Re: more glibc 2.0.109 fails on linuxppc

[Zack Weinberg <zack@rabi.columbia.edu>]

On Thu, 31 Dec 1998 00:13:07 -0500, "Jack Howarth" wrote:
>Zack,
>    Tonight to make sure that the newer linux 2.2.0-pre1 weren't causing
>problems with the glibc 2.0.109 build on linuxppc I decided to rebuild
>glibc 2.0.108 using these newer linuxppc headers. The resulting glibc
>passed all the standard make check tests except the known bug in 2.0.108
>which gives...

>runtest: line 94: 30225 Segmentation fault      $PROG -f $from -t $t $testdata
>/$from -o $temp1 2>&1  
>FAIL: iconvdata.30

This is the same as the iconv failure with the old testsuite.

># of expected passes            446
># of unexpected failures        9
># of expected failures          51
># of untested testcases         567
># of unsupported testcases      3

Your untested count is so high because the math tests failed to compile. 
Over half of the tests are of the math library (the testsuite's coverage is
very spotty).

I bet you have compiler problems.  Support for PPC in egcs has been evolving
rapidly, you may want to try a recent development snapshot.  This may also
fix the iconv problem.

In your build tree, in testsuite/libc.log, is a log of all the messages
produced by the test runs.  The compiler errors from the math test will be
in there.  Odds are it's an "internal compiler error" of some sort.  Can you
check that, please?  (Don't send me the whole file, it's huge.)

zw