[Bug fortran/95847] New: Internal error when processing pFUnit generated files with --coverage

[Bug fortran/95847] New: Internal error when processing pFUnit generated files with --coverage

[olebole at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

            Bug ID: 95847
           Summary: Internal error when processing pFUnit generated files
                    with --coverage
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: olebole at debian dot org
  Target Milestone: ---

Created attachment 48776
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48776&action=edit
Source file that reproduces the bug

With gcc 9 and 10, compiling the attached file with coverage info raises an
internal error:

$ gfortran-10 -c --coverage funit-bug.f90 
during IPA pass: profile
funit-bug.f90:8:0:

    8 |    use foo
      | 
internal compiler error: in coverage_begin_function, at coverage.c:656
0x7f6c0275ae0a __libc_start_main
        ../csu/libc-start.c:308
[…]
$ gfortran-10 --version
GNU Fortran (Debian 10.1.0-3) 10.1.0
[…]
$ gfortran-9 --version
GNU Fortran (Debian 9.3.0-14) 9.3.0

This does not happen with gfortran-8, so it is a regression:

$ gfortran-8 -c --coverage funit-bug.f90
$ gfortran-8 --version
GNU Fortran (Debian 8.4.0-4) 8.4.0

System is Debian 11 testing.
The attached file is derived from a pFUnit generated file where this was
originally observed <https://github.com/Goddard-Fortran-Ecosystem/pFUnit>.

[Bug gcov-profile/95847] Internal error when processing pFUnit generated files with --coverage

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-06-23
     Ever confirmed|0                           |1

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
The ICE is present in a significant number of tests in the test suite
(run make -k -j16 check-gfortran RUNTESTFLAGS="--target_board=unix
--tool_opts='--coverage'" in the gcc directory).

The change occurred between 2018-11-17 (OK) and 2018-11-24 (ICE).

[Bug gcov-profile/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |9.4
            Summary|Internal error when         |[9/10/11 Regression]
                   |processing pFUnit generated |Internal error when
                   |files with --coverage       |processing pFUnit generated
                   |                            |files with --coverage
           Keywords|                            |ice-on-valid-code

[Bug gcov-profile/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[johannes.ziegenbalg at mailbox dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

johannes.ziegenbalg at mailbox dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |johannes.ziegenbalg@mailbox
                   |                            |.org

--- Comment #2 from johannes.ziegenbalg at mailbox dot org ---
I get the same bug with GCC 10.2.0 in one of my c++ test-cases.
I also tested GCC 11 (GNU Fortran (GCC) 11.0.0 20201029) with the attached file
and got the same result.

After a little digging I found out that the expanded region for 'sbr' starts at
line 8 (`use foo`) and ends at line 4 (`end subroutine sbr`).

But I have no idea why `expand_location` returns line 8.

[Bug gcov-profile/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[johannes.ziegenbalg at mailbox dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #3 from johannes.ziegenbalg at mailbox dot org ---
I get the same bug with GCC 10.2.0 in one of my c++ test-cases.
I also tested GCC 11 (GNU Fortran (GCC) 11.0.0 20201029) with the attached file
and got the same result.

After a little digging I found out that the expanded region for 'sbr' starts at
line 8 (`use foo`) and ends at line 4 (`end subroutine sbr`).

But I have no idea why `expand_location` returns line 8.

[Bug gcov-profile/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |marxin at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
Thanks for the report. Started with r9-4216-g390e529e2b98983d and yes it's
about a function ending sooner than it starts.
I'll take a look.

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|gcov-profile                |fortran
             Status|ASSIGNED                    |NEW
           Assignee|marxin at gcc dot gnu.org          |unassigned at gcc dot gnu.org
                 CC|                            |burnus at gcc dot gnu.org,
                   |                            |kargl at gcc dot gnu.org,
                   |                            |pault at gcc dot gnu.org

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
I can only confirm it's a Fortran issue.
Can please anybody from Fortran folks take a look?

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #6 from kargl at gcc dot gnu.org ---
(In reply to Martin Liška from comment #5)
> I can only confirm it's a Fortran issue.
> Can please anybody from Fortran folks take a look?

First, the caveat, the Fortran code in the attached
example is invalid so a compiler can to do whatever
it wants, including dying with an ICE.

Split the example code into a file with only the 
module (say, d.f90) and a file with only the
function foo_suite (say, e.f90).  I get

% gfcx -c --coverage d.f90
% gfcx -c --coverage e.f90

with no ICE.  It seems that the strengthen assumption in
the git commit gets confused on where function information
comes from.

Just a SWAG.  Likely, gfortran reads the newly created
'foo.mod' to get information about the subroutine sbr,
so gfortran simply records the location where it obtains
the information about sbr.  That then means the location
for sbr is set to the line 'use foo'.  Why --coverage
finds the `end subroutine` at line 4 then becomes a
mystery.

Uncompressing the module shows that no location information
is stored in a module.

% cat foo.mod | gunzip
GFORTRAN module version '15' created from d.f90
(() () () () () () () () () () () () () () () () () () () () () () () ()
() () ())
()
()
()
()
()
(2 'foo' 'foo' '' 1 ((MODULE UNKNOWN-INTENT UNKNOWN-PROC UNKNOWN UNKNOWN
0 0) () (UNKNOWN 0 0 0 0 UNKNOWN ()) 0 0 () () 0 () () () 0 0)
3 'sbr' 'foo' '' 1 ((PROCEDURE UNKNOWN-INTENT MODULE-PROC DECL UNKNOWN 0
0 SUBROUTINE IMPLICIT_PURE) () (UNKNOWN 0 0 0 0 UNKNOWN ()) 0 0 () () 0
() () () 0 0)
)
('foo' 0 2 'sbr' 0 3)

So, it seems the strengthened assumption isn't so strong.

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Patch:  https://gcc.gnu.org/pipermail/gcc-patches/2020-November/558214.html

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Tobias Burnus <burnus@gcc.gnu.org>:

https://gcc.gnu.org/g:2d59593105d3b1e38e49f96b74fd81e4e5038c7f

commit r11-4866-g2d59593105d3b1e38e49f96b74fd81e4e5038c7f
Author: Tobias Burnus <tobias@codesourcery.com>
Date:   Tue Nov 10 10:31:33 2020 +0100

    Fortran: Fix function decl's location [PR95847]

    gcc/fortran/ChangeLog:

            PR fortran/95847
            * trans-decl.c (gfc_get_symbol_decl): Do not (re)set the location
            of an external procedure.
            (build_entry_thunks, generate_coarray_init, create_main_function,
            gfc_generate_function_code): Use fndecl's location in BIND_EXPR.

    gcc/testsuite/ChangeLog:

            PR fortran/95847
            * gfortran.dg/coverage.f90: New test.

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #9 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to johannes.ziegenbalg from comment #2)
> I get the same bug with GCC 10.2.0 in one of my c++ test-cases.

Johannes: Can you fill a bugreport for the C++ test case? This PR is only about
Fortran – and the fix is for the Fortran front end, only.

Ole: Thanks for the report!

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Tobias Burnus
<burnus@gcc.gnu.org>:

https://gcc.gnu.org/g:17563c1153d36440936ea0948ac689f84b2cf666

commit r10-9017-g17563c1153d36440936ea0948ac689f84b2cf666
Author: Tobias Burnus <tobias@codesourcery.com>
Date:   Tue Nov 10 10:31:33 2020 +0100

    Fortran: Fix function decl's location [PR95847]

    gcc/fortran/ChangeLog:

            PR fortran/95847
            * trans-decl.c (gfc_get_symbol_decl): Do not (re)set the location
            of an external procedure.
            (build_entry_thunks, generate_coarray_init, create_main_function,
            gfc_generate_function_code): Use fndecl's location in BIND_EXPR.

    gcc/testsuite/ChangeLog:

            PR fortran/95847
            * gfortran.dg/coverage.f90: New test.

    (cherry picked from commit 2d59593105d3b1e38e49f96b74fd81e4e5038c7f)

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

--- Comment #11 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-9 branch has been updated by Tobias Burnus
<burnus@gcc.gnu.org>:

https://gcc.gnu.org/g:547d07637bf09d74543f688454ce53a4c5eecd47

commit r9-9041-g547d07637bf09d74543f688454ce53a4c5eecd47
Author: Tobias Burnus <tobias@codesourcery.com>
Date:   Thu Nov 12 16:39:58 2020 +0100

    Fortran: Fix function decl's location [PR95847]

    gcc/fortran/ChangeLog:

            PR fortran/95847
            * trans-decl.c (gfc_get_symbol_decl): Do not (re)set the location
            of an external procedure.
            (build_entry_thunks, generate_coarray_init, create_main_function,
            gfc_generate_function_code): Use fndecl's location in BIND_EXPR.

    gcc/testsuite/ChangeLog:

            PR fortran/95847
            * gfortran.dg/coverage.f90: New test.

    (cherry picked from commit 2d59593105d3b1e38e49f96b74fd81e4e5038c7f)

[Bug fortran/95847] [9/10/11 Regression] Internal error when processing pFUnit generated files with --coverage

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95847

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #12 from Tobias Burnus <burnus at gcc dot gnu.org> ---
FIXED on the GCC 11 (mainline), GCC 10 and GCC 9.

Thanks again for the report.

@johannes: As mentioned earlier today – please fill a --coverage bug report for
your C++ testcase, if it still occurs. (This PR only fixed the Fortran issue.)