[Bug modula2/102342] New: gm2 testsuite failures for non-default multilib

[Bug modula2/102342] New: gm2 testsuite failures for non-default multilib

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102342

            Bug ID: 102342
           Summary: gm2 testsuite failures for non-default multilib
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: modula2
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
  Target Milestone: ---

When running the gm2 testsuite for non-default multilibs (e.g.
x86_64-pc-linux-gnu
for the default (-m64) and -m32, or i386-pc-solaris2.11 for the default -m32
and
-m64), many tests for the non-default multilib FAIL, e.g.

FAIL: gm2/calling-c/datatypes/unbounded/run/pass/m.mod compilation,  -g 
UNRESOLVED: gm2/calling-c/datatypes/unbounded/run/pass/m.mod execution,  -g 
(link failed)

/vol/gcc/bin/gld-2.35: i386:x86-64 architecture of input file
`/tmp/ccTE5Kph.a(c.o)' is incompatible with i386 output

While the main testcase is correctly compiled with the necessary multilib flags
(-m32 here), additional objects (c.o here) are not.  The common problem is that
those compilations are done manually with an explicit

set output [exec ${XGCC} -B[file dirname $rootme]/gcc -g -c
$srcdir/gm2/calling-c/datatypes/unbounded/run/pass/c.c -o c.o]

in gm2.exp, not taking multilib flags into account.

There are two ways to do this correctly.  One is to do the compilation with
target_compile like this:

-- gm2.exp      2021-07-08 14:07:03.408049578 +0000
+++ m2.exp      2021-09-15 11:50:06.344101169 +0000
@@ -29,14 +29,9 @@
 set gm2src ${srcdir}/../m2

 gm2_init_pim "${srcdir}/gm2/calling-c/datatypes/unbounded/run/pass"
-
-set XGCC [lookfor_file $rootme xgcc];
-
 gm2_link_with "c.o -lm2pim -lm2iso -lpthread"

-
-set output [exec rm -f c.o]
-set output [exec ${XGCC} -B[file dirname $rootme]/gcc -g -c
$srcdir/gm2/calling-c/datatypes/unbounded/run/pass/c.c -o c.o]
+set output [target_compile $srcdir/$subdir/c.c c.o object "-g"]

 foreach testcase [lsort [glob -nocomplain $srcdir/$subdir/*.mod]] {
     # If we're only testing specific files and this isn't one of them, skip
it.
@@ -47,5 +42,3 @@
     gm2_target_compile $srcdir/$subdir/m.mod m.o object "-g"
     gm2-torture-execute $testcase "" "pass"
 }
-
-set output [exec rm -f c.o]

The other, canonical one, is to use dg-additonal-sources instead, avoiding the
need for special driver files.

The above patch fixed the issue on i386-pc-solaris2.11 with -m32/-m64.

[Bug modula2/102342] gm2 testsuite failures for non-default multilib

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102342

--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> ---
Created attachment 51660
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51660&action=edit
proposed patch

While testing a recent version of the modula-2 branch
(386e7057d75043439f313085c4cbde8109459915),
I developed the attached patch to fully address this issue, following the
prove-of-concept one already included.  With this patch, x86_64-pc-linux-gnu
testresults (both -m64 and -m32 multilibs) are clean with the exception of

Running target unix/-m32
FAIL: gm2/cpp/pass/subaddr.mod,  -O
FAIL: gm2/cpp/pass/subaddr.mod,  -O -g
FAIL: gm2/cpp/pass/subaddr.mod,  -O3 -fomit-frame-pointer
FAIL: gm2/cpp/pass/subaddr.mod,  -O3 -fomit-frame-pointer -finline-functions
FAIL: gm2/cpp/pass/subaddr.mod,  -Os
FAIL: gm2/cpp/pass/subaddr.mod,  -g

which is preexisting (PR modula2/102343).  On i386-pc-solaris2.11, there are a
few more failures, also preexisting.

[Bug modula2/102342] gm2 testsuite failures for non-default multilib

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102342

Gaius Mulley <gaius at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gaius at gcc dot gnu.org
         Resolution|---                         |FIXED
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #2 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Many thanks for the patch - now applied to the modula2 branch of gcc-12.

[Bug modula2/102342] gm2 testsuite failures for non-default multilib

[ro at CeBiTec dot Uni-Bielefeld.DE]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102342

--- Comment #3 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> ---
> --- Comment #2 from Gaius Mulley <gaius at gcc dot gnu.org> ---
> Many thanks for the patch - now applied to the modula2 branch of gcc-12.

Thanks.  I can confirm that this worked fine on both
i386-pc-solaris2.11, i686-pc-linux-gnu, and x86_64-pc-linux-gnu.

Unfortunately, I hit quite a number of new issues breaking the build on
either or all of these targets.  I'll report them separately.