This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gcc 3.1 installed on hppa2.0w-hp-hpux11.00


I wrote:
> > I've successfully installed gcc 3.1 on hppa2.0w-hp-hpux11.00.

Harri Pasanen writes:
> I've taken a look at the test results through the build status page at:
> 
> http://gcc.gnu.org/gcc-3.1/buildstat.html
> 
> There seem to be a number of failures in the test cases, but as I'm not very 
> familiar with the test suite I don't know how to interpret the results.

Almost all of the unexpected failures seem to have to do with the handling
of weak symbols, mostly in corner cases.

However, going back and looking at the log, I notice that two compiler
crashes are hidden by XFAIL tests.  This is a nasty problem: the test
is presumably expected to fail because of missing functionality in the
compiler, the compiler is not expected to explode.

The two ICEs are for

gcc/testsuite/g++.old-deja/g++.oliva/typeof1.C:
 Internal compiler error in lookup_member, at cp/search.c:1411

gcc/testsuite/g++.old-deja/g++.pt/memtemp87.C:15
 Internal compiler error in grokdeclarator, at cp/decl.c:9914

I have the exact same two ICEs on GNU/Linux, so these appear to be
a processor-independent front end bugs.  They aren't relevant to the
failures you are seeing, I'm sure, since they will appear at any
optimization level.

The comment "crash test - XFAIL" appears in both tests, indicating that
whoever put that comment in knows that the compiler crashes, but marked it
XFAIL.  I don't think that we should be doing that.  I have no problem
with XFAIL if something other than an ICE happens.  For example, in
the lookup_member failure, instead of calling abort(), why can't the
compiler emit an error message and return the error node?  Something
similar could be done in grokdeclarator, it seems.

> We've been trying to compile OmniORB4 with gcc 3.1 on hpux, and it seems 
> that g++ crashes on some files if -O2 optimization is used, but without 
> optimization it works.
> 
> My question is, would it be useful to distil the compiler optimizer crash to 
> a test case, or is a similar failure mode already visible from the gcc 
> testsuite results on HP?

You have almost certainly found a new bug, unless the message you see is
identical to one of the two messages above.  In any case, please file
a bug report: see http://gcc.gnu.org/bugs.html for instructions.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]