Re: Small update to reversed_comparison_code

[Mike Stump <mrs at windriver dot com>]

> Date: Tue, 13 Mar 2001 17:22:40 +0000 (GMT)
> From: "Joseph S. Myers" <jsm28@cam.ac.uk>
> To: Mark Mitchell <mark@codesourcery.com>
> cc: <gcc@gcc.gnu.org>

> As a start, automated checkers that point out regressions on the primary
> platforms are needed.

I have machines, I have the diskspace.  They can check whatever I/we
want.  The problem is the lack of software.  What I would like, is a
single target that does whatever people want.

   make test, or maybe contrib/test

Given that, I would easily enough run it, and if it wanted to binary
search cvs regressions, or test an experimental branch, or collect
performance numbers, or ....; it could.  Usually, I just have them
run:

#!/usr/local/gnu/bin/bash

if [ -f Makefile ]; then nice make -k realclean; fi

( cd $HOME/net/gcc &&
  $HOME/net/gcc/contrib/gcc_update -A -d; )

ld=/usr/ccs/bin/ld
echo ld is $ld

if [ ! -f Makefile ]; then
    LD=$ld ../gcc/configure --prefix=$HOME/packages/gcc-`date "+%Y%m%d"`
fi

nice make -j5 LD=$ld LD_FOR_TARGET=$ld bootstrap &&
 { nice make LD=$ld -k check
   $HOME/net/gcc/contrib/test_summary | sh -x
 }

but notice the lack of reporting build errors that this script has.
Notice the lack of comparing two runs that it could do with
contrib/compare_tests, or diffing the ChangeLog that it could do, as
the other regression checker does.  I think we should have another
directory devoted to the testing of gcc, one that would not be in the
main release, but that people that want to devote test cycles to the
project could use.