First of all:

	s.ostanevich@s-ostanevich2:~/workspaces/t.sergos % uname -a                                                 
	Darwin s-ostanevich2 21.6.0 Darwin Kernel Version 21.6.0: Mon Aug 22 20:17:10 PDT 2022; root:xnu-8020.140.49~2/RELEASE_X86_64 x86_64

Then I just built Tarantool with ninja:

	Tarantool 2.11.0-entrypoint-494-gdc147ec91
	Target: Darwin-x86_64-debug

Then I brought the test as a patch from the mail and run ninja test.

The output says the test passes, while sources contains no patch. Am I missing something?

Sergos

> On 20 Sep 2022, at 12:58, Maxim Kokryashkin <m.kokryashkin@tarantool.org> wrote:
> 
> Hi!
> Are you sure you are doing everything right?
> I’ve tested the test case both on M1 and x86 platforms and it works perfectly fine.
> --
> Best regards,
> Maxim Kokryashkin
>  
>  
> Вторник, 20 сентября 2022, 11:53 +03:00 от sergos <sergos@tarantool.org>:
>  
> Hi!
>  
> Thanks for the patch!
>  
> Unfortunately, the test doesn’t fail for me, using tarantool test:
>  
> /Users/s.ostanevich/workspaces/t.sergos/third_party/luajit/test/tarantool-tests/lj-695-ffi-vararg-call.test.lua ...................... ok
> /Users/s.ostanevich/workspaces/t.sergos/third_party/luajit/test/tarantool-tests/lj-584-bad-renames-for-sunk-values.test.lua .......... ok
> /Users/s.ostanevich/workspaces/t.sergos/third_party/luajit/test/tarantool-tests/lj-408-tonumber-cdata-record.test.lua ................ ok
> /Users/s.ostanevich/workspaces/t.sergos/third_party/luajit/test/tarantool-tests/gh-6189-cur_L.test.lua ............................... ok
> /Users/s.ostanevich/workspaces/t.sergos/third_party/luajit/test/tarantool-tests/lj-418-assert-any-type.test.lua ...................... ok
>  
> The sources at the build time are:
>  
> void LJ_FASTCALL lj_crecord_tonumber(jit_State *J, RecordFFData *rd)
> {
>   CTState *cts = ctype_ctsG(J2G(J));
>   CType *d, *ct = lj_ctype_rawref(cts, cdataV(&rd->argv[0])->ctypeid);
>   if (ctype_isenum(ct->info)) ct = ctype_child(cts, ct);
>   if (ctype_isnum(ct->info) || ctype_iscomplex(ct->info)) {
>     if (ctype_isinteger_or_bool(ct->info) && ct->size <= 4 &&
>         !(ct->size == 4 && (ct->info & CTF_UNSIGNED)))
>       d = ctype_get(cts, CTID_INT32);
>     else
>       d = ctype_get(cts, CTID_DOUBLE);
>     J->base[0] = crec_ct_tv(J, d, 0, J->base[0], &rd->argv[0]);
>   } else {
>     J->base[0] = TREF_NIL;
>   }
> }
>  
> Means, test passess even without the patch.
>  
> Sergos
>  
>> On 19 Sep 2022, at 10:50, Maxim Kokryashkin <m.kokryashkin@tarantool.org <x-msg://e.mail.ru/compose/?mailto=mailto%3am.kokryashkin@tarantool.org>> wrote:
>>  
>> Hi, Sergey!
>> Thanks for the patch!
>> LGTM, except for a single nit below:
>> 
>> When `tonumber()` is recorded (as a part of a trace) for cdata argument
>> can't be converted to number the `nil` value is recorded as the yielded
>> result. But without special check on trace for cdata type this nil will
>> be returned for another type of cdata that can be converted.
>> The first sentence lacks commas and is completely unreadable.
>> I suggest the following fix:
>> | When `tonumber()` is recorded (as a part of a trace) for a cdata argument that can't be converted to number, the `nil` value is | recorded as the yielded result.
>> 
>> This patch adds the corresponding check for recoding of failed cdata
>> conversions.
>> Typo: s/recoding/recording
>>  
>> <snipped>
>> --
>> Best regards,
>> Maxim Kokryashkin
>>  
> 
>