[Bug other/109163] New: SARIF (and other JSON) output files are non-deterministic

[Bug other/109163] New: SARIF (and other JSON) output files are non-deterministic

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

            Bug ID: 109163
           Summary: SARIF (and other JSON) output files are
                    non-deterministic
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dmalcolm at gcc dot gnu.org
  Target Milestone: ---

gcc/json.cc's json::object uses a hash_map for tracking the key/value pairs,
and object::print iterates through them in arbitrary order, so every time we
emit json files they can potentially vary, which makes it much harder to
compare them from run to run (see e.g. PR 105959).

It would probably be much more user-friendly to use an ordered_hash_map here to
preserve insertion order and thus have deterministic output.  I don't know if
this would have a noticeable performance hit.

[Bug other/109163] SARIF (and other JSON) output files are non-deterministic

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

--- Comment #1 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
This would also help with one of the requests from a SARIF expert's review of
GCC's output:   
https://github.com/oasis-tcs/sarif-spec/issues/531#issuecomment-1181191100
which is that the "version" property should occur first in the file.

It also might make testcases easier to write.

[Bug other/109163] SARIF (and other JSON) output files are non-deterministic

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2023-03-16
           Assignee|unassigned at gcc dot gnu.org      |dmalcolm at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #2 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
I'm working on a fix for this.

[Bug other/109163] SARIF (and other JSON) output files are non-deterministic

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2023-March/6
                   |                            |14165.html
           Keywords|                            |patch
             Status|ASSIGNED                    |WAITING

--- Comment #3 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Patch posted for review:
  [PATCH] json: preserve key-insertion order [PR109163]
    https://gcc.gnu.org/pipermail/gcc-patches/2023-March/614165.html

[Bug other/109163] SARIF (and other JSON) output files are non-deterministic

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by David Malcolm <dmalcolm@gcc.gnu.org>:

https://gcc.gnu.org/g:7f1e15f743357e037d7c4f6f6000863c26f3dfc3

commit r13-6851-g7f1e15f743357e037d7c4f6f6000863c26f3dfc3
Author: David Malcolm <dmalcolm@redhat.com>
Date:   Fri Mar 24 11:38:14 2023 -0400

    json: preserve key-insertion order [PR109163]

    PR other/109163 notes that when we write out JSON files, we traverse
    the keys within each object via hash_map iteration, and thus the
    ordering is non-deterministic - it can arbitrarily vary from run to
    run and from different machines, making it harder for users to compare
    results and determine if anything has "really" changed.

    I'm running into this issue with SARIF output, but there are several
    places where we're currently emitting JSON:

      * -fsave-optimization-record emits SRCFILE.opt-record.json.gz
            "This option is experimental and the format of the data within
            the compressed JSON file is subject to change."; see
            optinfo-emit-json.{h,cc}, dumpfile.cc, etc
      * -fdiagnostics-format= with the various "sarif" and "json" options
      * -fdump-analyzer-json is a developer option in the analyzer
      * gcov has:
         "-j, --json-format: Output JSON intermediate format into
         .gcov.json.gz file"

    This patch adds an auto_vec to class json::object to preserve
    key-insertion order, and use it when writing out objects.  Potentially
    this slightly slows down JSON output, but I believe that this isn't
    normally a bottleneck, and that the benefits to the user of
    deterministic output are worth it.

    I had first attempted to use ordered_hash_map.h for this, but ran into
    impenetrable template errors, so this patch uses a simpler approach of
    just adding an auto_vec to json::object.

    Testing showed a failure of diagnostic-format-json-5.c, which was using
    a convoluted set of regexps to consume the output; I believe that this
    was brittle, and was intermittently failing for some of the random
    orderings of output.  I rewrote these regexps to work with the expected
    output order.  The other such tests seem to pass with the
    now-deterministic orderings.

    gcc/ChangeLog:
            PR other/109163
            * json.cc: Update comments to indicate that we now preserve
            insertion order of keys within objects.
            (object::print): Traverse keys in insertion order.
            (object::set): Preserve insertion order of keys.
            (selftest::test_writing_objects): Add an additional key to verify
            that we preserve insertion order.
            * json.h (object::m_keys): New field.

    gcc/testsuite/ChangeLog:
            PR other/109163
            * c-c++-common/diagnostic-format-json-1.c: Update comment.
            * c-c++-common/diagnostic-format-json-2.c: Likewise.
            * c-c++-common/diagnostic-format-json-3.c: Likewise.
            * c-c++-common/diagnostic-format-json-4.c: Likewise.
            * c-c++-common/diagnostic-format-json-5.c: Rewrite regexps.
            * c-c++-common/diagnostic-format-json-stderr-1.c: Update comment.

    Signed-off-by: David Malcolm <dmalcolm@redhat.com>

[Bug other/109163] SARIF (and other JSON) output files are non-deterministic

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109163

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

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

--- Comment #5 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Should be fixed by the above patch on trunk for GCC 13.