Re: [PATCH] Ignore dynamic references on forced local symbols

["Maciej W. Rozycki" <macro at mips dot com>]

H.J.,

On Fri, 12 Jan 2018, Alan Modra wrote:

> > >>       PR ld/22649
> > >>       * testsuite/ld-elf/pr22649-1.s: New file.
> > >>       * testsuite/ld-elf/pr22649-2a.s: Likewise.
> > >>       * testsuite/ld-elf/pr22649-2b.s: Likewise.
> > >>       * testsuite/ld-elf/pr22649.msg: Likewise.
> > >>       * testsuite/ld-elf/shared.exp: Run ld/22649 tests.
> > >
> > > OK.
> > >
> > 
> > I've noticed that 2 of the new tests fail:
> > FAIL:    ld:Build pr22649-2a.so
> > FAIL:    ld:Build pr22649-2b.so
> > 
> > on arm*linux* and arm-none-nacl targets
> > 
> > Can you check?
> 
> I have a patch to fix this under test at the moment.

 This is still broken for several MIPS targets:

.../ld/ld-new: Removing unused section '.reginfo' in file 'tmpdir/pr22649-2b.o'
.../ld/ld-new: Removing unused section '.reginfo' in file 'tmpdir/pr22649-2b.o'
FAIL: Build pr22649-2c.so

and:

.../ld/ld-new: Removing unused section '.reginfo' in file 'tmpdir/pr22649-2b.o'
.../ld/ld-new: Removing unused section '.reginfo' in file 'tmpdir/pr22649-2b.o'
FAIL: Build pr22649-2d.so

Unfortunately the change description is terse enough I cannot figure out 
how to address this properly.  E.g. is it OK to have:

#pass

as expected output from LD (and possibly verify afterwards with `objdump' 
or `readelf' that the section in question has been retained)?  Or can you 
make the test cases use a proper linker script selecting input and output 
concerned and letting target maintainers define target-specific data to 
/DISCARD/?  Or finally, should the test cases perhaps just use:

#failif
...

on the unwanted diagnostic output reporting section removal as noted in 
the PR?

 Would you please run proper cross-target testing with your generic 
changes you commit?  It doesn't require much effort, just some machine 
processing time and it would let you avoid introducing regressions or new 
failures due to mistakes in test cases.  Also I have asked you to do that 
before already I believe.

  Maciej