Re: [cocci] spatch 1.1.1 segmentation fault report

[Denis Efremov <efremov@linux.com>]

On 2/7/22 15:56, Julia Lawall wrote:
> 
> 
> On Mon, 7 Feb 2022, Denis Efremov wrote:
> 
>>
>>
>> On 2/7/22 12:54, Julia Lawall wrote:
>>>
>>>
>>> On Mon, 7 Feb 2022, Denis Efremov wrote:
>>>
>>>>
>>>>
>>>> On 2/7/22 09:56, Julia Lawall wrote:
>>>>>> I checked that the problem persist on latest commit
>>>>>> b2a4b9b77157 parsing_c: print correct attributes in pretty_print_c.ml
>>>>>>
>>>>>> and bisected the crash to commit
>>>>>> 7b79a2415a8c6e8b1c39a8183bf732fdc6d00453 is the first bad commit
>>>>>> commit 7b79a2415a8c6e8b1c39a8183bf732fdc6d00453
>>>>>> Author: Corentin De Souza <corentin.de-souza@inria.fr>
>>>>>> Date:   Tue May 25 19:21:51 2021 +0200
>>>>>>
>>>>>>     parsing_c: detect macro iterators from their bodies
>>>>>>
>>>>>>     This uses the same strategy as macro statements (see
>>>>>>     https://gitlab.inria.fr/coccinelle/coccinelle/-/merge_requests/92).
>>>>>>     The goal is to do fix https://gitlab.inria.fr/coccinelle/coccinelle/-/issues/24
>>>>>>
>>>>>>     The next commit will finally fix the assign_expr rule in the parser, removing
>>>>>>     the conflicts it introduced while not breaking test
>>>>>>     assign_deref_funcall_in_macro_it.
>>>>>>
>>>>>>  parsing_c/cpp_token_c.ml | 39 ++++++++++++++++++++++++++++++---------
>>>>>>  parsing_c/parser_c.mly   |  3 ++-
>>>>>>  2 files changed, 32 insertions(+), 10 deletions(-)
>>>>>
>>>>> Thanks for all of the investigation.  I'll take a look at this commit, but
>>>>> from the subject it seems hard to imagine how it would affect python...
>>>>>
>>>>> Could you take a look at the core to see if indicates what function caused
>>>>> the problem?
>>>>
>>>> Core was generated by `spatch segfault.cocci empty_dir'.
>>>> Program terminated with signal SIGSEGV, Segmentation fault.
>>>> #0  0x0000000000aa8b7e in pydecref (v=<optimized out>) at pyml_stubs.c:247
>>>> 247     pyml_stubs.c: Directory not empty.
>>>> (gdb) bt
>>>
>>> Does your pyml come from the bundle in Coccinelle or from opam?
>>>
>>> In my spatch 1.1.1 code, line 247 is just a variable declaration.
>>
>> Installed with opam pyml.20211015
>> Line 247:
>> 244:static void pydecref( value v )
>> 245:{
>> 246:    if (getcustom(v)) {
>> 247:        Py_DECREF((PyObject *)getcustom(v));
>> 248:    }
>> 249:}
>>
>> I uninstalled opam and rebuilded spatch with the bundled pyml on
>> latest commit: b2a4b9b77157 parsing_c: print correct attributes in pretty_print_c.ml
>>
>> Result is the same.
>>
>> #0  0x0000000000aa810e in pydecref ()
>> #1  0x0000000000ac512e in caml_empty_minor_heap () at minor_gc.c:409
>> #2  0x0000000000ac554c in caml_gc_dispatch () at minor_gc.c:475
>> #3  0x0000000000ac563e in caml_check_urgent_gc (extra_root=<optimized out>, extra_root@entry=1) at minor_gc.c:572
>> #4  0x0000000000ac56d3 in caml_alloc_small_dispatch (wosize=2, flags=flags@entry=1, nallocs=nallocs@entry=1,
>>     encoded_alloc_lens=encoded_alloc_lens@entry=0x0) at minor_gc.c:516
>> #5  0x0000000000ac6c71 in caml_alloc_small (wosize=wosize@entry=2, tag=tag@entry=0) at alloc.c:68
>> #6  0x0000000000accb2f in caml_ml_out_channels_list (unit=<optimized out>) at io.c:512
>> #7  0x0000000000a46645 in camlStdlib__flush_all_231 () at stdlib.ml:349
>> #8  0x00000000005ffeaf in camlStd_exit__entry () at stdlib.ml:556
>> #9  0x00000000005f5199 in caml_program ()
>> #10 0x0000000000adfca0 in caml_start_program ()
>> #11 0x0000000000ae0024 in caml_startup_common (argv=0x7ffcaf256848, pooling=<optimized out>, pooling@entry=0) at startup_nat.c:160
>> #12 0x0000000000ae009b in caml_startup_exn (argv=<optimized out>) at startup_nat.c:165
>> #13 caml_startup (argv=<optimized out>) at startup_nat.c:170
>> #14 caml_main (argv=<optimized out>) at startup_nat.c:177
>> #15 0x00000000005f3ddc in main (argc=<optimized out>, argv=<optimized out>) at main.c:44
> 
> Thierry has found the problem.  I will push the solution shortly to
> github.  We can make a new release if you need that.
>

Thank you! I don't need a new release. But if there is a way to slightly change the pattern
to avoid the crash this will be useful to me.

Regards,
Denis