Re: MPS: Crash when switching to buffer

[Gerd Möllmann]

Pip Cet <pipcet@protonmail.com> writes:

> Thanks, that helps clarify things. So would something like the
> following work? lface_id_to_name is never freed, it seems...

Yes, that works 👍

> Ihor, I'm not sure whether the bug is fully reproducible. If it is, could you 
> please try this?
>
> diff --git a/src/xfaces.c b/src/xfaces.c
> index 2bdd2f660fd..ee26a260ed4 100644
> --- a/src/xfaces.c
> +++ b/src/xfaces.c
> @@ -2970,9 +2970,15 @@ DEFUN ("internal-make-lisp-face", 
> Finternal_make_lisp_face,
>        The mapping from Lisp face to Lisp face id is given by the
>        property `face' of the Lisp face name.  */
>        if (next_lface_id == lface_id_to_name_size)
> +#ifdef HAVE_MPS
> +     lface_id_to_name =
> +       igc_xpalloc_ambig (lface_id_to_name, &lface_id_to_name_size, 1, 
> MAX_FACE_ID,
> +                          sizeof *lface_id_to_name);
> +#else
>       lface_id_to_name =
>         xpalloc (lface_id_to_name, &lface_id_to_name_size, 1, MAX_FACE_ID,
>                  sizeof *lface_id_to_name);
> +#endif
>  
>        Lisp_Object face_id = make_fixnum (next_lface_id);
>        lface_id_to_name[next_lface_id] = face;
> @@ -7326,7 +7332,11 @@ init_xfaces (void)
>       {
>         /* Allocate the lface_id_to_name[] array.  */
>         lface_id_to_name_size = next_lface_id = nfaces;
> +#ifdef HAVE_MPS
> +       lface_id_to_name = igc_xzalloc_ambig (next_lface_id * sizeof 
> *lface_id_to_name);
> +#else
>         lface_id_to_name = xnmalloc (next_lface_id, sizeof *lface_id_to_name);
> +#endif
>  
>         /* Store the faces.  */
>         struct Lisp_Hash_Table* table = XHASH_TABLE 
> (Vface_new_frame_defaults);

Alternatively, we could also use

  Lisp_Object *igc_alloc_lisp_obj_vec (size_t n)

to allocate the vector. That returns an object of type IGC_OBJ_OBJ_VEC
in MPS that eventually gets collected. We would then have to use

  void igc_root_create_exact_ptr (void *var_addr)

for the variable holding the reference to the object, where the
staticpros are done.

Just wanted to mention that this also exists, and has the advantage of
fewer ambiguous references.