Re: sparse-llvm array size computation issue

[Luc Van Oostenryck <luc.vanoostenryck@gmail.com>]

On Wed, Mar 29, 2017 at 4:41 PM, Dibyendu Majumdar
<mobile@majumdar.org.uk> wrote:
> Hi Luc,
>
>>
>> Okay thank you for the insight. It seems then that sparse-llvm is not
>> handing this correctly.
>>
>
> I looked at this again briefly today. I think that not having the size
> information on the array type poses some problems.
>
> + will instructions have access to the SYM_NODE always? It doesn't
> appear to be the case.

The SYM_NODE is present/only added when needed.
Of course, it's possible that it's somehow stripped in sparse-llvm.

> + sparse-llvm caches the symbol's type in symbol->aux. For array
> types, we would still need to do this - storing the type against
> SYM_NODE objects is probably not correct.

I don't see why it shouldn't be correct.

> So I feel that given that the size is an integral part of the array
> type then it makes sense that it should be present on the array type.

I feel some sympathy for the argument here but ...
there are other infos stored in the SYM_NODE that *can't* be
stored in the symbol underneath it. I'm thinking about the exact type,
the modifiers, for example. You will soon or later need to handle
the SYM_NODE anyway; stripping it and trying to directly use the
base type under is in general wrong.

-- Luc