Fwd: [Newcompiler] Status of the NewCompiler

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Fwd: [Newcompiler] Status of the NewCompiler

Stéphane Ducasse-3


Begin forwarded message:

> From: Marcus Denker <[hidden email]>
> Date: 9 avril 2007 11:29:51 HAEC
> To: "The New Compiler, AST, IRBuilder, Closures..."  
> <[hidden email]>
> Subject: Re: [Newcompiler] Status of the NewCompiler
> Reply-To: "The New Compiler, AST, IRBuilder, Closures..."  
> <[hidden email]>
>
>
> On 02.04.2007, at 12:04, Stéphane Ducasse wrote:
>
>> may be you should post that to squeak-dev too.
>>
>
> The problem is that I am not subscribed right now... maybe someone  
> else can do that? The
> list of things done by math would need to be extended with the  
> things mentioned later in the thread,
> and the mail should have a link to this mailinstlist's subscribe page.
>
>> Stef
>> On 1 avr. 07, at 01:32, Marcus Denker wrote:
>>
>>> Hello,
>>>
>>> Last summer/fall, Mathieu Suen worked (as part of the ESUG  
>>> SummerTalk) on the NewCompiler framework.
>>>
>>> He added
>>> -> Support for literalVariable read/write bytecode
>>> -> Human readable Error Messages for the Parser that say more  
>>> then "SmaCCParserError".
>>> -> Support for Squeak 3.9 Pragmas / A solution for parsing  
>>> primitives (which used the old parser with
>>>              nobody realizing it... ;-) )
>>>         -> An experiment for a Smalltalk Scripting Language's  
>>> parser (Saphire)
>>> -> Many of smaller and larger bugfixes and improvements
>>>
>>> Thanks a lot!
>>>
>>> The NewCompiler framework has been used in many projects, both at  
>>> SCG and elsewhere.
>>>
>>> Next things on the TODO
>>> ====================
>>>
>>> Bugs
>>> -------
>>>
>>> - variables in the debugger seem to be not work correctly when  
>>> selecting them for a doit / inspect it.
>>> - in the debugger make "debug it" work. See  
>>> ParagraphEditor>>#debugIt.
>>> - fix debugIt to not compile a #doit method but use the same  
>>> mechanism as #doIt.
>>> - instVar vs. Field in IRBuilder. Something is very  not nice  
>>> here. We should provide both concepts.
>>> - Fix Compiler IR-->RBNode decompile
>>> - Make to Block primitives work for BlockClosure so that  
>>> exceptions (#ifCurtailed:, ...) work
>>>
>>> Enhancements
>>> --------------
>>>
>>> - Integrate RB with Newcompiler Parser.
>>> -- step one: Put Parser into it's own package (NewParser)... or  
>>> the AST package
>>> -- step two: remove RBParser, check/ fix RefactoringEngine
>>> -  work-around for bytecode limitations (jump offsett, number of  
>>> temps, number of literals ...) in IRBuilder
>>>    (see http://lists.squeakfoundation.org/pipermail/squeak-dev/ 
>>> 2006-February/100691.html)
>>> - Look at optimizing the Closures generated by the compiler. VM  
>>> hacking included.
>>> - VMmaker code generator should use RB AST
>>>
>>>
>>> Documentation
>>> ------------
>>> - We need websites for AST, RB, SmaCC, NewCompiler
>>> -- Write Hitchhikers Guide to the New Compiler:
>>>   http://www.whysmalltalk.com/articles/bykov/HitchHiker.htm
>>>   But then, these examples are even simpler to realise with  
>>> Persphone/Geppetto aka. "The Reflectivity"...
>>> -- IRBuilder: comments for public method, provide documentation  
>>> for stack effects
>>> - Better bugtracker. At least I loose track of the Mantis bugs  
>>> very easily. Can't it
>>>   generate emails to this list? Maybe use Roundup like  
>>> Reflectivity and Moose?
>>>   (http://macamis.unibe.ch/trackers)
>>>
>>
>