[PATCH] D27551: [DebugInfo] Add regression test for __fp16, float, and double constants.
David Gross via llvm-commits
llvm-commits at lists.llvm.org
Mon Dec 12 16:11:51 PST 2016
Per the Details:
Prior to the change "DebugInfo: New metadata representation for global
variables." (D20147 <https://reviews.llvm.org/D20147>, D20415
<https://reviews.llvm.org/D20415>), clang emitted debug metadata for static
const values of floating-point types, but llvm did not translate this
debug metadata into DWARF. After that change, clang did not emit debug
metadata for static const values of floating-point types, but llvm is
capable of translating it into DWARF if it is emitted.
My (perhaps overly cautious) concern is that given what has occurred in the
past (where llvm was not emitting bytes for floating-point types), it's
possible that there might be a change in the future which would cause the
emission of the bytes to become type-sensitive again, and I'd like to be
sure we do continue to emit the bytes for floating-point types.
On Mon, Dec 12, 2016 at 3:47 PM, David Blaikie <dblaikie at gmail.com> wrote:
> On Mon, Dec 12, 2016 at 3:24 PM David Gross via Phabricator <
> reviews at reviews.llvm.org> wrote:
>> dgross added a comment.
>> In https://reviews.llvm.org/D27551#620373, @dblaikie wrote:
>> > Does this need the test coverage if it's just "hey, LLVM can splat
>> these bytes out into DWARF even if we (independently) say the type of those
>> bytes is a float"? The code that emits the bytes is presumably independent
>> of the code that emits the type description, so testing the combination
>> seems unnecessary to me.
>> Can you explain this further? What parts of this test case are you
>> suggesting may be unnecessary?
> All of it. (partly under the argument that no code changes were required
> to provide this functionality, but to be more complete:
> * We already have test coverage for emitting the correct DW_AT_type value
> * We already have test coverage that the arbitrary bytes in the constant
> value are emitted into the right place in the DW_AT_const_value or whatnot
> So what's the particularly interesting thing about testing the combination
> of those two (so far as I can see) orthogonal features?)
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits