Re: [patch] fix wrong gimple code with ptr to forward vla

[Mark Mitchell <mark at codesourcery dot com>]

Olivier Hainque wrote:

>  I don't think so. As Richard clarified later on, the idea was to check
>  whether the pointed-to (not pointer) type has a name.

OK.

> > In any case, I can't see any intuitive reason for making anonymity
> > have semantics in the middle end.  (It does have semantics in some
> > languages, of course, so in the front end I would not have the same
> > objection.)
>
>  I understand and share your concern.

OK, so we all agree we don't like the approach that's been proposed. 
The question now is one of pragmatism: what can we do in the relatively 
short term to solve the problem, without tying our hands for the future?

>  It did not seem unreasonable to consider variants of such assumptions
>  for a pointed-to type depending on whether it is anonymous or not (e.g.
>  there is no possible forward declaration involved), but I agree
>  exploiting such variations sure makes maintainance harder. Besides,
>  one's understanding of what can implicitly be relied upon might simply
>  be mistaken.

I think I understand why the heuristic works, in practice.  It's based 
on Richard's explanation, which assumes an Ada-like language in which 
types are elaborated at a well-defined point.  If you can't name the 
type, then its point of elaboration must be the point at which you first 
 encounter it.

I guess that you can either make the assumption that there will be 
TYPE_DECLs for all types at their point of elaboration (which then 
allows the gimplifier to transform the size calculations into temporary 
variables, etc.), or do as Joseph suggests and make the front ends 
themselves perform the transformations.

I like Joseph's choice better.  In particular, Richard's approach 
requires the C front end to make more TYPE_DECLs, which drives up memory 
usage.  And, Joseph's choice also makes the middle-end semantics more 
straightforward, which I think is important in helping as many people as 
possible work on the common code.  That said, I recognize that Joseph's 
change is not a simple one.

Have you looked into making Richard's proposed change to the C front 
end, as an alternate fix to Jakub's PR 21536?  If you made the C front 
end explictly declare the type, then the type-elaboration approach would 
work as expected, and Jakub's patch could be removed.  That would be a 
logically coherent approach, and we could always switch to Joseph's 
approach in the fullness of time.

In other words, I think I agree that Jakub's patch is incorrect: 
gimplification of pointer types should not imply gimplification of 
pointed-to types.  But, I'd much rather fix it in a coherent way than 
with a dependence on anonymity, which is a clever, but logically rather 
indefensible, approach.

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713