Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ] PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1 But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil." | nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek". Stephan and Diego
Hi Diego, Streams as they are currently defined, especially ReadStreams, can hardly be called IO streams. Many operations assume that there is a single collection over which you can move in any direction without restrictions. As a consequence most parsing code relies this ability, looking forward and going back multiple times, like using #skip: with a negative argument. Real IO ReadStreams like a SocketStreams or FileStreams cannot be assumed to keep more than just a buffer in memory, hence the unrestricted operations are not implementable. IMO, parsing code should be written against a much more restricted API, assuming at most a one element peek/buffer. I always try to do that, I even have Mock ReadStreams that ensure this. While benchmarking I have also found that slow #peek behaviour can be a bottleneck, especially on some of the more complex streams. Sven PS: Have a look at ZnCharacterReadStream for example, which implements this restricted behaviour: it is not positionable, but allows a one character peek using a one character internal buffer. On 04 Nov 2013, at 09:57, Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching. Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek: cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close. FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file) [cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.' [FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters. TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet. Cheers, Henry
Hi Henrik, Great writeup, thanks ! (more inline) On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption). I am still chasing a bug, but right now: [ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench. "135 per second.â BEFORE "3,310 per second.â AFTER But of course the input file is ASCII, so YMMV. Iâll let you know when I commit this code. Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
That's great! Remembering that commit message was part of the reason for benching, was sort of disappointed there was no significant difference between Zn in 2.0 and latest 3.0... I guess with the amount of hacks accumulating, it is indeed turning into a worthy successor of MultiByteFileStream ;) Cheers, Henry P.S: If you want another delightful one in the same vein (both from WTFy-ness and perf improvement POV), take a gander at UTF16TextConverter >> nextPutByteString:toStream: On Mon, Nov 4, 2013 at 5:12 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote:
Hi Henrik,
Great writeup, thanks !
(more inline)
On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
On 04 Nov 2013, at 22:25, Henrik Sperre Johansen <henrik.s.johansen@veloxit.no> wrote:
That's great! Remembering that commit message was part of the reason for benching, was sort of disappointed there was no significant difference between Zn in 2.0 and latest 3.0...
I guess with the amount of hacks accumulating, it is indeed turning into a worthy successor of MultiByteFileStream ;)
Bah, the hacking is limited to one method ( #optimizedReadInto:startingAt:count:fromStream: ) and the readable version is still present and used ( #originalReadInto:startingAt:count:fromStream: ). It is encapsulated and the rest of the code remains quite clear.
Cheers, Henry
P.S: If you want another delightful one in the same vein (both from WTFy-ness and perf improvement POV), take a gander at UTF16TextConverter >> nextPutByteString:toStream:
Pretty confusing ! BTW, do we still need UTF16 support ? For those encodings that we still want to support in the future, we should have new and more principled implementations under ZnCharacterEncoder. That is, if we ever want to fase out TextConverter.
On Mon, Nov 4, 2013 at 5:12 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote: Hi Henrik,
Great writeup, thanks !
(more inline)
On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
On 06 Nov 2013, at 10:38 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
BTW, do we still need UTF16 support ?
For those encodings that we still want to support in the future, we should have new and more principled implementations under ZnCharacterEncoder. That is, if we ever want to fase out TextConverter.
UTF16 is the encoding for string arguments to the Unicode-aware version of the Windows API's, so Iâd say yes. Not too sure about JNI, but I know Java also uses UTF16 internally. So the short answer is yes. Not as a storage encoding, but for calling foreign functions, itâs quite important. Cheers, Henry
On 06 Nov 2013, at 13:29, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 06 Nov 2013, at 10:38 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
BTW, do we still need UTF16 support ?
For those encodings that we still want to support in the future, we should have new and more principled implementations under ZnCharacterEncoder. That is, if we ever want to fase out TextConverter.
UTF16 is the encoding for string arguments to the Unicode-aware version of the Windows API's, so Iâd say yes. Not too sure about JNI, but I know Java also uses UTF16 internally.
So the short answer is yes. Not as a storage encoding, but for calling foreign functions, itâs quite important.
Ah, OK. Fancy giving it a try, to create a ânewâ implementation under ZnCharacterEncoder with some tests ? There are only 3 abstract methods to implement, if you should already know UTF16, it would not be too much work ;-)
Cheers, Henry
On 06 Nov 2013, at 13:59, Sven Van Caekenberghe <sven@stfx.eu> wrote:
On 06 Nov 2013, at 13:29, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 06 Nov 2013, at 10:38 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
BTW, do we still need UTF16 support ?
For those encodings that we still want to support in the future, we should have new and more principled implementations under ZnCharacterEncoder. That is, if we ever want to fase out TextConverter.
UTF16 is the encoding for string arguments to the Unicode-aware version of the Windows API's, so Iâd say yes. Not too sure about JNI, but I know Java also uses UTF16 internally.
So the short answer is yes. Not as a storage encoding, but for calling foreign functions, itâs quite important.
Ah, OK.
Fancy giving it a try, to create a ânewâ implementation under ZnCharacterEncoder with some tests ?
There are only 3 abstract methods to implement, if you should already know UTF16, it would not be too much work ;-)
A first attempt: === Name: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.27 Author: SvenVanCaekenberghe Time: 8 November 2013, 4:18:07.642898 pm UUID: 29824770-7b9d-4377-a934-7bb2fbeefefb Ancestors: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.26 Added new ZnUTF16Encoder Minor refactoring to ZnCharacterEncoder hierarchy === Name: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.14 Author: SvenVanCaekenberghe Time: 8 November 2013, 4:18:53.717919 pm UUID: 6309f553-1632-438d-825c-b7a0f89193f4 Ancestors: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.13 Added unit tests for the new ZnUTF16Encoder ===
Cheers, Henry
BTW, do we still need UTF16 support ?
For those encodings that we still want to support in the future, we should have new and more principled implementations under ZnCharacterEncoder. That is, if we ever want to fase out TextConverter.
:) You know what I think. Kill all the rotten code. Stef
Hi henrik why don't you give a try to change our lives and propose a new MultuByteFileStream and friends :) Stef
That's great! Remembering that commit message was part of the reason for benching, was sort of disappointed there was no significant difference between Zn in 2.0 and latest 3.0...
I guess with the amount of hacks accumulating, it is indeed turning into a worthy successor of MultiByteFileStream ;)
Cheers, Henry
P.S: If you want another delightful one in the same vein (both from WTFy-ness and perf improvement POV), take a gander at UTF16TextConverter >> nextPutByteString:toStream:
On Mon, Nov 4, 2013 at 5:12 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote: Hi Henrik,
Great writeup, thanks !
(more inline)
On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
that would be called XStreams, no? On 2013-11-06, at 13:43, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:
Hi henrik
why don't you give a try to change our lives and propose a new MultuByteFileStream and friends :)
Stef
That's great! Remembering that commit message was part of the reason for benching, was sort of disappointed there was no significant difference between Zn in 2.0 and latest 3.0...
I guess with the amount of hacks accumulating, it is indeed turning into a worthy successor of MultiByteFileStream ;)
Cheers, Henry
P.S: If you want another delightful one in the same vein (both from WTFy-ness and perf improvement POV), take a gander at UTF16TextConverter >> nextPutByteString:toStream:
On Mon, Nov 4, 2013 at 5:12 PM, Sven Van Caekenberghe <sven@stfx.eu> wrote: Hi Henrik,
Great writeup, thanks !
(more inline)
On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
On 04 Nov 2013, at 17:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:
Hi Henrik,
Great writeup, thanks !
(more inline)
On 04 Nov 2013, at 11:58, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 9:57 , Diego Lont <diego.lont@delware.nl> wrote:
Working on Petit Delphi we found a strange implementation for asPetitStream: Stream>asPetitStream ^ self contents asPetitStream
Further investigation showed that the basic peek was not fast enough for Petit Parser, as it is used a lot. So it implemented a "improved unchecked peek": PPStream>peek "An improved version of peek, that is slightly faster than the built in version." ^ self atEnd ifFalse: [ collection at: position + 1 ]
PPStream>uncheckedPeek "An unchecked version of peek that throws an error if we try to peek over the end of the stream, even faster than #peek." ^ collection at: position + 1
But in my knowledge a basic peek should be fast. The real problem is the peek in the underlying peek: PositionableStream>peek "Answer what would be returned if the message next were sent to the receiver. If the receiver is at the end, answer nil."
| nextObject | self atEnd ifTrue: [^nil]. nextObject := self next. position := position - 1. ^nextObject
That actually uses "self next". The least thing one should do is to cache the next object. But isn't there a primitive for peek in a file stream? Because al overriding peeks of PositionableStream have basically the same implementation: reading the next and restoring the state to before the peek (that is slow). So we would like to be able to remove PPStream without causing performance issues, as the only added method is the "improved peek".
Stephan and Diego
If you are reading from file, ZnCharacterStream should be a valid alternative. If not, ZnBufferedReadStream on an internal collection stream also does peek caching.
Beware with files though; itâs better to bench the overall operation for different alternatives. F.ex, ZnCharacterStream is much faster than the standard Filestream for peek:
cr := ZnCharacterReadStream on: 'PharoDebug.log' asFileReference readStream binary. [cr peek] bench. '49,400,000 per second.' cr close.
FileStream fileNamed: 'PharoDebug.log' do: [:fs | [fs peek] bench] '535,000 per second.â
but has different bulk reads characteristics (faster for small bulks, slower for large bulks, crossover-point at around 1k chars at once); (The actual values are of course also dependent on encoder/file contents, those given here obtained with UTF-8 and a mostly/all ascii text file)
[cr := ZnCharacterReadStream on: ('PharoDebug.log' asFileReference readStream binary ) readStream. cr next: 65536; close] bench '105 per second.' '106 per second.â
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
=== Name: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.26 Author: SvenVanCaekenberghe Time: 6 November 2013, 10:09:58.837656 am UUID: dcb21b30-9596-446d-801c-b091cf3f415e Ancestors: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.25 ZnCharacterReadStream and ZnCharacterWriteStream now use the optimized bulk reading/writing methods from ZnCharacterEncoder (#readInto:startingAt:count:fromStream: and #next:putAll:startingAt:toStream: respectively) Fixed a bug in ZnUTF8Encoder>>#optimizedReadInto:startingAt:count:fromStream: Some refactoring and cleanup of ZnUTF8Encoder and ZnByteEncoder Added 2 more unit tests === Name: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.13 Author: SvenVanCaekenberghe Time: 6 November 2013, 10:11:11.411621 am UUID: 71b503eb-548e-4e2f-b80d-4d2b805bd11e Ancestors: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.12 Added 2 more unit tests ZnCharacterReadStream and ZnCharacterWriteStream now use the optimized bulk reading/writing methods from ZnCharacterEncoder (#readInto:startingAt:count:fromStream: and #next:putAll:startingAt:toStream: respectively) Fixed a bug in ZnUTF8Encoder>>#optimizedReadInto:startingAt:count:fromStream: Some refactoring and cleanup of ZnUTF8Encoder and ZnByteEncoder ===
Sven
[FileStream fileNamed: 'PharoDebug.log' do: [:fs | fs next: 65536] ] bench '176 per second.â
If you use a StandardFilestream set to binary ( which has less overhead for binary nextâs compared to the MultiByteFileStream returned by asFileReference readStream)as the base stream instead, , the same general profile holds true, but with a crossover around 2k characters.
TL;DR: Benchmark the alternatives. The best replacement option depends on your results. Appropriately (according to source and actual use) set up Zn-streams are probably your best bet.
Cheers, Henry
On 04 Nov 2013, at 5:12 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
Yeah⦠sooo, I loaded the updated version, great improvement for streams on Latin1 content :D Maybe itâs just me, but I tested with actual wide source as well (it was as slow as youâd expect), and I think you need a notQuiteSoOptimizedReadInto* which uses the normal Byte -> Wide become: conversion machinery. Writing a ZnByteStringBecameWideString handler for every next: (and cousins) call where source may be non-latin1 is a real chore/nasty surprise for those used to dealing with legacy streams/converters... You can sort of kinda make up for the performance hit (at least on these sizes) with a faster replace:from:to:with:startingAt: in use after the fact, by using basicAt:put: to the string, thus avoiding converting the replacement value to character. Cheers, Henry PS: Why is ZnByteStringBecameWideString a notification and not a resumable exception? I would assume those who run into it without a handler would rather have an actual error, than a result where their string has been read up to the first wide char...
On 06 Nov 2013, at 13:20, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 5:12 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
Yeah⦠sooo, I loaded the updated version, great improvement for streams on Latin1 content :D
Maybe itâs just me, but I tested with actual wide source as well (it was as slow as youâd expect), and I think you need a notQuiteSoOptimizedReadInto* which uses the normal Byte -> Wide become: conversion machinery. Writing a ZnByteStringBecameWideString handler for every next: (and cousins) call where source may be non-latin1 is a real chore/nasty surprise for those used to dealing with legacy streams/converters... You can sort of kinda make up for the performance hit (at least on these sizes) with a faster replace:from:to:with:startingAt: in use after the fact, by using basicAt:put: to the string, thus avoiding converting the replacement value to character.
Cheers, Henry
PS: Why is ZnByteStringBecameWideString a notification and not a resumable exception? I would assume those who run into it without a handler would rather have an actual error, than a result where their string has been read up to the first wide charâ¦
Henrik, thanks again for the feedback, it is really useful. You are right, the ZnByteStringBecameWideString hack was introduced specifically to avoid the become, because benchmarking showed that it took too long. The original user is ZnStringEntity>>#readFrom: Now, if you know up front that your string will be wide anyway, then there is #decodeBytesIntoWideString: Yes, ZnByteStringBecameWideString should probably be a resumable exception, as not handling it properly will lead to errors. I will have to think and reconsider the generality/specificity of the optimisation hack in ZnUTF8Encoder. I got some ideas, but it will take some time. At least the code works now, time to process the constructive criticism. Regards, Sven
On 06 Nov 2013, at 13:56, Sven Van Caekenberghe <sven@stfx.eu> wrote:
On 06 Nov 2013, at 13:20, Henrik Johansen <henrik.s.johansen@veloxit.no> wrote:
On 04 Nov 2013, at 5:12 , Sven Van Caekenberghe <sven@stfx.eu> wrote:
Well, I just realised that ZnCharacterReadStream and ZnCharacterWriteStream did not yet make use of the optimisations that I did for ZnCharacterEncoding some time ago. More specifically, they were not yet using #next:putAll:startingAt:toStream: and #readInto:startingAt:count:fromStream: which are overwritten for ZnUTF8Encoder with (super hacked) versions that assume most of the input will be ASCII (a reasonable assumption).
I am still chasing a bug, but right now:
[ (ZnCharacterReadStream on: ('timezones.json' asFileReference readStream binary)) next: 65536; close ] bench.
"135 per second.â BEFORE "3,310 per second.â AFTER
But of course the input file is ASCII, so YMMV.
Iâll let you know when I commit this code.
Sven
Yeah⦠sooo, I loaded the updated version, great improvement for streams on Latin1 content :D
Maybe itâs just me, but I tested with actual wide source as well (it was as slow as youâd expect), and I think you need a notQuiteSoOptimizedReadInto* which uses the normal Byte -> Wide become: conversion machinery. Writing a ZnByteStringBecameWideString handler for every next: (and cousins) call where source may be non-latin1 is a real chore/nasty surprise for those used to dealing with legacy streams/converters... You can sort of kinda make up for the performance hit (at least on these sizes) with a faster replace:from:to:with:startingAt: in use after the fact, by using basicAt:put: to the string, thus avoiding converting the replacement value to character.
Cheers, Henry
PS: Why is ZnByteStringBecameWideString a notification and not a resumable exception? I would assume those who run into it without a handler would rather have an actual error, than a result where their string has been read up to the first wide charâ¦
Henrik, thanks again for the feedback, it is really useful.
You are right, the ZnByteStringBecameWideString hack was introduced specifically to avoid the become, because benchmarking showed that it took too long. The original user is ZnStringEntity>>#readFrom:
Now, if you know up front that your string will be wide anyway, then there is #decodeBytesIntoWideString:
Yes, ZnByteStringBecameWideString should probably be a resumable exception, as not handling it properly will lead to errors.
I will have to think and reconsider the generality/specificity of the optimisation hack in ZnUTF8Encoder. I got some ideas, but it will take some time.
At least the code works now, time to process the constructive criticism.
=== Name: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.28 Author: SvenVanCaekenberghe Time: 13 November 2013, 3:04:00.925461 pm UUID: f104e1ee-7fe5-4a23-a650-197dd9462783 Ancestors: Zinc-Character-Encoding-Core-SvenVanCaekenberghe.27 ZnByteStringBecameWideString is now a resumable Error instead of a Notification (as suggested by henrik sperre johansen); Added ZnByteStringBecameWideString>>#becomeForward convenience method Fixed ZnCharacterReadStream>>#readInto:startingAt:count: to do the actual becomeForward when needed so that clients are not affected === Name: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.15 Author: SvenVanCaekenberghe Time: 13 November 2013, 3:05:18.26391 pm UUID: 8c41ee29-9906-4626-a9da-cb576980cd4c Ancestors: Zinc-Character-Encoding-Tests-SvenVanCaekenberghe.14 Added a unit test to make sure ZnCharacterReadStream>>#readInto:startingAt:count: does an actual becomeForward when needed === At least the situation is correct now. Sven
participants (6)
-
Camillo Bruni -
Diego Lont -
Henrik Johansen -
Henrik Sperre Johansen
-
Stéphane Ducasse -
Sven Van Caekenberghe