Re: gcc 3.4.0 testsuite gives over 5000 g++ errors

[llewelly at xmission dot com]

"Paul C. Leopardi" <leopardi@bigpond.net.au> writes:

> Hi,
> Answers and many questions below.
> Best regards
> 
> On Tuesday 04 May 2004 02:08, llewelly@xmission.com wrote:
> > "Paul C. Leopardi" <leopardi@bigpond.net.au> writes:
> > > Hi,
> > > I checked versions as you suggested.
> > > runtest -V
> > > Expect version is       5.38.0
> > > Tcl version is          8.4
> > > Framework version is    1.4.3
> > >
> > > I also rebuilt gcc 3.4.0 and re-ran the tests.
> > > http://gcc.gnu.org/ml/gcc-testresults/2004-05/msg00137.html
> > >
> > > I don't understand why the heading is "Results for 3.3.2 20031216
> > > (prerelease)
> >
> > This is really strange. I guess the dejagnu is running you old g++
> >     3.3.2 instead of the g++ you just built. I can't imagine *why*,
> >     however. Maybe an alias is interfering?
> 
> I tried the test suite again, after renaming /usr/local/bin 
> to /usr/local/bin.old, and the test used /usr/bin/g++ this time.
> 
> 		=== g++ Summary ===
> 
> # of expected passes		8454
> # of unexpected failures	1404
> # of unexpected successes	7
> # of expected failures		61
> # of unresolved testcases	15
> # of untested testcases		30
> # of unsupported tests		68
> /usr/bin/g++ version 3.3.2 20031216 (prerelease) (SuSE Linux)
> 
> It turns out that I had colorgcc installed and the SuSE rpm had put links 
> in /usr/local/bin from eg g++ to colorgcc.

I haven't had colorgcc installed on any machine I own for over 3
    years, so I had totally forgotten about it. But this seems to be
    only part of the problem; it shouldn't have been using
    /usr/local/bin/g++ in the first place.

> 
> I uninstalled colorgcc, renamed gcc-3.4.0.obj and ran configure and "make 
> bootstrap" in a fresh directory, then ran "make -k check" again.
> 
>                 === g++ Summary ===
> 
> # of expected passes            8455
> # of unexpected failures        1403
> # of unexpected successes       7
> # of expected failures          61
> # of unresolved testcases       15
> # of untested testcases         30
> # of unsupported tests          68
> /usr/bin/g++ version 3.3.2 20031216 (prerelease) (SuSE Linux)

This is completely mindboggling. Unless you have colorgcc installed in
    /usr/bin too?

> 
> I think the build must have failed in some cryptic way. 
> The output from "make bootstrap" is 12.9MB in size!

Maybe I'm jumping at shadows, but that sounds a little large ... I
    have  7028873 bytes (< 7MB) for the output of make boostrap-lean
    of 3.4.0 (c, objc, c++, java, f77) on i686-freeBSD5.2 .

> How do I check it to see 
> whether it succeeded? I thought that if it went all the way to the end it 
> must have succeeded?

I still think it is more likely that something went wrong with 
    make check than with make bootstrap. Remember, almost all of the
    libstdc++ tests succeed, and those use the 3.4.0 you just
    built. So I *think* the newly built g++ binary is ok. It just isn't
    getting run by 'make check'.

> No?
> Where does g++ get built to?

It should be in objdir/gcc . 

> What file am I looking for?
> How does runtest know where to find g++?

I don't know. I seem to lack dejagnu info files for some reason. Could
    you try aliasing 'runtest' to 'runtest -v' ?

> Sorry for all the questions. I am such a noob. Maybe this is all documented 
> somewhere?
> Should I direct these questions to the main gcc mailing list, or one of the 
> suse mailing lists?
> 
> > > (SuSE Linux) testsuite on x86_64-suse-linux-gnu". I thought I was testing
> > > 3.4.0? The test results themselves say, "LAST_UPDATED: Obtained from CVS:
> > > -rgcc_3_4_0_release".
> > >
> > > Anyway, the result is about the same. Over 5000 g++ errors.
> > >
> > > I must have something configured wrong. What could it be?
> >
> > [snip]