Re: [PATCH] gdb/testsuite: Fix typos in infcall-nested-structs.c

[Tom de Vries <tdevries@suse.de>]

On 10-10-2019 22:26, Tom de Vries wrote:
> On 10-10-2019 20:30, Tom de Vries wrote:
>> On 10-10-2019 19:24, Andreas Arnez wrote:
>>> On Thu, Oct 10 2019, Tom de Vries wrote:
>>>
>>>> I see these new failures on x86_64-linux:
>>>> ...
>>>> FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-tc-tf: p/d
>>>> check_arg_struct_02_01 (ref_val_struct_02_01)
>>>> FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-ts-tf: p/d
>>>> check_arg_struct_02_01 (ref_val_struct_02_01)
>>>> FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-ti-tf: p/d
>>>> check_arg_struct_02_01 (ref_val_struct_02_01)
>>>> ...
>>>
>>> Maybe the test case caught a real bug then, right?  Or do you see a
>>> problem with the test case?
>>
>> I think it's a real bug.
>>
>> I've minimized the types-ti-tf FAIL to:
>> ...
>> $ cat test.c
>> typedef int ti;
>> typedef float tf;
>> struct struct_02_01
>> {
>>   struct { } es1;
>>   struct {
>>     struct {
>>       ti a;
>>       tf b;
>>     } s1;
>>   } s2;
>> };
>>
>> struct struct_02_01 ref_val_struct_02_01 = {
>>   {},
>>   {
>>     {
>>       'a',
>>       'b'
>>     }
>>   }
>> };
>>
>> int cmp_struct_02_01 (struct struct_02_01 a, struct struct_02_01 b)
>> { return a.s2.s1.a == b.s2.s1.a && a.s2.s1.b == b.s2.s1.b; }
>>
>> int
>> check_arg_struct_02_01 (struct struct_02_01 arg) {
>>   return cmp_struct_02_01 (arg, ref_val_struct_02_01);
>> }
>>
>> int
>> main (void)
>> {
>>   return check_arg_struct_02_01 (ref_val_struct_02_01);
>> }
>> $ g++ test.c -g
>> $ ./a.out; echo $?
>> 1
>> $ gdb a.out -batch -ex start -ex "p check_arg_struct_02_01
>> (ref_val_struct_02_01)"
>> Temporary breakpoint 1 at 0x400563: file test.c, line 35.
>>
>> Temporary breakpoint 1, main () at test.c:35
>> 35        return check_arg_struct_02_01 (ref_val_struct_02_01);
>> $1 = 0
>> ...
>>
> 
> The discrepancy is that the code generated by gcc passes the struct in
> registers %rdi and %xmm0, but amd64_push_arguments classifies the struct as:
> ...
> (gdb) p theclass
> $57 = {AMD64_INTEGER, AMD64_INTEGER}
> ...
> and therefore passes it in %rdi and %rsi.
> 

I've simplified the test-case a bit further, and filed as:
https://sourceware.org/bugzilla/show_bug.cgi?id=25096.

Thanks,
- Tom