[Pharo-project] XTream was Re: Re: Ocean (was Re: Less plugins, more Smalltalk code!)

Nicolas Cellier nicolas.cellier.aka.nice at gmail.com
Thu Oct 7 15:41:23 EDT 2010


Yes, every time I must warn and apologize because I'm guilty of
hijacking the name.
The goals are more or less the same, because I also hijacked some
ideas (though not all invented at VW):
- separate read/write stream
- use stream wrappers
- implement main sequenceable collection API
(more to be found at http://www.squeaksource.com/XTream.html wiki page)

There are also Squeak specific goals:
- clean the mess by first principles (just an excerpt
http://lists.squeakfoundation.org/pipermail/squeak-dev/2009-December/141786.html
)
- accelerate MultiByteFileStream

At time Squeak Xtream started, there was no license for VW Xtream, so
it's a full rewrite from scratch.
Thus implementation differs. Mainly I did not reify the Buffer (the
master piece of original VW Xtream).
Due to this, I've got a BufferedReadWriteStream, which is
contradictory to goal 1).
It might change in future.
I also do not want to depend too much on Exception handling, which
current VW Buffer implementation does.

Also, I did not replace the whole protocol, but rather reused existing one.
That's because further goal is to provide a replacement for Squeak
Streams (could apply to Pharo too, but there's a possible overlap with
Nile* See note below).
However, I currently made no plan to reach this goal. I just have some
toy images that I break regularly due to lack of compatibility...
Either I implement all the messy protocol that flourished in Squeak,
or I try to stay a bit cleaner but with uncompatibilities...
Maybe a separate compatibility package with a deprecation policy might help...

Be aware that recent Squeak and Pharo acceleration thanks to buffering
was first experimented in Squeak Xtream, and then ported by talents of
Levente, that's already a little reward.
Some big acceleration could be gained other UTF8 streams when a
majority of characters are ASCII (the case of source code management
and most latin characters texts), both in cog and traditional vm.
My current tests (cog) are:

{
[| tmp | tmp := (MultiByteFileStream readOnlyFileNamed: (SourceFiles
at: 2) name) ascii; wantsLineEndConversion: false; converter:
UTF8TextConverter new.
       1 to: 20000 do: [:i | tmp upTo: Character cr]. tmp close] timeToRun.
[| tmp | tmp := ((StandardFileStream readOnlyFileNamed: (SourceFiles
at: 2) name) readXtream binary buffered ~> UTF8Decoder) buffered.
       1 to: 20000 do: [:i | tmp upTo: Character cr]. tmp close] timeToRun.
}
 #(163 21)

So, Xtream is around 8x faster at reading first 20000 lines of change
log... No surprise it's more than 99% pure ASCII.

Last, concerning the errors, it's more a matter of fixing the tests
themselves...
They use #squeakToUtf8 as a reference, and some tests are for other
unloaded XTream packages.
I started to change the test last night but did not publish all...

See also:
http://lists.squeakfoundation.org/pipermail/squeak-dev/2010-September/153578.html
http://lists.squeakfoundation.org/pipermail/squeak-dev/2009-December/141539.html
http://lists.squeakfoundation.org/pipermail/squeak-dev/2009-December/141647.html
and more mails in november/december 2009

more dated material http://bugs.squeak.org/view.php?id=6755
and various attempts like LazyStream/LazyCollections...

Nicolas

* Note about Nile:
Nile has a different approach: do not eliminate ReadWriteStreams, but
compose with traits.
Nile also has high test coverage level, which is good.
Nile was designed as a Stream replacement for Squeak Stream...
As such, it carries some cruft (same dilemna as Squeak Xtream).
Pharo goal is to perform major clean-up. As a mean to reach this goal,
it shall not encumber with backward compatibility. This means that
once Nile adopted (with historical compatibility cruft), it should be
cleaned up again...

2010/10/7 Stéphane Ducasse <stephane.ducasse at inria.fr>:
> pay attention that there are two xtreams package frameworks from what I understand
>
>        one presented at esug and one developed by nicolas
>
> Stef
>
> On Oct 7, 2010, at 4:58 PM, Sven Van Caekenberghe wrote:
>
>>
>> On 06 Oct 2010, at 08:48, Nicolas Cellier wrote:
>>
>>> See also hijacked  http://www.squeaksource.com/XTream.html borrowing
>>> some of the ideas, but somehow less extreme.
>>> Most packages should now load in Pharo.
>>
>> Hi Nicolas,
>>
>> I have been looking at the ESUG 2010 slides & google code project pages of xstreams and I must say that I like this very much.
>>
>> Your code on SS does indeed load in Pharo 1.1. 8 tests fail out of 37 with errors that I think are probably fixable. I am browsing it now.
>>
>> Now my question is, first, do you have some write up somewhere explaining what you did and why, and second, what is the current state of your project and what are your future plans/goals ?
>>
>> Thx,
>>
>> Sven
>>
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> Pharo-project at lists.gforge.inria.fr
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> Pharo-project at lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>




More information about the Pharo-dev mailing list