Re: SSI crashes on Win32.

[Graham Dumpleton <gr...@dscpl.com.au>]

CC'd to developers list so others are aware of issue and maybe can offer
insight.

Nicolas Lehuen wrote ..
> Hi Graham,
> 
> The latest trunk version yields multiple segfaults and failures in
> different places :
> 
>   * Testing req.add_handler() for empty phase
> E
>   * Testing req.add_handler() directory
> E
>   * Testing interpreter per directive
> E

I'll review my changes visually to make sure I didn't miss something
obvious like variable initialisation or something. It is a pain when it
works for me and then dies on another platform. :-(

>   * Testing phase status
> F

If you can send me section from error log for this one would help
as specifically log stuff at each point so I can track that handlers
where executed in order as expected. Would expect to see:

[Tue May 09 09:10:25 2006] [error] phase_status_1
[Tue May 09 09:10:25 2006] [error] phase_status_2
[Tue May 09 09:10:25 2006] [error] phase_status_4
[Tue May 09 09:10:25 2006] [error] phase_status_5
[Tue May 09 09:10:25 2006] [error] phase_status_6
[Tue May 09 09:10:25 2006] [error] phase_status_7

>   * Testing server side include
> F

Based on the logs below, this may be what I suspected. That is there
is some issue with CR/LF differences on Win32 which causing the
code some grief. Because of the crash by Apache when exception
occurred in SSI code, real issue was probably being masked. I'll
do some more playing with CR/LF differences on UNIX and see if
I can recreate it.

BTW, when I first add ssi.shtml to subversion, I didn't set eol:native
property. It was added by someone later but probably after you checked
it out. Not sure if you need to remove that file and check it out again
to ensure things okay. But then, in release tar ball it is always going
to be be UNIX style EOLs anyway.

Thanks for checking for me after I made these latest changes.

Graham

> The E correspond to segfaults. The child process terminates and a new
> one is relaunched by the parent process, so that the other tests can
> continue.
> 
> The SSI bug gives this in the error log :
> 
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml: Traceback (most recent
> call last):
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml:   File
> "E:\\Python24\\Lib\\site-packages\\mod_python\\apache.py", line 459,
> in IncludeDispatch\n    exec(code, filter.req.ssi_globals)
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml:   File "<string>", line 1
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml:
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml:     ^
> [Mon May 08 15:35:07 2006] [error] includes
> C:/projets/mod_python/test/htdocs/ssi.shtml: SyntaxError: invalid
> syntax
> 
> As for the segfaults, I'll try to make a debug build of mod_python and
> launch the MSVC debugger. Right now I can debug the Apache code but
> not the mod_python one, so I must find how to coerce the build system
> into making a debug build.
> 
> Regards,
> Nicolas
> 
> 
> 2006/5/8, Graham Dumpleton <gr...@dscpl.com.au>:
> > Nicolas, can you try the test suite on subversion trunk on Win32 and
> > see if crash
> > in SSI test still occurs. If it doesn't crash, you may find that SSI
> > test still fails and
> > hopefully the error log will show the reason why, whereas before it
> > may not have
> > got a chance to flush out error before Apache died.
> >
> > Thanks in advance.
> >
> > Graham
> >