Message-Id:	<200105260916.f4Q9Go329723@hal.astr.lu.lv>
From:	Andris Pavenis <pavenis AT latnet DOT lv>
To:	djgpp-workers AT delorie DOT com
Subject:	Test binaries of gcc-3.0 20010523 for DJGPP
Date:	Sat, 26 May 2001 12:16:50 +0300
X-Mailer:	KMail [version 1.2.1]
MIME-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com

WARNING: IT'S NOT A RELEASE. IT'S ONLY RATHER RANDOMLY CHOOSED SNAPSHOT
OF gcc-3_0-branch. 
 
I have uploaded test binaries of gcc-3.0 prerelease updated with CVS in 23 May 2001
built for DJGPP. They are now in directory v2gnu/alphas:
 
v2gnu/alphas/readme-gcc30-prerelease
 
v2gnu/alphas/gcc30b.zip		C compiler binaries for DJGPP
v2gnu/alphas/gpp30b.zip		C++ compiler binaries for DJGPP
v2gnu/alphas/g7730b.zip		Fortran compiler binaries for DJGPP (untested)
v2gnu/alphas/objc30b.zip	Objective C compiler binaries forDJGPP (even more untested)
 
v2gnu/alphas/gcc30s2.zip	Some scripts and diffs against gcc30-20010523.tar.gz
 				
v2gnu/alphas/gcc30-20010523.tar.gz   Source archive. Next time I'll remove stuff not needed for
				DJGPP and the size will be about 10Mb
 
Binaries are built under Win98SE from rather randomly choosed CVS snapshot (I only looked
that I don't have too much testsuite failures under Linux). They works for me in DOS session 
under Win98SE. I renamed header files where needed and created header.gcc files to avoid
filename conflicts for binaries when there is no LFN support available. So I hope binaries will
work without LFN support, but I haven't tested that

Software requirements:
==================
 
binutils-2.11 or above 
 
djdev203.zip or current CVS version (recommended). I built binaries used CVS version updated
	in May 23 to build these test binaries, but they seems to work Ok also with 2.03. At least
	I didn't saw any difference while messing with libstdc++-v3 testsuite
 
Problems:
========
 
There were problems with iostream library as it was designed so that data size written to file
is the same as one in memory (but we have LF <===> CR LF conversion for text files). As result
iostream classes were seriously broken. I suceeded to workaround patch by patching
bits/std_fstream.h. It causes testsuite failure for filebuf but anyway I think it's better than 
wild seeks to wrong place. See gcc30s2.zip for diffs. Perhaps it would be nice if somebody
would look into that more deeply.

I had problems compiling libstdc++-v3/src/locale.cc. More exactly generated object code
contained infinite recursion in std::strchr(). Simplified test case was submitted to gcc bug
data base (c++/2853). I'm still afraid that the source of problem could be in binutils. 
I workarounded problem by patching gcc/sdbout.c (first diffs, the second one is from Mark,
see. c++/2606 for details)

Andris

- Raw text -