Re: [cdt-dev] Torture tests (was Re: Applied [HEAD]: Second attempt at [

[John Camelon <jcamelon@xxxxxxxxxx>]

I inspected some of the failed tests
as well and saw things that were definitely not ANSI that gcc gobbles up.
 Our support of ANSI C99 hasn't been as thorough as ANSI C++ but it
should be.   

I would say for the time being, our support is limited to the latest ANSIs
and also GNU, as its our preferred platform.  K&R is the only
exception to this rule.  

The problem with supporting too many
variants in the same parser is that performance suffers due to excessive
backtracking and investigating paths that never work, and the code becomes
obfuscated.  I would hope that we could subclass parsers and then
add more options to ParserFactory.createParser() to allow for the client
to choose the variant.  This may be something else that needs to be
managed through the build extension point with the
IScannerInfoProvider family of interfaces.


If you plan on putting anything in the
current parser to help it accept something that is GNU but non-ANSI, please
mark it with a comment so that way if we restructure things later we'll
know where to look.  

JohnC

"Victor Mozgin" <vic_mozgin@xxxxxxxxxxx>
Sent by: cdt-dev-admin@xxxxxxxxxxx

07/03/2003 03:28 PM

Please respond to
cdt-dev@xxxxxxxxxxx

To	<cdt-dev@xxxxxxxxxxx>
cc
Subject	[cdt-dev] Torture tests (was Re: Applied [HEAD]: Second attempt at [cdt-patch] Fix for PR 39501)

        That was the
idea. I took some time to analyze ~400 failures we had before, and added
filters for the things that are not relevant. Now all these ~100 failures
are real problems that need to be fixed. All of them are reported in bugzilla
(you've probably seen ~25 new parser/scanner PRs ;) ), except some template
instantiation/specialization cases I am not sure about. So after all these
cases are resolved, the failure count will be about zero. I am planning
to add simplified tests for these PRs into FailedTests, so it won't be
necessary to run TortureTest all the time.
        Now, question:
what C/C++ variants do we want to support in CDT? For instance, there are
K&R C, C89/C90, C94/C95, C9X/C99. C99 has several features that CDT
does not support yet...
 
        /Vic
----- Original Message ----- 
From: John
Camelon 
To: cdt-patch@xxxxxxxxxxx

Sent: Thursday, July 03, 2003 8:51 PM
Subject: Applied [HEAD]: Second attempt at [cdt-patch]
Fix for PR 39501


Thanks for the help Victor.  That fix gets us down to under 100 failures
out of ~6900 tests in TortureTest.   
I try and run TortureTest in quick parse mode on a regular basis now, I
look forward to when we can get the failure count down to zero so that
I can include it in my regression for validating parser patches.

JohnC

"Victor Mozgin" <vic_mozgin@xxxxxxxxxxx>
Sent by: cdt-patch-admin@xxxxxxxxxxx

07/03/2003 01:23 PM

Please respond to
cdt-patch@xxxxxxxxxxx

To	<cdt-patch@xxxxxxxxxxx>
cc
Subject	Re: Rejected: [cdt-patch] Fix for PR 39501

        That wasn't difficult (it wasn't used anyway).
But are platform requirements documented? Is it 1.3 that should be supported,
or anything earlier, or?.. 
  
            /Vic 
----- Original Message ----- 
From: John
Camelon 
To: cdt-patch@xxxxxxxxxxx
Sent: Thursday, July 03, 2003 5:31 PM 
Subject: Rejected: [cdt-patch] Fix for PR 39501 


I'm sorry Victor, I cannot apply this patch.  In the core.tests you
use java.util.regexp which is exclusively a Java 1.4 package.  
As far as I know, we still have to support Java 1.3.  

Hopefully it won't be too hard for you to fix ... 

JohnC

"Victor Mozgin" <vic_mozgin@xxxxxxxxxxx>
Sent by: cdt-patch-admin@xxxxxxxxxxx

07/02/2003 12:00 PM

Please respond to
cdt-patch@xxxxxxxxxxx

To	<cdt-patch@xxxxxxxxxxx>
cc
Subject	[cdt-patch] Fix for PR 39501

   - Fixed PR 39501 : Parser problems with throw clauses.
  - Improved filtering of expected failures/inconclusives in TortureTest.

          /Vic[attachment "core20030702.patch"
deleted by John Camelon/Ottawa/IBM] [attachment "core.tests20030702.patch"
deleted by John Camelon/Ottawa/IBM]