[PATCH] Fix libstdc++ tests requiring atomic support on hppa-hpux

[Jonathan Wakely]

On 11/03/19 13:54 +0000, Jonathan Wakely wrote:
>On 11/03/19 09:27 -0400, John David Anglin wrote:
>>On 2019-03-11 9:16 a.m., Jonathan Wakely wrote:
>>>On 11/03/19 14:13 +0100, Andreas Schwab wrote:
>>>>On Mär 11 2019, Jonathan Wakely <jwakely@redhat.com> wrote:
>>>>
>>>>>What do you think about adding the following?
>>>>>
>>>>>--- a/libstdc++-v3/testsuite/lib/dg-options.exp
>>>>>+++ b/libstdc++-v3/testsuite/lib/dg-options.exp
>>>>>@@ -257,6 +257,15 @@ proc add_options_for_net_ts { flags } {
>>>>>Â Â Â  return $flags
>>>>>}
>>>>>
>>>>>+# Add to FLAGS all the target-specific flags to link to libatomic, if required.
>>>>>+
>>>>>+proc add_options_for_libatomic { flags } {
>>>>>+Â Â Â  if { [istarget hppa*-*-hpux*] } {
>>>>>+Â Â Â Â Â Â  return "$flags -L../../libatomic/.libs -latomic"
>>>>>+Â Â Â  }
>>>>>+Â Â Â  return $flags
>>>>>+}
>>>>
>>>>That should probably also be enabled for riscv*-*-*.
>>>
>>>So all the more reason to do it this way, so the list of targets is
>>>only maintained in one place
>>Your suggestion looks good to me.  We still need hunk that sets up LD_LIBRARY_PATH.
>
>Right.
>
>>This is what is done in fortran tests:
>>! { dg-additional-options "-latomic" { target libatomic_available } }
>
>That adds it for every target that has a libatomic though, right? We
>have tests that check certain uses of atomics work without libatomic,
>which would no longer check anything if we added that to them :-)

Ah, but unless -L../../libatomic./libs has been added to the flags,
-latomic won't find anything, so it wouldn't get added.

I still think {target libatomic_available } doesn't buy us much, due
to ...

>If we use { dg-add-options libatomic } it'll only be added for the
>listed targets (hpux and riscv to begin with) and if somebody disables
>building libatomic on those targets they probably should get test
>failures to say there's something wrong.
>