Re: [Emacs-diffs] trunk r117464: Shrink Lisp_Sub_Char_Table by preferring C integers to Lisp_Objects.

[Dmitry Antipov]

On 07/03/2014 08:49 PM, Stefan Monnier wrote:

> Could you try to use something like the patch below to eliminate this
> alignment assumption?
>
> === modified file 'src/alloc.c'
> --- src/alloc.c 2014-07-02 03:26:19 +0000
> +++ src/alloc.c 2014-07-03 16:47:25 +0000
> @@ -5962,13 +5962,18 @@
>   {
>     int size = ptr->header.size & PSEUDOVECTOR_SIZE_MASK;
>     /* Consult the Lisp_Sub_Char_Table layout before changing this.  */
> -  int i, idx = (pvectype == PVEC_SUB_CHAR_TABLE ? SUB_CHAR_TABLE_OFFSET : 0);
> +  int i;
> +  Lisp_Object *contents
> +    = pvectype == PVEC_SUB_CHAR_TABLE
> +    ? (size -= SUB_CHAR_TABLE_OFFSET,
> +       ((struct Lisp_Sub_Char_Table *)ptr)->contents)
> +    : ((struct Lisp_Char_Table *)ptr)->contents;

This is obviously wrong because Lisp_Object slots of Lisp_Char_Table
between 'header' and 'contents' are never marked. This code should
looks like:

  Lisp_Object *contents
    = pvectype == PVEC_SUB_CHAR_TABLE
    ? (size -= SUB_CHAR_TABLE_OFFSET,
       ((struct Lisp_Sub_Char_Table *)ptr)->contents)
    : ptr->contents;


And, IIUC, 'size -= SUB_CHAR_TABLE_OFFSET' still assumes that
SUB_CHAR_TABLE_OFFSET is a multiple of sizeof (Lisp_Object)
(otherwise subtracting it from size, which is a number of
Lisp_Object slots, makes no sense).

Dmitry