[PATCH] D14275: Make createReferenceType take size and align

[David Blaikie via llvm-commits]

On Tue, Nov 3, 2015 at 8:17 AM, Adrian Prantl <aprantl at apple.com> wrote:

>
> On Nov 2, 2015, at 10:44 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>
>
> On Mon, Nov 2, 2015 at 10:24 PM, Keno Fischer <
> kfischer at college.harvard.edu> wrote:
>
>> I'm happy to have either representation, but I would like it to be
>> consistent between pointers and references. One point that speaks for not
>> having it in the debug info is that it makes the debug info dependent on
>> the size of the pointer (when fixing up the test cases I had to add quite a
>> number of explicit `target triple`s to fix the pointer size).
>>
>
> More or less agree that consistency is a good idea here (the only marginal
> counterargument involves weird stuff about how C++ references don't
> necessarily have storage, etc, but it's not terribly practical/relevant
> when you get right down to it I don't think)
>
> & I would tend towards removing the info, if it's not being used in DWARF
> emission, just to avoid confusion, but I'm open to other ideas from the
> usual suspects (including yourself)
>
>
> Since the size/alignment doesn’t take up any additional space in the IR,
> I’m leaning towards keeping them — if it allows us to write better/stricter
> debug info verifiers.
>

Fair - let's go with it then. We already have the emission+suppression for
pointers, so let's just do the same for references.


>
>
>
>> However, is it possible that different pointer types (e.g. in different
>> address spaces) could have different size? In that case it might be needed
>> in either metadata or DWARF or both.
>>
>
> Possible, though I don't think we've tried representing address-spaced
> variables as yet. Not 100% sure though. I know it's been mentioned.
>
>
> 16-bit x86 near/far pointers would be another example, but I think
> differently sized address spaces are more likely to come to LLVM :-)
>
>
>
>>
>> On Tue, Nov 3, 2015 at 1:13 AM, David Blaikie <dblaikie at gmail.com> wrote:
>>
>>>
>>>
>>> On Mon, Nov 2, 2015 at 10:04 PM, Keno Fischer via llvm-commits <
>>> llvm-commits at lists.llvm.org> wrote:
>>>
>>>> loladiro added a comment.
>>>>
>>>> Ah, interesting we skip emitting the size of a pointer in the backend.
>>>> Will add the same for reference types.
>>>>
>>>
>>> Oh, interesting...
>>>
>>> Suspicious why we add it to the metadata then... that seems strange.
>>> (admittedly it comes in handy for the verification you're adding - but I'm
>>> not sure how much we want to add to the metadata to facilitate that sort of
>>> thing & make the debug info less clear in its purpose by doing so)
>>>
>>>
>>>>
>>>>
>>>> Repository:
>>>>   rL LLVM
>>>>
>>>> http://reviews.llvm.org/D14275
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>>
>>>
>>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151103/20b48f91/attachment.html>