Re: [Tarantool-patches] [PATCH luajit] Fix predict_next() in parser (for real now).

[Sergey Bronnikov via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

[-- Attachment #1: Type: text/plain, Size: 1145 bytes --]


On 10.01.2025 14:10, Sergey Kaplun wrote:
> Hi, Sergey!
> Thanks for the review.
>
> On 10.01.25, Sergey Bronnikov wrote:
>> Hi, Sergey,
>>
>> thanks for the patch!
>>
>> test is passed with reverted patch.
>>
>> With original reproducer luajit segfaults.
>>
>> CMake options:  cmake -S . -B build -DCMAKE_BUILD_TYPE=Debug
>> -DLUA_USE_APICHECK=ON -DLUA_USE_ASSERT=ON
> You must run this test with Valgrind enabled to see the failure as it is
> mentioned in the commit.


cmake arguments required for reproducing a bug by proposed test:

  cmake -S . -B build -DCMAKE_BUILD_TYPE=Debug -DLUA_USE_APICHECK=ON 
-DLUA_USE_ASSERT=ON -DLUAJIT_USE_VALGRIND=ON -DLUAJIT_ENABLE_GC64=ON 
-DLUAJIT_USE_SYSMALLOC=ON


> The original reproducer is clumsy and unstable
> -- any slight change of the parser may break it. This reproducer is much
> simpler, robust, and more readable.
>
>>
>> Sergey
>>
>> On 09.01.2025 18:01, Sergey Kaplun wrote:
> <snipped>
>
>>> +-- The parsing of the `for` iterator uses the incorrect check for
>>> +-- `fs->bclim`, which allows the usage of an uninitialized value,
>>> +-- so the test fails under Valgrind.
> <snipped>
>

[-- Attachment #2: Type: text/html, Size: 2222 bytes --]