Re: mes call encoding bug with struct return

[Jan Nieuwenhuizen]

Danny Milosavljevic writes:

Hi!

> when encoding a function call, mescc assumes that a huge struct fits into a
> register directly.  That is not actually possible.

Eh, struct by value...only if it's size pointer (int) or smaller... :-)

> Example:
>
> typedef struct
> {
>   long quot;
>   long rem;
> } ldiv_t;
>
> ldiv_t __mesabi_ldiv(long a, long b) {
>   ldiv_t result = {2,3};
>   return result;
> }
>
> long
> __aeabi_idiv (long a, long b)
> {
>   ldiv_t result = __mesabi_ldiv(a, b); // actual result: Call is
> encoded as if RESULT can fit into r0, which is not possible.
>   return result.quot; // actual result: not 2
> }
>
> That means at runtime we will get an erroneous result.

Yes.

> On the other hand, GCC automatically converts that into:
>
>   void __mesabi_ldiv(long a, long b, ldiv_t* result);
>
> long
> __aeabi_idiv (long a, long b)
> {
>   ldiv_t result;
>   __mesabi_ldiv(a, b, &result);
>   return result.quot;
> }

Hehe, I wrote a bug report with a patch for something very similar in
g++ once.

> Possible fixes in mescc:
>
> * Fail with an error message when someone tries to use that, or
> * Implement the same way as GCC does

(there's also the 0-option: do nothing and silently generate broken
code)

At the least, I'd like to see a test for this that fails; failing
with an error is an acceptable first improvement for now.  WDYT?

Thanks,
Janneke

-- 
Jan Nieuwenhuizen <janneke@gnu.org> | GNU LilyPond http://lilypond.org
Freelance IT http://JoyofSource.com | Avatar® http://AvatarAcademy.com