Re: [Emacs-diffs] master d02c2f7 4/4: Speed up maybe_gc when GC is inhibited

[Pip Cet]

On Sun, Jul 21, 2019 at 6:24 PM Paul Eggert <address@hidden> wrote:
>  static void
> -allow_garbage_collection (void)
> +allow_garbage_collection (void *ptr)
>  {
> +  object_ct *p = ptr;
> +  consing_until_gc = *p;
>    garbage_collection_inhibited--;
>  }
>
> @@ -5513,9 +5516,10 @@ ptrdiff_t
>  inhibit_garbage_collection (void)
>  {
>    ptrdiff_t count = SPECPDL_INDEX ();
> -
> -  record_unwind_protect_void (allow_garbage_collection);
> +  object_ct consing = consing_until_gc;
> +  record_unwind_protect_ptr (allow_garbage_collection, &consing);
>    garbage_collection_inhibited++;
> +  consing_until_gc = OBJECT_CT_MAX;
>    return count;
>  }

This looks unsafe to me. `consing' goes out of scope when
inhibit_garbage_collection returns, at which point the stack space for
`consing' will be reused for something else.
`allow_garbage_collection' is called later, and accesses what's
essentially random stack data.

Or am I missing something?