[testsuite] note pitfall in how outputs.exp sets gld

[testsuite] note pitfall in how outputs.exp sets gld

[Alexandre Oliva]

This patch documents a glitch in gcc.misc-tests/outputs.exp: it checks
whether the linker is GNU ld, and uses that to decide whether to
expect collect2 to create .ld1_args files under -save-temps, but
collect2 bases that decision on whether HAVE_GNU_LD is set, which may
be false zero if the linker in use is GNU ld.  Configuring
--with-gnu-ld fixes this misalignment.  Without that, atsave tests are
likely to fail, because without HAVE_GNU_LD, collect2 won't use @file
syntax to run the linker (so it won't create .ld1_args files).

Long version: HAVE_GNU_LD is set when (i) DEFAULT_LINKER is set during
configure, pointing at GNU ld; (ii) --with-gnu-ld is passed to
configure; or (iii) config.gcc sets gnu_ld=yes.  If a port doesn't set
gnu_ld, and the toolchain isn't configured so as to assume GNU ld,
configure and thus collect2 conservatively assume the linker doesn't
support @file arguments.

But outputs.exp can't see how configure set HAVE_GNU_LD (it may be
used to test an installed compiler), and upon finding that the linker
used by the compiler is GNU ld, it will expect collect2 to use @file
arguments when running the linker.  If that assumption doesn't hold,
atsave tests will fail.

Does it make sense to put this in?  I'd like to preserve this knowledge
somehow, and I suppose this would be most useful for someone observing
these failures and trying to figure out why they come about, so this
seems the best place for them.  Ok to install?


for  gcc/testsuite/ChangeLog

	* outputs.exp (gld): Note a known mismatch and record a
	workaround.
---
 gcc/testsuite/gcc.misc-tests/outputs.exp |   10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/gcc/testsuite/gcc.misc-tests/outputs.exp b/gcc/testsuite/gcc.misc-tests/outputs.exp
index 7ee355034ccaf..9f44cbdb0b5d5 100644
--- a/gcc/testsuite/gcc.misc-tests/outputs.exp
+++ b/gcc/testsuite/gcc.misc-tests/outputs.exp
@@ -50,7 +50,15 @@ if !$skip_lto {
     set ltop [check_linker_plugin_available]
 }
 
-# Check for GNU LD.  Some files like .ld1_args depend on this.
+# Check for GNU LD.  Some files like .ld1_args depend on this.  This
+# should really be testing whether HAVE_GNU_LD was set by configure.
+# If we find GNU ld here, but the compiler wasn't configured
+# --with-gnu-ld or with DEFAULT_LINKER pointing at GNU ld, on a target
+# that doesn't set gnu_ld=yes unconditionally, configure and thus
+# collect2 will conservatively assume there's no support for @file in
+# the linker, but our atfile tests will expect ld1_args files to be
+# created, and thus fail.  Configuring the compiler --with-gnu-ld
+# fixes this.
 set gld [check_effective_target_gld]
 
 # Prepare additional options to be used for linking.

-- 
Alexandre Oliva, happy hacker                https://FSFLA.org/blogs/lxo/
   Free Software Activist                       GNU Toolchain Engineer
Disinformation flourishes because many people care deeply about injustice
but very few check the facts.  Ask me about <https://stallmansupport.org>

Re: [testsuite] note pitfall in how outputs.exp sets gld

[Mike Stump]

On Jun 22, 2023, at 10:35 PM, Alexandre Oliva <oliva@adacore.com> wrote:
> 
> This patch documents a glitch in gcc.misc-tests/outputs.exp: it checks
> whether the linker is GNU ld, and uses that to decide whether to
> expect collect2 to create .ld1_args files under -save-temps, but
> collect2 bases that decision on whether HAVE_GNU_LD is set, which may
> be false zero if the linker in use is GNU ld.  Configuring
> --with-gnu-ld fixes this misalignment.  Without that, atsave tests are
> likely to fail, because without HAVE_GNU_LD, collect2 won't use @file
> syntax to run the linker (so it won't create .ld1_args files).
> 
> Long version: HAVE_GNU_LD is set when (i) DEFAULT_LINKER is set during
> configure, pointing at GNU ld; (ii) --with-gnu-ld is passed to
> configure; or (iii) config.gcc sets gnu_ld=yes.  If a port doesn't set
> gnu_ld, and the toolchain isn't configured so as to assume GNU ld,
> configure and thus collect2 conservatively assume the linker doesn't
> support @file arguments.
> 
> But outputs.exp can't see how configure set HAVE_GNU_LD (it may be
> used to test an installed compiler), and upon finding that the linker
> used by the compiler is GNU ld, it will expect collect2 to use @file
> arguments when running the linker.  If that assumption doesn't hold,
> atsave tests will fail.
> 
> Does it make sense to put this in?  I'd like to preserve this knowledge
> somehow, and I suppose this would be most useful for someone observing
> these failures and trying to figure out why they come about, so this
> seems the best place for them.  Ok to install?

Ok.