[Bug d/115295] New: [15 regression] Various gdc testsuite regressions

[Bug d/115295] New: [15 regression] Various gdc testsuite regressions

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

            Bug ID: 115295
           Summary: [15 regression] Various gdc testsuite regressions
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: d
          Assignee: ibuclaw at gdcproject dot org
          Reporter: ro at gcc dot gnu.org
  Target Milestone: ---

Between 20240529 (18f477980c8597fe3dca2c2e8bd533c0c2b17aa6) and 20240530
(71899d6b4e18145f005046681208db0137c95ede),
quite a number of gdc.* tests regressed:

+FAIL: gdc.dg/attr_module.d    (test for warnings, line 12)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 18)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 24)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 36)
+FAIL: gdc.dg/attr_module.d   (test for excess errors)

Excess errors:
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:6:2: error:
undefined identifier 'value_ignored'
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:12:2: error:
undefined identifier 'type_symbol'
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:18:2: error:
undefined identifier 'template_symbol'
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:24:2: error:
undefined identifier 'struct_wrong_field'
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:30:2: error:
undefined identifier 'struct_void_init'
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/attr_module.d:36:2: error:
undefined identifier 'unknown_attribute'

+FAIL: gdc.dg/gdc231.d   (test for excess errors)

Excess errors:
/vol/gcc/src/hg/master/local/gcc/testsuite/gdc.dg/gdc231.d:5:8: error: unable
to read module 'gdc231'
import path[0] =
/var/gcc/regression/master/11.4-gcc/build/i386-pc-solaris2.11/libphobos/libdruntime
import path[1] = /vol/gcc/src/hg/master/local/libphobos/libdruntime
import path[2] = /vol/gcc/src/hg/master/local/libphobos/src

+FAIL: gdc.dg/gdc253a.d   (test for excess errors)
+FAIL: gdc.dg/gdc27.d   (test for excess errors)
+FAIL: gdc.dg/pr108050.d   (test for excess errors)
+FAIL: gdc.test/compilable/test16348.d   output-exists test16348.s
+FAIL: gdc.test/compilable/test17541.d   output-exists test17541.s

compilable/test17541.d:8:8: error: unable to read module 'two'
compilable/test17541.d:8:8: note: Expected 'two.d' or 'two/package.d' in one of
the following import paths:
import path[0] =
/var/gcc/regression/master/11.4-gcc/build/i386-pc-solaris2.11/libphobos/libdruntime
import path[1] = /vol/gcc/src/hg/master/local/libphobos/libdruntime
import path[2] = /vol/gcc/src/hg/master/local/libphobos/src
import path[3] =
/var/gcc/regression/master/11.4-gcc/build/gcc/testsuite/gdc3/compilable
compiler exited with status 1

+FAIL: gdc.test/fail_compilation/diag4479.d   output-exists-not diag4479.o
+FAIL: gdc.test/fail_compilation/fail11453a.d   output-exists-not fail11453a.o
+FAIL: gdc.test/fail_compilation/fail11453b.d   output-exists-not fail11453b.o
+FAIL: gdc.test/fail_compilation/fail18938.d   output-exists-not fail18938.o
+FAIL: gdc.test/fail_compilation/fail19911b.d   output-exists-not fail19911b.o
+FAIL: gdc.test/fail_compilation/fail23109.d   output-exists-not fail23109.o
+FAIL: gdc.test/fail_compilation/ice10598.d   output-exists-not ice10598.o
+FAIL: gdc.test/fail_compilation/ice11136.d   output-exists-not ice11136.o
+FAIL: gdc.test/fail_compilation/ice14116.d   output-exists-not ice14116.o
+FAIL: gdc.test/fail_compilation/no_Throwable.d   output-exists-not
no_Throwable.o
+FAIL: gdc.test/fail_compilation/no_TypeInfo.d   output-exists-not
no_TypeInfo.o
+FAIL: gdc.test/fail_compilation/verifyhookexist.d   output-exists-not
verifyhookexist.o

I'm seeing this on at least Solaris/x86 (32 and 64-bit) and Linux/x86_64 (64
and 32-bit).

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |15.0

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ibuclaw at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-05-31
           Assignee|ibuclaw at gdcproject dot org      |aoliva at gcc dot gnu.org

--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> ---
I've now run a reghunt, which again found this patch as the culprit:

commit bdc264a16e327c63d133131a695a202fbbc0a6a0
Author: Alexandre Oliva <oliva@adacore.com>
Date:   Thu May 30 02:06:48 2024 -0300

    [testsuite] conditionalize dg-additional-sources on target and type

Taking gdc.dg/attr_module.d, which I've use as test for the reghunt, I see

-PASS: gdc.dg/attr_module.d    (test for warnings, line 12)
-PASS: gdc.dg/attr_module.d    (test for warnings, line 18)
-PASS: gdc.dg/attr_module.d    (test for warnings, line 24)
-PASS: gdc.dg/attr_module.d    (test for warnings, line 36)
-PASS: gdc.dg/attr_module.d   (test for excess errors)
+UNSUPPORTED: imports/attributes.d: additional-source will not be used to build
attr_module.s
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 12)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 18)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 24)
+FAIL: gdc.dg/attr_module.d    (test for warnings, line 36)
+FAIL: gdc.dg/attr_module.d   (test for excess errors)

when comparing the gdc.sum files.

Please fix.  This patch should really have gotten way more testing before
committing!

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[aoliva at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

--- Comment #2 from Alexandre Oliva <aoliva at gcc dot gnu.org> ---
Ugh, it looks like D deviates from one of the fundamental assumptions behind
the change, namely, that for each named source file, the compiler would attempt
to generate one (set of) output files, so when not linking, an explicitly named
output file would be rejected by the compiler.

I suppose we could make the new behavior of omitting additional sources in
compile tests dependent on the input language.

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[ibuclaw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

--- Comment #3 from Iain Buclaw <ibuclaw at gcc dot gnu.org> ---
(In reply to Alexandre Oliva from comment #2)
> Ugh, it looks like D deviates from one of the fundamental assumptions behind
> the change, namely, that for each named source file, the compiler would
> attempt to generate one (set of) output files, so when not linking, an
> explicitly named output file would be rejected by the compiler.
> 
> I suppose we could make the new behavior of omitting additional sources in
> compile tests dependent on the input language.

This is because D (and at a glance Go and Rust too) sets this setting in
lang-specs.h

```
int combinable;       /* If nonzero, compiler can deal with
                         multiple source files at once (IMA).  */
```

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[aoliva at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

--- Comment #4 from Alexandre Oliva <aoliva at gcc dot gnu.org> ---
Created attachment 58410
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58410&action=edit
candidate patch

This patch reverts dg-additional-sources to the earlier behavior, in which
sources are added regardless of the test, introducing an explicit flag to
activate the link-only mode.  This should bring the D testsuite back to a
functioning state (I'd appreciate if someone who has a D compiler could give it
a spin; I'm new to it), and the additional of the flag to the pr95401.cc test
should address the original problem.

[Bug d/115295] [15 regression] Various gdc testsuite regressions

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115295

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Alexandre Oliva from comment #4)
> Created attachment 58410 [details]
> candidate patch
> 
> This patch reverts dg-additional-sources to the earlier behavior, in which
> sources are added regardless of the test, introducing an explicit flag to
> activate the link-only mode.  This should bring the D testsuite back to a
> functioning state (I'd appreciate if someone who has a D compiler could give
> it a spin; I'm new to it), and the additional of the flag to the pr95401.cc
> test should address the original problem.

So I fixed the other `dg-additional-sources` issues in the vect testsuite by
adding `dg-do run` to them (PR 113899, r14-8978-g948dbc5ee45f9f). I suspect we
should just do the same for g++.dg/vect/pr95401.cc . I had missed the c++
testsuite's vect testcases when I was looking for dg-additional-sources when I
fixed that one.