[PATCH] D100630: [Debug-Info][DBX] DW_TAG_rvalue_reference_type should not be generated when dwarf version is smaller than 4

[David Blaikie via Phabricator via cfe-commits]

dblaikie added a comment.

In D100630#2749625 <https://reviews.llvm.org/D100630#2749625>, @shchenz wrote:

> In D100630#2749594 <https://reviews.llvm.org/D100630#2749594>, @dblaikie wrote:
>
>> In D100630#2749550 <https://reviews.llvm.org/D100630#2749550>, @shchenz wrote:
>>
>>> In D100630#2748899 <https://reviews.llvm.org/D100630#2748899>, @dblaikie wrote:
>>>
>>>> Does this cause duplicate DW_TAG_reference_types in the output, if the input IR Has both DW_TAG_reference_type and DW_TAG_rvalue_reference_types?
>>>
>>> Yes, it will cause such issue in theory. I can not cook a case that has a DW_TAG_reference_type to a DW_TAG_rvalue_reference_types or has a DW_TAG_rvalue_reference_types to a DW_TAG_reference_type for now. Is there such usage that we both be rvalue-reference and lvalue-reference to a basic type?
>>
>> Ah, I was less thinking of a case of an rvalue reference and a normal reference in the same chain - I was thinking more about two unrelated reference types.
>>
>> Try some code like:
>>
>>   void f1(int &&x, int &y) {
>>   }
>>
>> I'm guessing without this change - you get 'x's type pointing to the rvalue_reference_type, 'y's type pointing to the reference_type, and both of those pointing to the 'int' DW_TAG_basic_type.
>>
>> But with this change I'm worried you'll get two copies of the DW_TAG_reference_type - one that replaces the rvalue reference, and then the real one. Instead you should get one copy that both 'x' and 'y' refer to, I think.
>
> Hmm, yeah, indeed, a duplicated `DW_TAG_reference_type` is generated.
>
>   0x00000069:   DW_TAG_reference_type
>                   DW_AT_type      (0x0000006e "int")
>   
>   0x0000006e:   DW_TAG_base_type
>                   DW_AT_name      ("int")
>                   DW_AT_encoding  (DW_ATE_signed)
>                   DW_AT_byte_size (0x04)
>   
>   0x00000075:   DW_TAG_reference_type
>                   DW_AT_type      (0x0000006e "int")
>
> I think rather than fixing the duplicated `DW_TAG_reference_type` in the backend, maybe it is better to fix this strict DWARF issue in the FE like the fix in the previous patch https://reviews.llvm.org/D100630?id=338755
>
> What do you think? @dblaikie

I'd still hope this issue could be addressed in the backend, but yeah - it might be non-trivial.

@aprantl @probinson etc - any thoughts on this?



================
Comment at: llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp:398-399
+
+    // Assertion for other cases.
+    assert(true && "Tag is generated not following strict DWARF!");
+  }
----------------
side note: Not sure what this assertion is for/does?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D100630/new/

https://reviews.llvm.org/D100630