DebugInfo: emit DW_AT_alignment for all types.

[dimitry-]

Bugzilla Link	39009
Version	unspecified
OS	All
CC	@dwblaikie

Extended Description

DWARF5 introduced DW_AT_alignment attribute.

As far as I can see this is currently mostly unused by clang/llvm. It would be helpful to have this attribute emitted for all base types as well as structure/union/class types.

[dwblaikie]

Worth measuring the impact - I'd expect to only put alignment on types that have non-standard alignment (ones where the user has specified a custom different alignment - like LLVM's habit of overaligning data structures so it can use low bits in pointers for other things).

If the impact is low enough to put it everywhere, I guess that'd be OK and save consumers from having to compute the alignment in many complex cases.

[llvmbot]

@llvm/issue-subscribers-debuginfo

Author: None (c1ed623a-df99-497b-81ca-bca6dae737a4)

| | | | --- | --- | | Bugzilla Link | [39009](https://llvm.org/bz39009) | | Version | unspecified | | OS | All | | CC | @dwblaikie |

Extended Description

DWARF5 introduced DW_AT_alignment attribute.

As far as I can see this is currently mostly unused by clang/llvm. It would be helpful to have this attribute emitted for all base types as well as structure/union/class types.

[Michael137]

The scope of when this attribute is to be emitted was narrowed in https://reviews.llvm.org/D24426

Discussions where this came up in LLDB:

• DebugInfo: No alignment info for "transitive" alignment-enforced structures. #73623
• [clang][RecordLayoutBuilder] Be stricter about inferring packed-ness in ExternalLayouts #97443

But in both cases the solution was for LLDB to be less fragile against lack of alignment information.

(FWIW, GCC doesn't do this either last time I checked)