Re: sparse-llvm array size computation issue

[Dibyendu Majumdar <mobile@majumdar.org.uk>]

On 29 March 2017 at 19:12, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> On 29 March 2017 at 17:41, Linus Torvalds <torvalds@linux-foundation.org> wrote:
>> On Wed, Mar 29, 2017 at 9:21 AM, Dibyendu Majumdar
>> <mobile@majumdar.org.uk> wrote:
>>>
>>> I am trying out an approach. If a SYM_NODE has a base type of SYM_NODE
>>> then which of the nodes should be used as the source for information
>>> you mention?
>>
>> Does that actually happen? It shouldn't. A symbol node contains the C
>> name of the symbol, but you should never have a SYM_NODE that points
>> to another SYM_NODE, it always points to some actual type (ie ptr,
>> whatever).
>>
>> So the rule should be that the node can have specific information
>> about that particular named symbol (so: name, array size, modifiers,
>> address space, initializer etc), and then the node->ctype.base_type
>> should point to a non-NODE symbol describing the base type.
>>
>
> Okay thank you - that's good to know. It wasn't obvious to me looking
> at the code, and I thought I had seen an example where a node
> contained another node ... but I cannot find this now, so I may have
> been mistaken.
>

I have implemented a solution in my repository (link below). Basically
I split the symbol_type() function into three:

get_symnode_type() - when we know the symbol is a SYM_NODE
get_symnode_or_basetype() - when we do not know whether a symbol is
SYM_NODE or not

The original symbol_type() is renamed to type_to_llvmtype() and it:
a) takes extra argument that if not NULL is a SYM_NODE
b) the original symbol parameter can never be a SYM_NODE.

My tests pass. I do not know that I have been as conservative I could
- i.e. where the symbol is known to be a SYM_NODE or a base type - it
would be good to be as specific as possible.

Here is the link to my version:

https://github.com/dibyendumajumdar/dmr_c/blob/master/llvm-backend/sparse-llvm.c

Thanks and Regards
Dibyendu