorks and it works well. One can avoid using atEnd and lion for the sentinel instead, which is much cleaner and faster and less problematic than atEnd.

Hi Eliot, Using self is a nice hack for saving an inst. var. :)

If we want to afford an inst.var. then we can also reify an endOfStreamAction inst.var. like I did in ExtendedStream ^endOfStreamAction value We are then free to use Notification/Error, default value (reified EndOfStream object, nil, ...), block with block return... The problem I had is when I started using stream wrappers like in Xtreams. Each wrapper has it's own action, and it's not obvious how we should compose these. Nicolas Sorry to double post, I should have respond-to-all as default action...

2018-04-04 12:00 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>:

...

...

On 4 Apr 2018, at 11:38, Nicolas Cellier <nicolas.cellier.aka.nice@gmai l.com> wrote:

Hi Sven, See also discussion at https://github.com/OpenSmallta lk/opensmalltalk-vm/pull/232

Thanks Nicolas, I had already seen parts of it.

Now, I still want image level changes to be based on clear semantic definitions of the abstract stream API, we're not just talking about file or unix streams, the general Smalltalk concept.

For reading, they are IMHO,

#next #readInto:startingAt:count: #peek #atEnd #upToEnd (can be derived but still the semantics are important in relation to #atEnd)

For writing, we have

#nextPut: #next:putAll:startingAt: #flush

For both, we have

#atEnd #close #closed (new)

So, I know, #next returning nil exists, but is it universally/officially defined as such ? Where is that documented ?

Positioning, sizing are not universal, IMHO, but should be clearly defined as well.

...

2018-04-04 11:32 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>: Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Now, all kinds of changes are being done image size to work around this.

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

Hi Nicolas, On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look. Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare? It would keep the same coding style, but avoid the problems with nil. Thanks, Alistair

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long. [1] Time began when stdio stream support was introduced. :-)

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates. As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error. We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required. Cheers, Alistair

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

Thanks for this discussion. On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

I have trouble understanding your problem analysis, and how your proposed solution, would solve it. Where is it being said that #next and/or #atEnd should be blocking or non-blocking ? How is this related to how EOF is signalled ? It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking. Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible). I have a feeling that too much functionality is being pushed into too small an API.

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse <stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

Alistair, I am replying in-between, please don't take my remarks as being negative, I just like to straighten things out as clear as possible.

On 10 Apr 2018, at 22:14, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Sven,

On 10 April 2018 at 19:36, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

I have trouble understanding your problem analysis, and how your proposed solution, would solve it.

The discussion started with my proposal to modify the Zinc streams to check the return value of #next for nil rather than using #atEnd to iterate over the underlying stream. You correctly pointed out that the intention behind #atEnd is that it can be used to control iterating over streams.

OK

...

Where is it being said that #next and/or #atEnd should be blocking or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file-scripting-language/answe... https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

Thanks, Alistair

...

...

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse <stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

<stdio.cs>

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

Where is it being said that #next and/or #atEnd should be blocking or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file- scripting-language/answer/Philippe-Back-1?share=c19bfc95 https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

...

...

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse <stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <

akgrant0710@gmail.com>:

...

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van

Caekenberghe wrote:

...

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed

which is

...

...

related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

<stdio.cs>

On 11 Apr 2018, at 10:29, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Denis,

On 11 April 2018 at 10:02, Denis Kudriashov <dionisiydk@gmail.com> wrote:

...

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

...

Where is it being said that #next and/or #atEnd should be blocking or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

The code is loaded in squeak using:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/buildspurtr...

for 32 bit images. It loads:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/LoadReader....

which loads package CogTools-Listener in http://source.squeak.org/VMMaker

An image that automatically runs the code and nothing else is created in:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/StartReader...

If you want to run it interactively you can load CogTools-Listener and do something like:

StdioListener new quitOnEof: false; run

What does #quitOnEof: do ? Can the StdioListener code be browsed/viewed online somewhere ?

If you modify #atEnd to block it will result in the "squeak>" input prompt being printed in the terminal after the input has been entered.

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ? ^ self peek isNil ? PS: I liked your runnable example better, I will try it later on. Thx!

The code can be loaded in to Pharo and basically works, but the output tends to be hidden behind the next input prompt because it uses #cr instead of #lf. You can easily modify StdioListener>>initialize to set the line end convention in stdout.

NOTE: It is not intended to be a release quality implementation of a evaluation loop. The whole purpose as I understand it is for it to be as simple as possible to assist in tracking down issues using the VM simulator. It runs minimal code to get to the point of waiting for user input and then allows an expression that causes problems to be entered and traced using the simulator.

Cheers, Alistair

...

...

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file-scripting-language/answe... https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

...

...

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse <stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

<stdio.cs>

Hi Alistair. I don't think anybody is annoyed by you. You are doing really good job. And nice thing that you are super patient to continue :) What I try to understand is why blocking atEnd is bad? Here is code from VMMaker: [stdin atEnd] whileFalse: [| nextChunk | stdout nextPutAll: 'squeak> '; flush. nextChunk := stdin nextChunkNoTag. [nextChunk notEmpty and: [nextChunk first isSeparator]] whileTrue: [nextChunk := nextChunk allButFirst]. Transcript cr; nextPutAll: nextChunk; cr; flush. [stdout print: (Compiler evaluate: nextChunk); cr; flush] on: Error do: [:ex| self logError: ex description inContext: ex signalerContext to: stderr]]. quitOnEof ifTrue: [SourceFiles at: 2 put: nil. Smalltalk snapshot: false andQuit: true] I am not see why it breaks with blocking #atEnd. Can you explain? 2018-04-11 11:41 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

Hi Sven,

Oh dear. I feel as though I'm not getting my concerns across at all well, and I'm pushing hard enough that all I'm going to do is make people annoyed. So let me try to restate the issue one last time before answering your questions directly.

Pharo & Squeak have unwritten rules about stream usage that I suspect have just emerged over time without being designed.

If you want to be able to iterate over any stream, and in particular stdin from a terminal (which, as far as I know, is the outlier that causes all the problems) you have to follow these rules:

1. If the stream is character / byte oriented you have to check for EOF using "stream next == nil". #atEnd can be used, but you'll still have to do the nil check.

2. All other streams have to check for EOF (end of stream) using #atEnd. "stream next == nil" can be used, but you'll still need to test #atEnd to determine whether nil is a value returned by the stream.

If you write code that you want to be able to consume characters, bytes or any other object, you'll have to test both "stream next == nil" and #atEnd.

The rules are the result of the original blue book design being that #atEnd should be used, and then character input from a terminal being added later, but always returning an EOF character (nil) before #atEnd answers correctly.

At the moment, ZnCharacterEncoder uses #atEnd on character / byte streams, so fails for stdin on a terminal.

Back to your questions:

On 11 April 2018 at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 10:29, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Denis,

On 11 April 2018 at 10:02, Denis Kudriashov <dionisiydk@gmail.com>

wrote:

...

...

...

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

...

Where is it being said that #next and/or #atEnd should be blocking

or

...

...

...

...

...

non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

The code is loaded in squeak using:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/ buildspurtrunkreaderimage.sh

for 32 bit images. It loads:

https://github.com/OpenSmalltalk/opensmalltalk- vm/blob/Cog/image/LoadReader.st

which loads package CogTools-Listener in http://source.squeak.org/ VMMaker

An image that automatically runs the code and nothing else is created in:

https://github.com/OpenSmalltalk/opensmalltalk- vm/blob/Cog/image/StartReader.st

If you want to run it interactively you can load CogTools-Listener and do something like:

StdioListener new quitOnEof: false; run

What does #quitOnEof: do ? Can the StdioListener code be browsed/viewed online somewhere ?

I just referenced this as an example of making #atEnd (really FilePlugin>>primitiveFileAtEnd) blocking causing problems. I wasn't expecting people to go and look at the code or use it as a test.

If you really want to look at it (from Pharo):

1. Add http://source.squeak.org/VMMaker as a repository. 2. Browse the CogTools-Listener package

...

...

If you modify #atEnd to block it will result in the "squeak>" input prompt being printed in the terminal after the input has been entered.

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

I meant the primitive, i.e. FilePlugin>>primitiveFileAtEnd / FilePluginPrims>>atEnd:.

...

^ self peek isNil

?

PS: I liked your runnable example better, I will try it later on. Thx!

Right. My code is meant to be minimal and trigger the problem I'm actually focused on - that ZnCharacterEncoder doesn't work with stdin from a terminal.

Sven has expressed a hesitation to change the internal operation of the Zinc streams from using #atEnd to "stream peek == nil" and this whole discussion is really about us trying to resolve our different perspective of the best path forward. I respect Sven and his work so I'm trying to justify the change (but I'm not expressing it at all well, obviously).

Cheers, Alistair

...

...

...

...

...

...

...

...

...

...

...

...

...

...

...

...

Now, all kinds of changes are being done image size to

work

...

...

around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position

...

...

The code can be loaded in to Pharo and basically works, but the output tends to be hidden behind the next input prompt because it uses #cr instead of #lf. You can easily modify StdioListener>>initialize to set the line end convention in stdout.

NOTE: It is not intended to be a release quality implementation of a evaluation loop. The whole purpose as I understand it is for it to be as simple as possible to assist in tracking down issues using the VM simulator. It runs minimal code to get to the point of waiting for user input and then allows an expression that causes problems to be entered and traced using the simulator.

Cheers, Alistair

...

...

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file- scripting-language/answer/Philippe-Back-1?share=c19bfc95 https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

...

...

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse < stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe < sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com

...

...

...

...

...

...

...

...

...

...

wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant < akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the

implementation

...

...

...

of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size

...

...

...

...

...

...

...

...

...

...

...

...

...

...

...

1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

<stdio.cs>

Hi Sven & Dennis, On 11 April 2018 at 12:04, Denis Kudriashov <dionisiydk@gmail.com> wrote:

Hi Alistair.

I don't think anybody is annoyed by you. You are doing really good job. And nice thing that you are super patient to continue :)

On 11 April 2018 at 12:13, Sven Van Caekenberghe <sven@stfx.eu> wrote:

Yes, Alistair, you are a top notch open source contributor !

For me, this discussion is about the difference between looking from low level details/issues/changes up, vs, from a higher level down.

Thanks for your kind words.

What I try to understand is why blocking atEnd is bad? Here is code from VMMaker:

[stdin atEnd] whileFalse: [| nextChunk | stdout nextPutAll: 'squeak> '; flush. nextChunk := stdin nextChunkNoTag. [nextChunk notEmpty and: [nextChunk first isSeparator]] whileTrue: [nextChunk := nextChunk allButFirst]. Transcript cr; nextPutAll: nextChunk; cr; flush. [stdout print: (Compiler evaluate: nextChunk); cr; flush] on: Error do: [:ex| self logError: ex description inContext: ex signalerContext to: stderr]]. quitOnEof ifTrue: [SourceFiles at: 2 put: nil. Smalltalk snapshot: false andQuit: true]

I am not see why it breaks with blocking #atEnd. Can you explain?

First consider the case where #atEnd doesn't block and we just want to evaluate 4+3: 1. #atEnd will return false 2. the loop will print the prompt 3. wait for input (stdin nextChunkNoTag) 4. print the result 5. goto 1. So the screen will look like: squeak> 4+3! 7 squeak> [cursor here] Which is what we expect (prompt, input, result, prompt). If #atEnd is blocking the VM will hang at step 1 until the user enters something in the terminal. In Ubuntu at least terminal input appears to be line buffered, so for the example above the terminal will look like: 4+3! squeak> 7 [cursor here] We don't get the prompt when the program is started, the result is printed after the prompt, and then there's just a cursor sitting at the start of the next line. Obviously the program could be re-written to have the correct output with #atEnd blocking. But I'm arguing that this program is representative of many others, and we don't want to break backward compatibility in this case. Cheers, Alistair

2018-04-11 11:41 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

Oh dear. I feel as though I'm not getting my concerns across at all well, and I'm pushing hard enough that all I'm going to do is make people annoyed. So let me try to restate the issue one last time before answering your questions directly.

Pharo & Squeak have unwritten rules about stream usage that I suspect have just emerged over time without being designed.

If you want to be able to iterate over any stream, and in particular stdin from a terminal (which, as far as I know, is the outlier that causes all the problems) you have to follow these rules:

1. If the stream is character / byte oriented you have to check for EOF using "stream next == nil". #atEnd can be used, but you'll still have to do the nil check.

2. All other streams have to check for EOF (end of stream) using #atEnd. "stream next == nil" can be used, but you'll still need to test #atEnd to determine whether nil is a value returned by the stream.

If you write code that you want to be able to consume characters, bytes or any other object, you'll have to test both "stream next == nil" and #atEnd.

The rules are the result of the original blue book design being that #atEnd should be used, and then character input from a terminal being added later, but always returning an EOF character (nil) before #atEnd answers correctly.

At the moment, ZnCharacterEncoder uses #atEnd on character / byte streams, so fails for stdin on a terminal.

Back to your questions:

On 11 April 2018 at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 10:29, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Denis,

On 11 April 2018 at 10:02, Denis Kudriashov <dionisiydk@gmail.com> wrote:

...

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

...

Where is it being said that #next and/or #atEnd should be blocking or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

The code is loaded in squeak using:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/buildspurtr...

for 32 bit images. It loads:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/LoadReader....

which loads package CogTools-Listener in http://source.squeak.org/VMMaker

An image that automatically runs the code and nothing else is created in:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/StartReader...

If you want to run it interactively you can load CogTools-Listener and do something like:

StdioListener new quitOnEof: false; run

What does #quitOnEof: do ? Can the StdioListener code be browsed/viewed online somewhere ?

I just referenced this as an example of making #atEnd (really FilePlugin>>primitiveFileAtEnd) blocking causing problems. I wasn't expecting people to go and look at the code or use it as a test.

If you really want to look at it (from Pharo):

1. Add http://source.squeak.org/VMMaker as a repository. 2. Browse the CogTools-Listener package

...

...

If you modify #atEnd to block it will result in the "squeak>" input prompt being printed in the terminal after the input has been entered.

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

I meant the primitive, i.e. FilePlugin>>primitiveFileAtEnd / FilePluginPrims>>atEnd:.

...

^ self peek isNil

?

PS: I liked your runnable example better, I will try it later on. Thx!

Right. My code is meant to be minimal and trigger the problem I'm actually focused on - that ZnCharacterEncoder doesn't work with stdin from a terminal.

Sven has expressed a hesitation to change the internal operation of the Zinc streams from using #atEnd to "stream peek == nil" and this whole discussion is really about us trying to resolve our different perspective of the best path forward. I respect Sven and his work so I'm trying to justify the change (but I'm not expressing it at all well, obviously).

Cheers, Alistair

...

...

The code can be loaded in to Pharo and basically works, but the output tends to be hidden behind the next input prompt because it uses #cr instead of #lf. You can easily modify StdioListener>>initialize to set the line end convention in stdout.

NOTE: It is not intended to be a release quality implementation of a evaluation loop. The whole purpose as I understand it is for it to be as simple as possible to assist in tracking down issues using the VM simulator. It runs minimal code to get to the point of waiting for user input and then allows an expression that causes problems to be entered and traced using the simulator.

Cheers, Alistair

...

...

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file-scripting-language/answe... https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

OS pipes are a similar case. On Pharo, you can run CommandShellTestCase to provide some test coverage for this. Dave On Wed, Apr 11, 2018 at 03:13:35PM +0200, Denis Kudriashov wrote:

Thanks for explanation.

I think it would be same scenario for socket stream where #atEnd is not blocking. So I agree that it is expected behaviour.

The example is general enough to expect it to be working for any given pair of in and out streams. So our streams should support this.

2018-04-11 14:56 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven & Dennis,

On 11 April 2018 at 12:04, Denis Kudriashov <dionisiydk@gmail.com> wrote:

...

Hi Alistair.

I don't think anybody is annoyed by you. You are doing really good job. And nice thing that you are super patient to continue :)

On 11 April 2018 at 12:13, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Yes, Alistair, you are a top notch open source contributor !

For me, this discussion is about the difference between looking from low

level details/issues/changes up, vs, from a higher level down.

Thanks for your kind words.

...

What I try to understand is why blocking atEnd is bad? Here is code from VMMaker:

[stdin atEnd] whileFalse: [| nextChunk | stdout nextPutAll: 'squeak> '; flush. nextChunk := stdin nextChunkNoTag. [nextChunk notEmpty and: [nextChunk first isSeparator]] whileTrue: [nextChunk := nextChunk allButFirst]. Transcript cr; nextPutAll: nextChunk; cr; flush. [stdout print: (Compiler evaluate: nextChunk); cr; flush] on: Error do: [:ex| self logError: ex description inContext: ex signalerContext to: stderr]]. quitOnEof ifTrue: [SourceFiles at: 2 put: nil. Smalltalk snapshot: false andQuit: true]

I am not see why it breaks with blocking #atEnd. Can you explain?

First consider the case where #atEnd doesn't block and we just want to evaluate 4+3:

1. #atEnd will return false 2. the loop will print the prompt 3. wait for input (stdin nextChunkNoTag) 4. print the result 5. goto 1.

So the screen will look like:

squeak> 4+3! 7 squeak> [cursor here]

Which is what we expect (prompt, input, result, prompt).

If #atEnd is blocking the VM will hang at step 1 until the user enters something in the terminal. In Ubuntu at least terminal input appears to be line buffered, so for the example above the terminal will look like:

4+3! squeak> 7 [cursor here]

We don't get the prompt when the program is started, the result is printed after the prompt, and then there's just a cursor sitting at the start of the next line.

Obviously the program could be re-written to have the correct output with #atEnd blocking. But I'm arguing that this program is representative of many others, and we don't want to break backward compatibility in this case.

Cheers, Alistair

...

2018-04-11 11:41 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

Oh dear. I feel as though I'm not getting my concerns across at all well, and I'm pushing hard enough that all I'm going to do is make people annoyed. So let me try to restate the issue one last time before answering your questions directly.

Pharo & Squeak have unwritten rules about stream usage that I suspect have just emerged over time without being designed.

If you want to be able to iterate over any stream, and in particular stdin from a terminal (which, as far as I know, is the outlier that causes all the problems) you have to follow these rules:

1. If the stream is character / byte oriented you have to check for EOF using "stream next == nil". #atEnd can be used, but you'll still have to do the nil check.

2. All other streams have to check for EOF (end of stream) using #atEnd. "stream next == nil" can be used, but you'll still need to test #atEnd to determine whether nil is a value returned by the stream.

If you write code that you want to be able to consume characters, bytes or any other object, you'll have to test both "stream next == nil" and #atEnd.

The rules are the result of the original blue book design being that #atEnd should be used, and then character input from a terminal being added later, but always returning an EOF character (nil) before #atEnd answers correctly.

At the moment, ZnCharacterEncoder uses #atEnd on character / byte streams, so fails for stdin on a terminal.

Back to your questions:

On 11 April 2018 at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 10:29, Alistair Grant <akgrant0710@gmail.com>

wrote:

...

...

...

Hi Denis,

On 11 April 2018 at 10:02, Denis Kudriashov <dionisiydk@gmail.com> wrote:

...

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

...

Where is it being said that #next and/or #atEnd should be

blocking

...

...

...

...

...

...

or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

The code is loaded in squeak using:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/ buildspurtrunkreaderimage.sh

for 32 bit images. It loads:

https://github.com/OpenSmalltalk/opensmalltalk- vm/blob/Cog/image/LoadReader.st

which loads package CogTools-Listener in http://source.squeak.org/VMMaker

An image that automatically runs the code and nothing else is created in:

https://github.com/OpenSmalltalk/opensmalltalk- vm/blob/Cog/image/StartReader.st

If you want to run it interactively you can load CogTools-Listener and do something like:

StdioListener new quitOnEof: false; run

What does #quitOnEof: do ? Can the StdioListener code be browsed/viewed online somewhere ?

I just referenced this as an example of making #atEnd (really FilePlugin>>primitiveFileAtEnd) blocking causing problems. I wasn't expecting people to go and look at the code or use it as a test.

If you really want to look at it (from Pharo):

1. Add http://source.squeak.org/VMMaker as a repository. 2. Browse the CogTools-Listener package

...

...

If you modify #atEnd to block it will result in the "squeak>" input prompt being printed in the terminal after the input has been entered.

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

I meant the primitive, i.e. FilePlugin>>primitiveFileAtEnd / FilePluginPrims>>atEnd:.

...

^ self peek isNil

?

PS: I liked your runnable example better, I will try it later on. Thx!

Right. My code is meant to be minimal and trigger the problem I'm actually focused on - that ZnCharacterEncoder doesn't work with stdin from a terminal.

Sven has expressed a hesitation to change the internal operation of the Zinc streams from using #atEnd to "stream peek == nil" and this whole discussion is really about us trying to resolve our different perspective of the best path forward. I respect Sven and his work so I'm trying to justify the change (but I'm not expressing it at all well, obviously).

Cheers, Alistair

...

...

The code can be loaded in to Pharo and basically works, but the output tends to be hidden behind the next input prompt because it uses #cr instead of #lf. You can easily modify StdioListener>>initialize to set the line end convention in stdout.

NOTE: It is not intended to be a release quality implementation of a evaluation loop. The whole purpose as I understand it is for it to be as simple as possible to assist in tracking down issues using the VM simulator. It runs minimal code to get to the point of waiting for user input and then allows an expression that causes problems to be entered and traced using the simulator.

Cheers, Alistair

...

...

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file- scripting-language/answer/Philippe-Back-1?share=c19bfc95 https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

On Wed, Apr 11, 2018 at 12:04 PM, Denis Kudriashov <dionisiydk@gmail.com> wrote:

Hi Alistair.

I don't think anybody is annoyed by you. You are doing really good job. And nice thing that you are super patient to continue :)

+100000

What I try to understand is why blocking atEnd is bad? Here is code from VMMaker:

[stdin atEnd] whileFalse: [| nextChunk | stdout nextPutAll: 'squeak> '; flush. nextChunk := stdin nextChunkNoTag. [nextChunk notEmpty and: [nextChunk first isSeparator]] whileTrue: [nextChunk := nextChunk allButFirst]. Transcript cr; nextPutAll: nextChunk; cr; flush. [stdout print: (Compiler evaluate: nextChunk); cr; flush] on: Error do: [:ex| self logError: ex description inContext: ex signalerContext to: stderr]]. quitOnEof ifTrue: [SourceFiles at: 2 put: nil. Smalltalk snapshot: false andQuit: true]

I am not see why it breaks with blocking #atEnd. Can you explain?

2018-04-11 11:41 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

Oh dear. I feel as though I'm not getting my concerns across at all well, and I'm pushing hard enough that all I'm going to do is make people annoyed. So let me try to restate the issue one last time before answering your questions directly.

Pharo & Squeak have unwritten rules about stream usage that I suspect have just emerged over time without being designed.

If you want to be able to iterate over any stream, and in particular stdin from a terminal (which, as far as I know, is the outlier that causes all the problems) you have to follow these rules:

1. If the stream is character / byte oriented you have to check for EOF using "stream next == nil". #atEnd can be used, but you'll still have to do the nil check.

2. All other streams have to check for EOF (end of stream) using #atEnd. "stream next == nil" can be used, but you'll still need to test #atEnd to determine whether nil is a value returned by the stream.

If you write code that you want to be able to consume characters, bytes or any other object, you'll have to test both "stream next == nil" and #atEnd.

The rules are the result of the original blue book design being that #atEnd should be used, and then character input from a terminal being added later, but always returning an EOF character (nil) before #atEnd answers correctly.

At the moment, ZnCharacterEncoder uses #atEnd on character / byte streams, so fails for stdin on a terminal.

Back to your questions:

On 11 April 2018 at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 10:29, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Denis,

On 11 April 2018 at 10:02, Denis Kudriashov <dionisiydk@gmail.com> wrote:

...

2018-04-11 8:32 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

...

...

...

Where is it being said that #next and/or #atEnd should be blocking or non-blocking ?

There is existing code that assumes that #atEnd is non-blocking and that #next is allowed block. I believe that we should keep those conditions.

I fail to see where that is written down, either way. Can you point me to comments stating that, I would really like to know ?

I'm not aware of it being written down, just that ever existing implementation I'm aware of behaves this way.

On the other hand, making #atEnd blocking breaks Eliot's REPL sample (in Squeak).

Could you write here this example, please?

The code is loaded in squeak using:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/buildspurtr...

for 32 bit images. It loads:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/LoadReader....

which loads package CogTools-Listener in http://source.squeak.org/VMMaker

An image that automatically runs the code and nothing else is created in:

https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/Cog/image/StartReader...

If you want to run it interactively you can load CogTools-Listener and do something like:

StdioListener new quitOnEof: false; run

What does #quitOnEof: do ? Can the StdioListener code be browsed/viewed online somewhere ?

I just referenced this as an example of making #atEnd (really FilePlugin>>primitiveFileAtEnd) blocking causing problems. I wasn't expecting people to go and look at the code or use it as a test.

If you really want to look at it (from Pharo):

1. Add http://source.squeak.org/VMMaker as a repository. 2. Browse the CogTools-Listener package

...

...

If you modify #atEnd to block it will result in the "squeak>" input prompt being printed in the terminal after the input has been entered.

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

I meant the primitive, i.e. FilePlugin>>primitiveFileAtEnd / FilePluginPrims>>atEnd:.

...

^ self peek isNil

?

PS: I liked your runnable example better, I will try it later on. Thx!

Right. My code is meant to be minimal and trigger the problem I'm actually focused on - that ZnCharacterEncoder doesn't work with stdin from a terminal.

Sven has expressed a hesitation to change the internal operation of the Zinc streams from using #atEnd to "stream peek == nil" and this whole discussion is really about us trying to resolve our different perspective of the best path forward. I respect Sven and his work so I'm trying to justify the change (but I'm not expressing it at all well, obviously).

Cheers, Alistair

...

...

The code can be loaded in to Pharo and basically works, but the output tends to be hidden behind the next input prompt because it uses #cr instead of #lf. You can easily modify StdioListener>>initialize to set the line end convention in stdout.

NOTE: It is not intended to be a release quality implementation of a evaluation loop. The whole purpose as I understand it is for it to be as simple as possible to assist in tracking down issues using the VM simulator. It runs minimal code to get to the point of waiting for user input and then allows an expression that causes problems to be entered and traced using the simulator.

Cheers, Alistair

...

...

...

...

...

How is this related to how EOF is signalled ?

Because, combined with terminal EOF not being known until the user explicitly flags it (with Ctrl-D) it means that #atEnd can't be used for iterating over input from stdin connected to a terminal.

This seems to me like an exception that only holds for one particular stream in one particular scenario (interactive stdin). I might be wrong.

...

...

It seems to me that there are quite a few classes of streams that are 'special' in the sense that #next could be blocking and/or #atEnd could be unclear - socket/network streams, serial streams, maybe stdio (interactive or not). Without a message like #isDataAvailable you cannot handle those without blocking.

Right. I think this is a distraction (I was trying to explain some details, but it's causing more confusion instead of helping).

The important point is that #atEnd doesn't work for iterating over streams with terminal input

Maybe you should also point to the actual code that fails. I mean you showed a partial stack trace, but not how you got there, precisely. How does the application reading from an interactive stdin do to get into trouble ?

Included below.

...

...

...

Reading from stdin seems like a very rare case for a Smalltalk system (not that it should not be possible).

There's been quite a bit of discussion and several projects recently related to using pharo for scripting, so it may become more common. E.g.

https://www.quora.com/Can-Smalltalk-be-a-batch-file-scripting-language/answe... https://github.com/rajula96reddy/pharo-cli

Still, it is not common at all.

...

...

I have a feeling that too much functionality is being pushed into too small an API.

This is just about how should Zinc streams be iterating over the underlying streams. You didn't like checking the result of #next for nil since it isn't general, correctly pointing out that nil is a valid value for non-byte oriented streams. But #atEnd doesn't work for stdin from a terminal.

At this point I think there are three options:

1. Modify Zinc to check the return value of #next instead of using #atEnd.

This is what all existing character / byte oriented streams in Squeak and Pharo do. At that point the Zinc streams can be used on all file / stdio input and output.

I agree that such code exists in many places, but there is lots of stream reading that does not check for nils.

Right. Streams can be categorised in many ways, but for this discussion I think streams are broken in to two types:

1) Byte / Character oriented 2) All others

For historical reasons, byte / character oriented streams need to check for EOF by using "stream next == nil" and all other streams should use #atEnd.

This avoids the "nil being part of the domain" issue that was discussed earlier in the thread.

...

...

2. Modify all streams to signal EOF in some other way, i.e. a sentinel or notification / exception.

This is what we were discussing below. But it is a decent chunk of work with significant impact on the existing code base.

Agreed. This would be a future extension.

...

3. Require anyone who wants to read from stdin to code around Zinc's inability to handle terminal input.

I'd prefer to avoid this option if possible.

See higher for a more concrete usage example request.

testAtEnd.st -- | ch stream string stdin |

'stdio.cs' asFileReference fileIn. "stdin := FileStream stdin." stdin := ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). stream := (String new: 100) writeStream. ch := stdin next. [ ch == nil ] whileFalse: [ stream nextPut: ch. ch := stdin next. ]. string := stream contents. FileStream stdout nextPutAll: string; lf; nextPutAll: 'Characters read: '; nextPutAll: string size asString; lf. Smalltalk snapshot: false andQuit: true. --

Execute with:

./pharo --headless Pharo7.0-64bit-e76f1a2.image testAtEnd.st

and type Ctrl-D gives:

'Errors in script loaded from testAtEnd.st' MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt OpalCompiler>>evaluate

Using #atEnd to control the loop instead of "stdin next == nil" produces the same result.

Replacing stdin with FileStream stdin makes the script work.

stdio.cs fixes a bug in StdioStream which really isn't part of this discussion (PR to be submitted).

Cheers, Alistair

...

...

Does that clarify the situation?

Yes, it helps. Thanks. But questions remain.

...

Thanks, Alistair

...

...

On 10 Apr 2018, at 18:30, Alistair Grant <akgrant0710@gmail.com> wrote:

First a quick update:

After doing some work on primitiveFileAtEnd, #atEnd now answers correctly for files that don't report their size correctly, e.g. /dev/urandom and /proc/cpuinfo, whether the files are opened directly or redirected through stdin.

However determining whether stdin from a terminal has reached the end of file can't be done without making #atEnd blocking since we have to wait for the user to flag the end of file, e.g. by typing Ctrl-D. And #atEnd is assumed to be non-blocking.

So currently using ZnCharacterReadStream with stdin from a terminal will result in a stack dump similar to:

MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next UndefinedObject>>DoIt

Going back through the various suggestions that have been made regarding using a sentinel object vs. raising a notification / exception, my (still to be polished) suggestion is to:

1. Add an endOfStream instance variable 2. When the end of the stream is reached answer the value of the instance variable (i.e. the result of sending #value to the variable). 3. The initial default value would be a block that raises a Deprecation warning and then returns nil. This would allow existing code to function for a changeover period. 4. At the end of the deprecation period the default value would be changed to a unique sentinel object which would answer itself as its #value.

At any time users of the stream can set their own sentinel, including a block that raises an exception.

Cheers, Alistair

On 4 April 2018 at 19:24, Stephane Ducasse <stepharo.self@gmail.com> wrote:

...

Thanks for this discussion.

On Wed, Apr 4, 2018 at 1:37 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

First off, thanks for the discussions and your contributions, I really appreciate them.

But I want to have a discussion at the high level of the definition and semantics of the stream API in Pharo.

...

On 4 Apr 2018, at 13:20, Alistair Grant <akgrant0710@gmail.com> wrote:

On 4 April 2018 at 12:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Playing a bit devil's advocate, the idea is that, in general,

[ stream atEnd] whileFalse: [ stream next. "..." ].

is no longer allowed ?

It hasn't been allowed "forever" [1]. It's just been misused for almost as long.

[1] Time began when stdio stream support was introduced. :-)

I am still not convinced. Another way to put it would be that the old #atEnd or #upToEnd do not make sense for these streams and some new loop is needed, based on a new test (it exists for socket streams already).

[ stream isDataAvailable ] whileTrue: [ stream next ]

...

...

And you want to replace it with

[ stream next ifNil: [ false ] ifNotNil: [ :x | "..." true ] whileTrue.

That is a pretty big change, no ?

That's the way quite a bit of code already operates.

As Denis pointed out, it's obviously problematic in the general sense, since nil can be embedded in non-byte oriented streams. I suspect that in practice not many people write code that reads streams from both byte oriented and non-byte oriented streams.

Maybe yes, maybe no. As Denis' example shows there is a clear definition problem.

And I do use streams of byte arrays or strings all the time, this is really important. I want my parsers to work on all kinds of streams.

...

...

I think/feel like a proper EOF exception would be better, more correct.

[ [ stream next. "..." true ] on: EOF do: [ false ] ] whileTrue.

I agree, but the email thread Nicolas pointed to raises some performance questions about this approach. It should be straightforward to do a basic performance comparison which I'll get around to if other objections aren't raised.

Reading in bigger blocks, using #readInto:startingAt:count: (which is basically Unix's (2) Read sys call), would solve performance problems, I think.

...

...

Will we throw away #atEnd then ? Do we need it if we cannot use it ?

Unix file i/o returns EOF if the end of file has been reach OR if an error occurs. You should still check #atEnd after reading past the end of the file to make sure no error occurred. Another part of the primitive change I'm proposing is to return additional information about what went wrong in the event of an error.

I am sorry, but this kind of semantics (the OR) is way too complex at the general image level, it is too specific and based on certain underlying implementation details.

Sven

...

We could modify the read primitive so that it fails if an error has occurred, and then #atEnd wouldn't be required.

Cheers, Alistair

...

...

On 4 Apr 2018, at 12:41, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Nicolas,

On 4 April 2018 at 12:36, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

...

2018-04-04 12:18 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On Wed, Apr 04, 2018 at 11:32:02AM +0200, Sven Van Caekenberghe wrote:

...

Somehow, somewhere there was a change to the implementation of the primitive called by some streams' #atEnd.

That's a proposed change by me, but it hasn't been integrated yet. So the discussion below should apply to the current stable vm (from August last year).

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size

And, as you say, plenty of virtual unix files report size 0.

...

Now, all kinds of changes are being done image size to work around this.

I would phrase this slightly differently :-)

Some code does the right thing, while other code doesn't. E.g.:

MultiByteFileStream>>upToEnd is good, while FileStream>>contents is incorrect

...

I am a strong believer in simple, real (i.e. infinite) streams, but I am not sure we are doing the right thing here.

Point is, I am not sure #next returning nil is official and universal.

Consider the comments:

Stream>>#next "Answer the next object accessible by the receiver."

ReadStream>>#next "Primitive. Answer the next object in the Stream represented by the receiver. Fail if the collection of this stream is not an Array or a String. Fail if the stream is positioned at its end, or if the position is out of bounds in the collection. Optional. See Object documentation whatIsAPrimitive."

Note how there is no talk about returning nil !

I think we should discuss about this first.

Was the low level change really correct and the right thing to do ?

The primitive change proposed doesn't affect this discussion. It will mean that #atEnd returns false (correctly) sometimes, while currently it returns true (incorrectly). The end result is still incorrect, e.g. #contents returns an empty string for /proc/cpuinfo.

You're correct about no mention of nil, but we have:

FileStream>>next

(position >= readLimit and: [self atEnd]) ifTrue: [^nil] ifFalse: [^collection at: (position := position + 1)]

which has been around for a long time (I suspect, before Pharo existed).

Having said that, I think that raising an exception is a better solution, but it is a much, much bigger change than the one I proposed in https://github.com/pharo-project/pharo/pull/1180.

Cheers, Alistair

Hi, yes, if you are after universal behavior englobing Unix streams, the Exception might be the best way. Because on special stream you can't allways say in advance, you have to try. That's the solution adopted by authors of Xtreams. But there is a runtime penalty associated to it.

The penalty once was so high that my proposal to generalize EndOfStream usage was rejected a few years ago by AndreaRaab. http://forum.world.st/EndOfStream-unused-td68806.html

Thanks for this, I'll definitely take a look.

Do you have a sense of how Denis' suggestion of using an EndOfStream object would compare?

It would keep the same coding style, but avoid the problems with nil.

Thanks, Alistair

...

I have regularly benched Xtreams, but stopped a few years ago. Maybe i can excavate and pass on newer VM.

In the mean time, i had experimented a programmable end of stream behavior (via a block, or any other valuable) http://www.squeaksource.com/XTream.htm so as to reconcile performance and universality, but it was a source of complexification at implementation side.

Nicolas

...

...

Note also that a Guille introduced something new, #closed which is related to the difference between having no more elements (maybe right now, like an open network stream) and never ever being able to produce more data.

Sven

<stdio.cs>

On 11 April 2018 at 15:11, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

On 11 Apr 2018, at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

^ self peek isNil

?

Still the same question, how do you implement a blocking #atEnd for stdin ?

I have seen your stdio.cs which is indeed needed as the current StdioStream>>#atEnd is bogus for sure.

But that is still a non-blocking one, right ?

Since there is a peekBuffer in StdioStream, why can't that be used ?

I think you've created a chicken-and-egg problem with this question, but ignoring that for now: StdioStream>>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil. " self atEnd ifTrue: [^ nil ]. peekBuffer ifNotNil: [ ^ peekBuffer ]. ^ peekBuffer := self next. So when we first start the program, i.e. the user hasn't entered any input yet, and #peek is called: 1. #atEnd returns false because Ctrl-D (or similar) hasn't been entered (assuming it is non-blocking). 2. peekBuffer is nil because we haven't previously called #peek. 3. The system now blocks on "self next". Just a reminder: for terminal input the end-of-file isn't reached until the user explicitly enters the end of file key (Ctrl-D). So, if there is no buffered input (either none has been entered yet, or all input has been consumed) #atEnd (after the patch) calls #primAtEnd:. At the moment, #primAtEnd: ends up calling the libc function feof(), which is non-blocking and answers the end-of-file flag for the FILE*. Since the user hasn't entered Ctrl-D, that's false. If we want to control iteration over the stream and ensure that we don't need to do a "stream next == nil" check, then #primAtEnd: is going to have to peek for the next character, and that means waiting for the user to enter that character. In c that is typically done using: atEnd = ungetc(fgetc(fp), fp); and fgetc() will block until the user enters something.

I have run your example testAtEnd.st now, and it works/fails as advertised.

:-) Cheers, Alistair

On 11 Apr 2018, at 16:36, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I can make your example, using the Zn variants, work with the following change:

StdioStream>>#atEnd ^ peekBuffer isNil or: [ (peekBuffer := self next) isNil ]

Argh, make that atEnd ^ peekBuffer isNil and: [ (peekBuffer := self next) isNil ] but I am still testing, this is probably not the final answer/solution.

Which is a literal implementation of your statement that you can only know that you are atEnd by reading (and thus waiting/blocking) and checking for nil, which seems logical to me, given the fact that you *are* waiting for user input.

BTW, at least on macOS you have to enter ctrl-D (^D) on a separate line, I am not sure how relevant that is, but that is probably another argument that stdin is special (being line-buffered by the OS, EOF needing to be on a separate line).

And FWIW, I have been writing networking code in Pharo for years, and I have never had issues with unclear semantics of these primitives (#atEnd, #next, #peek) on network streams, either the classic SocketStream or the Zdc* streams (TLS or not). That is why I think we have to be careful.

That being said, it is important to continue this discussion, I find it very interesting. I am trying to write some test code using stdin myself, to better understand the topic.

...

On 11 Apr 2018, at 16:06, Alistair Grant <akgrant0710@gmail.com> wrote:

On 11 April 2018 at 15:11, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

^ self peek isNil

?

Still the same question, how do you implement a blocking #atEnd for stdin ?

I have seen your stdio.cs which is indeed needed as the current StdioStream>>#atEnd is bogus for sure.

But that is still a non-blocking one, right ?

Since there is a peekBuffer in StdioStream, why can't that be used ?

I think you've created a chicken-and-egg problem with this question, but ignoring that for now:

StdioStream>>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil. "

self atEnd ifTrue: [^ nil ].

peekBuffer ifNotNil: [ ^ peekBuffer ].

^ peekBuffer := self next.

So when we first start the program, i.e. the user hasn't entered any input yet, and #peek is called:

1. #atEnd returns false because Ctrl-D (or similar) hasn't been entered (assuming it is non-blocking). 2. peekBuffer is nil because we haven't previously called #peek. 3. The system now blocks on "self next".

Just a reminder: for terminal input the end-of-file isn't reached until the user explicitly enters the end of file key (Ctrl-D).

So, if there is no buffered input (either none has been entered yet, or all input has been consumed)

#atEnd (after the patch) calls #primAtEnd:.

At the moment, #primAtEnd: ends up calling the libc function feof(), which is non-blocking and answers the end-of-file flag for the FILE*. Since the user hasn't entered Ctrl-D, that's false.

If we want to control iteration over the stream and ensure that we don't need to do a "stream next == nil" check, then #primAtEnd: is going to have to peek for the next character, and that means waiting for the user to enter that character.

In c that is typically done using:

atEnd = ungetc(fgetc(fp), fp);

and fgetc() will block until the user enters something.

...

I have run your example testAtEnd.st now, and it works/fails as advertised.

:-)

Cheers, Alistair

On 11 Apr 2018, at 17:16, Denis Kudriashov <dionisiydk@gmail.com> wrote:

2018-04-11 17:02 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>:

...

On 11 Apr 2018, at 16:36, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I can make your example, using the Zn variants, work with the following change:

StdioStream>>#atEnd ^ peekBuffer isNil or: [ (peekBuffer := self next) isNil ]

Argh, make that

atEnd ^ peekBuffer isNil and: [ (peekBuffer := self next) isNil ]

But discussion exactly about "self next isNil": how to avoid it.

I know, but like this it could/might become an implementation detail. The more things that I try, the more that I feel that stdin is so special that it does not fit in the rest of the stream zoo.

but I am still testing, this is probably not the final answer/solution.

...

Which is a literal implementation of your statement that you can only know that you are atEnd by reading (and thus waiting/blocking) and checking for nil, which seems logical to me, given the fact that you *are* waiting for user input.

BTW, at least on macOS you have to enter ctrl-D (^D) on a separate line, I am not sure how relevant that is, but that is probably another argument that stdin is special (being line-buffered by the OS, EOF needing to be on a separate line).

And FWIW, I have been writing networking code in Pharo for years, and I have never had issues with unclear semantics of these primitives (#atEnd, #next, #peek) on network streams, either the classic SocketStream or the Zdc* streams (TLS or not). That is why I think we have to be careful.

That being said, it is important to continue this discussion, I find it very interesting. I am trying to write some test code using stdin myself, to better understand the topic.

...

On 11 Apr 2018, at 16:06, Alistair Grant <akgrant0710@gmail.com> wrote:

On 11 April 2018 at 15:11, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

^ self peek isNil

?

Still the same question, how do you implement a blocking #atEnd for stdin ?

I have seen your stdio.cs which is indeed needed as the current StdioStream>>#atEnd is bogus for sure.

But that is still a non-blocking one, right ?

Since there is a peekBuffer in StdioStream, why can't that be used ?

I think you've created a chicken-and-egg problem with this question, but ignoring that for now:

StdioStream>>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil. "

self atEnd ifTrue: [^ nil ].

peekBuffer ifNotNil: [ ^ peekBuffer ].

^ peekBuffer := self next.

So when we first start the program, i.e. the user hasn't entered any input yet, and #peek is called:

1. #atEnd returns false because Ctrl-D (or similar) hasn't been entered (assuming it is non-blocking). 2. peekBuffer is nil because we haven't previously called #peek. 3. The system now blocks on "self next".

Just a reminder: for terminal input the end-of-file isn't reached until the user explicitly enters the end of file key (Ctrl-D).

So, if there is no buffered input (either none has been entered yet, or all input has been consumed)

#atEnd (after the patch) calls #primAtEnd:.

At the moment, #primAtEnd: ends up calling the libc function feof(), which is non-blocking and answers the end-of-file flag for the FILE*. Since the user hasn't entered Ctrl-D, that's false.

If we want to control iteration over the stream and ensure that we don't need to do a "stream next == nil" check, then #primAtEnd: is going to have to peek for the next character, and that means waiting for the user to enter that character.

In c that is typically done using:

atEnd = ungetc(fgetc(fp), fp);

and fgetc() will block until the user enters something.

...

I have run your example testAtEnd.st now, and it works/fails as advertised.

:-)

Cheers, Alistair

Hi Sven, On 11 April 2018 at 17:33, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I think we have to reset this whole discussion.

FileStream stdin

and

Stdio stdin

are completely different !

We'll have to check that first, before talking about the issues raised in this thread.

Are you sure you're comparing (roughly) equal things? I would compare: FileStream stdin (very roughly) to: ZnCharacterReadStream on: (ZnBufferedReadStream on: Stdio stdin). Actually it is still not a fair comparison because MultiByteFileStream attempts to be writable as well. If you could notionally do: FileStream stdin readOnly; binary; unbuffered; yourself. you could compare FileStream and Stdio :-) One important similarity: At the bottom they both use the same set of primitives to communicate with the OS stdio streams (FilePlugin). HTH, Alistair

And BTW these terminal streams are a real pain to test ;-)

Hi Sven, On 11 April 2018 at 18:53, Sven Van Caekenberghe <sven@stfx.eu> wrote:

Something is off (and/or I am getting crazy, probably both).

$ ./pharo --headless Pharo.image eval '(FileStream stdin binary; next: 3)' 123 #[49]

??

This should return #[49 50 51] AFAIK.

I haven't checked properly, but I can take a guess at this... The file read primitive in FilePlugin tries to read from stdin 1 character at a time (presumably to get around the line buffering that is done, but I haven't gone through the history. At one time I think it modified the terminal settings, but that code is #if'd out). For all other streams it attempts to read the requested number of characters. StandardFileStream>>readInto:startingAt:count: assumes that primitiveFileRead will always attempt to read count bytes, but it actually only attempts to read 1. Cheers, Alistair

Alistair,

On 11 Apr 2018, at 19:42, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I will send you some code later on.

Today I arranged for my NeoConsole code (that normally works over a network connection) to work over stdio. Although I am not yet happy with every aspect of the implementation, it does work (using unaltered Zn streams and code). The foll $ cat /etc/issue Ubuntu 16.04.4 LTS $ mkdir pharo7 $ cd pharo7/ $ curl get.pharo.org/70+vm | bash % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 3036 100 3036 0 0 36799 0 --:--:-- --:--:-- --:--:-- 37024 Downloading the latest 70 Image: http://files.pharo.org/get-files/70/pharo.zip Pharo.image Downloading the latest pharoVM: http://files.pharo.org/get-files/70/pharo-linux-stable.zip pharo-vm/pharo Creating starter scripts pharo and pharo-ui On a 64-bit system? You must enable and install the 32-bit libraries Please see http://pharo.org/gnu-linux-installation for detailed instructions $ ./pharo Pharo.image config http://mc.stfx.eu/Neo ConfigurationOfNeoConsole --install=bleedingEdge 'Installing ConfigurationOfNeoConsole bleedingEdge' Loading 1-baseline of ConfigurationOfNeoConsole... Fetched -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- http://mc.stfx.eu/Neo Loaded -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- cache ...finished 1-baseline $ ./pharo Pharo.image eval NeoConsoleStdio run Neo Console Pharo-7.0+alpha.build.760.sha.d2734dcabda799803c307365bcd120f92211d34a (32 Bit) pharo> 1+2 3 pharo> 42 factorial 1405006117752879898543142606244511569936384000000000 pharo> Stdio stdin StdioStream: #stdin pharo> == self: StdioStream: #stdin class: StdioStream file: a File handle: #[148 213 25 107 160 197 105 247 0 0 0 0 0 0 0 0 0 1 255 1] forWrite: false peekBuffer: nil pharo> show StdioStream>>#atEnd StdioStream>>#atEnd atEnd ^ file atEnd pharo> get process.list Morphic UI Process Delay Scheduling Process Low Space Watcher Input Event Fetcher Process Idle Process WeakArray Finalization Process CommandLine handler process pharo> quit Bye! a NeoConsoleStdio I know there are many approaches to a REPL, I don't claim mine is best, it is just the one that I have been using for years. In the above, I do not depend on EOF - just to be clear. The point being that there is no immediate fundamental problem. But there is something wrong with what is returned by Stdio stdin Sven

On 11 Apr 2018, at 21:44, Stephane Ducasse <stepharo.self@gmail.com> wrote:

I did not know about the NeoConsole. Nice because I wanted to build a little REPL for my minilanguage implementation.

You are of course welcome to look at it. But it is Pharo specific. I use it to be able to hook/look into running headless server images. For this it is super handy.

Stef

On Wed, Apr 11, 2018 at 8:47 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

...

On 11 Apr 2018, at 19:42, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I will send you some code later on.

Today I arranged for my NeoConsole code (that normally works over a network connection) to work over stdio. Although I am not yet happy with every aspect of the implementation, it does work (using unaltered Zn streams and code). The foll

$ cat /etc/issue Ubuntu 16.04.4 LTS $ mkdir pharo7 $ cd pharo7/ $ curl get.pharo.org/70+vm | bash % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 3036 100 3036 0 0 36799 0 --:--:-- --:--:-- --:--:-- 37024 Downloading the latest 70 Image: http://files.pharo.org/get-files/70/pharo.zip Pharo.image Downloading the latest pharoVM: http://files.pharo.org/get-files/70/pharo-linux-stable.zip pharo-vm/pharo Creating starter scripts pharo and pharo-ui On a 64-bit system? You must enable and install the 32-bit libraries Please see http://pharo.org/gnu-linux-installation for detailed instructions $ ./pharo Pharo.image config http://mc.stfx.eu/Neo ConfigurationOfNeoConsole --install=bleedingEdge 'Installing ConfigurationOfNeoConsole bleedingEdge'

Loading 1-baseline of ConfigurationOfNeoConsole... Fetched -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- http://mc.stfx.eu/Neo Loaded -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- cache ...finished 1-baseline $ ./pharo Pharo.image eval NeoConsoleStdio run Neo Console Pharo-7.0+alpha.build.760.sha.d2734dcabda799803c307365bcd120f92211d34a (32 Bit) pharo> 1+2

3 pharo> 42 factorial

1405006117752879898543142606244511569936384000000000 pharo> Stdio stdin

StdioStream: #stdin pharo> == self: StdioStream: #stdin class: StdioStream file: a File handle: #[148 213 25 107 160 197 105 247 0 0 0 0 0 0 0 0 0 1 255 1] forWrite: false peekBuffer: nil pharo> show StdioStream>>#atEnd StdioStream>>#atEnd atEnd

^ file atEnd pharo> get process.list Morphic UI Process Delay Scheduling Process Low Space Watcher Input Event Fetcher Process Idle Process WeakArray Finalization Process CommandLine handler process pharo> quit Bye! a NeoConsoleStdio

I know there are many approaches to a REPL, I don't claim mine is best, it is just the one that I have been using for years.

In the above, I do not depend on EOF - just to be clear. The point being that there is no immediate fundamental problem.

But there is something wrong with what is returned by Stdio stdin

Sven

On 12 Apr 2018, at 03:15, Eliot Miranda <eliot.miranda@gmail.com> wrote:

Hi Sven,

On Wed, Apr 11, 2018 at 1:25 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

On 11 Apr 2018, at 21:44, Stephane Ducasse <stepharo.self@gmail.com> wrote:

I did not know about the NeoConsole. Nice because I wanted to build a little REPL for my minilanguage implementation.

You are of course welcome to look at it. But it is Pharo specific. I use it to be able to hook/look into running headless server images. For this it is super handy.

Cool usage! Could you tell me whether you type Smalltalk expressions into this to examine your running server? The answer will be used in a related discussion on a mailing list not too distant from this one ;-)

Yes of course, that it the whole purpose, to type in expressions and to manipulate objects in a running image. I recently added options to look at code and change/add methods. All very primitive, but when in trouble, it works well.

...

Stef

On Wed, Apr 11, 2018 at 8:47 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

...

On 11 Apr 2018, at 19:42, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I will send you some code later on.

Today I arranged for my NeoConsole code (that normally works over a network connection) to work over stdio. Although I am not yet happy with every aspect of the implementation, it does work (using unaltered Zn streams and code). The foll

$ cat /etc/issue Ubuntu 16.04.4 LTS $ mkdir pharo7 $ cd pharo7/ $ curl get.pharo.org/70+vm | bash % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 3036 100 3036 0 0 36799 0 --:--:-- --:--:-- --:--:-- 37024 Downloading the latest 70 Image: http://files.pharo.org/get-files/70/pharo.zip Pharo.image Downloading the latest pharoVM: http://files.pharo.org/get-files/70/pharo-linux-stable.zip pharo-vm/pharo Creating starter scripts pharo and pharo-ui On a 64-bit system? You must enable and install the 32-bit libraries Please see http://pharo.org/gnu-linux-installation for detailed instructions $ ./pharo Pharo.image config http://mc.stfx.eu/Neo ConfigurationOfNeoConsole --install=bleedingEdge 'Installing ConfigurationOfNeoConsole bleedingEdge'

Loading 1-baseline of ConfigurationOfNeoConsole... Fetched -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- http://mc.stfx.eu/Neo Loaded -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- cache ...finished 1-baseline $ ./pharo Pharo.image eval NeoConsoleStdio run Neo Console Pharo-7.0+alpha.build.760.sha.d2734dcabda799803c307365bcd120f92211d34a (32 Bit) pharo> 1+2

3 pharo> 42 factorial

1405006117752879898543142606244511569936384000000000 pharo> Stdio stdin

StdioStream: #stdin pharo> == self: StdioStream: #stdin class: StdioStream file: a File handle: #[148 213 25 107 160 197 105 247 0 0 0 0 0 0 0 0 0 1 255 1] forWrite: false peekBuffer: nil pharo> show StdioStream>>#atEnd StdioStream>>#atEnd atEnd

^ file atEnd pharo> get process.list Morphic UI Process Delay Scheduling Process Low Space Watcher Input Event Fetcher Process Idle Process WeakArray Finalization Process CommandLine handler process pharo> quit Bye! a NeoConsoleStdio

I know there are many approaches to a REPL, I don't claim mine is best, it is just the one that I have been using for years.

In the above, I do not depend on EOF - just to be clear. The point being that there is no immediate fundamental problem.

But there is something wrong with what is returned by Stdio stdin

Sven

-- _,,,^..^,,,_ best, Eliot

On 12 Apr 2018, at 04:07, Ben Coman <btc@openinworld.com> wrote:

On 12 April 2018 at 04:25, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

On 11 Apr 2018, at 21:44, Stephane Ducasse <stepharo.self@gmail.com> wrote:

I did not know about the NeoConsole. Nice because I wanted to build a little REPL for my minilanguage implementation.

You are of course welcome to look at it. But it is Pharo specific. I use it to be able to hook/look into running headless server images. For this it is super handy.

A description of how to set that up would make a nice blog post.

It is part of, and partially documented in https://github.com/svenvc/pharo-server-tools But maybe I'll write something about it, yes.

cheers -ben

On 12 Apr 2018, at 08:33, Alistair Grant <akgrant0710@gmail.com> wrote:

Hi Sven,

On 11 April 2018 at 20:47, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Alistair,

...

On 11 Apr 2018, at 19:42, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I will send you some code later on.

Today I arranged for my NeoConsole code (that normally works over a network connection) to work over stdio. Although I am not yet happy with every aspect of the implementation, it does work (using unaltered Zn streams and code). The foll

$ cat /etc/issue Ubuntu 16.04.4 LTS $ mkdir pharo7 $ cd pharo7/ $ curl get.pharo.org/70+vm | bash % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 3036 100 3036 0 0 36799 0 --:--:-- --:--:-- --:--:-- 37024 Downloading the latest 70 Image: http://files.pharo.org/get-files/70/pharo.zip Pharo.image Downloading the latest pharoVM: http://files.pharo.org/get-files/70/pharo-linux-stable.zip pharo-vm/pharo Creating starter scripts pharo and pharo-ui On a 64-bit system? You must enable and install the 32-bit libraries Please see http://pharo.org/gnu-linux-installation for detailed instructions $ ./pharo Pharo.image config http://mc.stfx.eu/Neo ConfigurationOfNeoConsole --install=bleedingEdge 'Installing ConfigurationOfNeoConsole bleedingEdge'

Loading 1-baseline of ConfigurationOfNeoConsole... Fetched -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- http://mc.stfx.eu/Neo Loaded -> Neo-Console-Core-SvenVanCaekenberghe.24 --- http://mc.stfx.eu/Neo --- cache ...finished 1-baseline $ ./pharo Pharo.image eval NeoConsoleStdio run Neo Console Pharo-7.0+alpha.build.760.sha.d2734dcabda799803c307365bcd120f92211d34a (32 Bit) pharo> 1+2

3 pharo> 42 factorial

1405006117752879898543142606244511569936384000000000 pharo> Stdio stdin

StdioStream: #stdin pharo> == self: StdioStream: #stdin class: StdioStream file: a File handle: #[148 213 25 107 160 197 105 247 0 0 0 0 0 0 0 0 0 1 255 1] forWrite: false peekBuffer: nil pharo> show StdioStream>>#atEnd StdioStream>>#atEnd atEnd

^ file atEnd pharo> get process.list Morphic UI Process Delay Scheduling Process Low Space Watcher Input Event Fetcher Process Idle Process WeakArray Finalization Process CommandLine handler process pharo> quit Bye! a NeoConsoleStdio

I know there are many approaches to a REPL, I don't claim mine is best, it is just the one that I have been using for years.

In the above, I do not depend on EOF - just to be clear. The point being that there is no immediate fundamental problem.

Cool. I've wanted to try NeoConsole for a while (I've got a Raspberry Pi monitoring temperatures and wanted to use it to work on the Pi) but hadn't made the time.

But... :-)

$ vm/pharo Pharo7.0-64bit-d2734dc.image eval NeoConsoleStdio run Neo Console Pharo-7.0+alpha.build.760.sha.d2734dcabda799803c307365bcd120f92211d34a (64 Bit) pharo> 4+3

7 pharo> [Ctrl-D] MessageNotUnderstood: receiver of "<" is nil UndefinedObject(Object)>>doesNotUnderstand: #< ZnUTF8Encoder>>nextCodePointFromStream: ZnUTF8Encoder(ZnCharacterEncoder)>>nextFromStream: ZnCharacterReadStream>>nextElement ZnCharacterReadStream(ZnEncodedReadStream)>>next [ :out | | eol char | eol := false. [ eol ] whileFalse: [ char := readStream next. (char isNil or: [ char == lf ]) ifTrue: [ eol := true ] ifFalse: [ char == cr ifTrue: [ eol := true. readStream peekFor: lf ] ifFalse: [ out nextPut: char ] ] ] ] in ZnFastLineReader>>nextLine in Block: [ :out | ... ZnFastLineReader>>streamContents: ZnFastLineReader>>nextLine [ lineReader nextLine trimBoth ] in NeoConsoleStdio>>readInputFrom: in Block: [ lineReader nextLine trimBoth ] BlockClosure>>on:do: NeoConsoleStdio>>readInputFrom: NeoConsoleStdio>>executeRequestResponseLoop NeoConsoleStdio>>run NeoConsoleStdio class>>run UndefinedObject>>DoIt

In your example you've carefully exited by some other means than signalling end-of-file (Ctrl-D).

I don't think we can reasonably say "if you write code with Zinc streams, ensure that none of your users ever press Ctrl-D, they must always type 'quit' (or whatever)".

I left that open as an exercise for the reader - I knew you would try that ;-) The code was originally written for network streams and there it catches ConnectionClosed which is the network stream equivalent of EOF (more or less). I think we need something like an EndOfStream exception which should be thrown in that case, instead of returning nil. Returning nil from ZnCharacterReadStream or ZnCharacterEncoder when the stream is EOF is wrong IMHO. The stream being EOF is an exceptional situation (even though it is relatively common) and there is always #atEnd that works well in 99% of the cases. AFAIK that is also the design decision taking by Xtreams (where I believe the Incomplete exception replaces #atEnd).

...

But there is something wrong with what is returned by Stdio stdin

Right, that's the bug I was addressing stdio.cs. I've got a note to open an issue and submit a PR.

Cheers, Alistair

On 12 April 2018 at 10:48, Alistair Grant <akgrant0710@gmail.com> wrote:

So I'm going to respectfully disagree and say that I think to raise a MNU instead of returning nil is wrong, and that changing the Zinc streams isn't such a big deal.

Sorry Sven, this sounds like I'm suggesting that your planning to leave the MNU in there, which would be rather rude of me, but isn't what I meant at all. But I do think Zinc streams should handle #next returning nil before #atEnd answers true. Cheers, Alistair

Thanks!!!! On Thu, Apr 12, 2018 at 5:06 PM, Alistair Grant <akgrant0710@gmail.com> wrote:

On 11 April 2018 at 20:47, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

But there is something wrong with what is returned by Stdio stdin

I've opened https://pharo.fogbugz.com/f/cases/21692/StdioStream-incorrectly-delegates-at...

PR in progress.

Cheers, Alistair

On 11 April 2018 at 19:42, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:

2018-04-11 19:08 GMT+02:00 Alistair Grant <akgrant0710@gmail.com>:

...

Hi Sven,

On 11 April 2018 at 18:53, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

Something is off (and/or I am getting crazy, probably both).

$ ./pharo --headless Pharo.image eval '(FileStream stdin binary; next: 3)' 123 #[49]

??

This should return #[49 50 51] AFAIK.

I haven't checked properly, but I can take a guess at this...

The file read primitive in FilePlugin tries to read from stdin 1 character at a time (presumably to get around the line buffering that is done, but I haven't gone through the history. At one time I think it modified the terminal settings, but that code is #if'd out). For all other streams it attempts to read the requested number of characters.

StandardFileStream>>readInto:startingAt:count: assumes that primitiveFileRead will always attempt to read count bytes, but it actually only attempts to read 1.

Cheers, Alistair

https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/150

Thanks! I'd already looked at this routine and found it strangely constructed. Having some background discussion will help a lot. Cheers, Alistair

On Wed, Apr 11, 2018 at 5:33 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I think we have to reset this whole discussion.

FileStream stdin

and

Stdio stdin

are completely different !

Could you explain the difference because I'm dead tired and got confused :)

We'll have to check that first, before talking about the issues raised in this thread.

And BTW these terminal streams are a real pain to test ;-)

...

On 11 Apr 2018, at 17:20, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

On 11 Apr 2018, at 17:16, Denis Kudriashov <dionisiydk@gmail.com> wrote:

2018-04-11 17:02 GMT+02:00 Sven Van Caekenberghe <sven@stfx.eu>:

...

On 11 Apr 2018, at 16:36, Sven Van Caekenberghe <sven@stfx.eu> wrote:

I can make your example, using the Zn variants, work with the following change:

StdioStream>>#atEnd ^ peekBuffer isNil or: [ (peekBuffer := self next) isNil ]

Argh, make that

atEnd ^ peekBuffer isNil and: [ (peekBuffer := self next) isNil ]

But discussion exactly about "self next isNil": how to avoid it.

I know, but like this it could/might become an implementation detail.

The more things that I try, the more that I feel that stdin is so special that it does not fit in the rest of the stream zoo.

...

but I am still testing, this is probably not the final answer/solution.

...

Which is a literal implementation of your statement that you can only know that you are atEnd by reading (and thus waiting/blocking) and checking for nil, which seems logical to me, given the fact that you *are* waiting for user input.

BTW, at least on macOS you have to enter ctrl-D (^D) on a separate line, I am not sure how relevant that is, but that is probably another argument that stdin is special (being line-buffered by the OS, EOF needing to be on a separate line).

And FWIW, I have been writing networking code in Pharo for years, and I have never had issues with unclear semantics of these primitives (#atEnd, #next, #peek) on network streams, either the classic SocketStream or the Zdc* streams (TLS or not). That is why I think we have to be careful.

That being said, it is important to continue this discussion, I find it very interesting. I am trying to write some test code using stdin myself, to better understand the topic.

...

On 11 Apr 2018, at 16:06, Alistair Grant <akgrant0710@gmail.com> wrote:

On 11 April 2018 at 15:11, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

...

On 11 Apr 2018, at 11:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:

How does one modify #atEnd to block ? I suppose you are talking about StdioStream>>#atEnd ?

^ self peek isNil

?

Still the same question, how do you implement a blocking #atEnd for stdin ?

I have seen your stdio.cs which is indeed needed as the current StdioStream>>#atEnd is bogus for sure.

But that is still a non-blocking one, right ?

Since there is a peekBuffer in StdioStream, why can't that be used ?

I think you've created a chicken-and-egg problem with this question, but ignoring that for now:

StdioStream>>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil. "

self atEnd ifTrue: [^ nil ].

peekBuffer ifNotNil: [ ^ peekBuffer ].

^ peekBuffer := self next.

So when we first start the program, i.e. the user hasn't entered any input yet, and #peek is called:

1. #atEnd returns false because Ctrl-D (or similar) hasn't been entered (assuming it is non-blocking). 2. peekBuffer is nil because we haven't previously called #peek. 3. The system now blocks on "self next".

Just a reminder: for terminal input the end-of-file isn't reached until the user explicitly enters the end of file key (Ctrl-D).

So, if there is no buffered input (either none has been entered yet, or all input has been consumed)

#atEnd (after the patch) calls #primAtEnd:.

At the moment, #primAtEnd: ends up calling the libc function feof(), which is non-blocking and answers the end-of-file flag for the FILE*. Since the user hasn't entered Ctrl-D, that's false.

If we want to control iteration over the stream and ensure that we don't need to do a "stream next == nil" check, then #primAtEnd: is going to have to peek for the next character, and that means waiting for the user to enter that character.

In c that is typically done using:

atEnd = ungetc(fgetc(fp), fp);

and fgetc() will block until the user enters something.

...

I have run your example testAtEnd.st now, and it works/fails as advertised.

:-)

Cheers, Alistair

On Wednesday 04 April 2018 04:06 PM, Nicolas Cellier wrote:

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized.

Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as:

atEnd := stream position >= stream size I see a confusion between Stream and its underlying collection. Stream is an iterator and just does next, nextPut, peek, reset etc. But methods like size or atEnd depend on its collection and there is no guarantee that this collection has a known and finite size.

Essentially, a collection's size may be known and finite, unknown but finite size or infinite. This has nothing do with file descriptor being std{in,out,err}. If std* is a regular file or special file like /dev/mem, /dev/null, it's size is known at open time. With console streams or sysfs files, size is unknown until EOT (^D) or NUL is received. Lastly, special files like /dev/zero, /dev/random or /proc/cpuinfo don't have a finite size but report it as zero (!). [ stream atEnd ] whileFalse: [ stream next. .. ] will only terminate if its collection size is finite. It won't terminate for infinite collections. Regards .. Subbu

On Wed, Apr 4, 2018 at 8:37 AM, Sven Van Caekenberghe <sven@stfx.eu> wrote:

...

On 4 Apr 2018, at 17:32, K K Subbu <kksubbu.ml@gmail.com> wrote:

On Wednesday 04 April 2018 04:06 PM, Nicolas Cellier wrote:

...

...

IIRC, someone said it is implemented as 'remaining size being zero' and some virtual unix files like /dev/random are zero sized. Currently, for files other than sdio (stdout, stderr, stdin) it is effectively defined as: atEnd := stream position >= stream size I see a confusion between Stream and its underlying collection. Stream is an iterator and just does next, nextPut, peek, reset etc. But methods like size or atEnd depend on its collection and there is no guarantee that this collection has a known and finite size.

Essentially, a collection's size may be known and finite, unknown but finite size or infinite. This has nothing do with file descriptor being std{in,out,err}. If std* is a regular file or special file like /dev/mem, /dev/null, it's size is known at open time. With console streams or sysfs files, size is unknown until EOT (^D) or NUL is received. Lastly, special files like /dev/zero, /dev/random or /proc/cpuinfo don't have a finite size but report it as zero (!).

[ stream atEnd ] whileFalse: [ stream next. .. ]

will only terminate if its collection size is finite. It won't terminate for infinite collections.

Regards .. Subbu

Good summary, I agree.

Still, what are the semantics of #next - does the caller always have to check for nil ? Do we think this is ugly (as the return value is outside the domain) ?

The problem is not that the value is outside the domain. The problem is that it may be within the domain. Answering an element within the domain when there are no more elements is insane. Not even C does that.

Do we then still need #atEnd ?

For backward compatibility, yes, even if deprecated. There is so much existing code that uses it, it'll be useful to have.

Sven

_,,,^..^,,,_ best, Eliot