[PATCH] ld: testsuite: Skip "Run with libpr19553c.so" test on Solaris

[Rainer Orth]

Hi Jan,

> On 28.07.2025 11:49, Rainer Orth wrote:
>> The
>> 
>> FAIL: Run with libpr19553c.so
>> 
>> test FAILs on Solaris (32 and 64-bit, sparc and x86):
>> 
>> Running: tmpdir/pr19553c > tmpdir/pr19553c.out
>> diff tmpdir/pr19553c.out /vol/src/gnu/binutils/hg/master/local/ld/testsuite/ld-elf/pr19553c.out
>> 1c1
>> < pr19553b
>> ---
>>> pr19553c
>> child process exited abnormally
>> 
>> The test uses .symver, resulting in versioned symbols which the Solaris
>> ld.so.1 doesn't support and never will.  Running it with LD_DEBUG=all
>> shows
>> 
>> 26493: 1: symbol=foo;  lookup in file=tmpdir/pr19553c  [ ELF ]
>> 26493: 1: symbol=foo;  lookup in file=tmpdir/libpr19553c.so  [ ELF ]
>> 26493: 1: symbol=foo;  skipping entry in file=tmpdir/libpr19553c.so, index[7], version=FOO, due to GNU version hidden bit
>> 26493: 1: symbol=foo;  continuing lookup in file=tmpdir/libpr19553c.so  [ ELF ]
>> 26493: 1: symbol=foo;  lookup in file=tmpdir/libpr19553b.so  [ ELF ]
>> 26493: 1: binding file=tmpdir/pr19553c to file=tmpdir/libpr19553b.so: symbol 'foo'
>> so this patch skips the test.
>> 
>> Ok for trunk?
>
> With the patch subject changed to match this email's subject: Yes. Yet better

Certainly: I always take the subject from the mail.  In this case, I'd
meant to just xfail the test until I realized that there's no easy way
to do so.

> would be to keep the tests in sorted order (i.e. c not pulled out of the middle
> of b and d), but I can see how this would cause some extra churn (more or less
> depending on how the issue would be tackled).

I can do so if you prefer: in another (yet not submitted patch) I've
used something like

# PR 33177
if { ![istarget *-*-solaris2*] } {
    lappend run_tests \
[...]
}

lappend run_tests \
[...]

which keeps the order and isn't too intrusive.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University