[HarfBuzz] harfbuzz-ng: Branch 'master'

[Behdad Esfahbod]

On 09/06/2012 10:39 PM, Harshula wrote:
> On Thu, 2012-09-06 at 15:17 -0400, Behdad Esfahbod wrote:
>> On 09/06/2012 02:47 PM, Harshula wrote:
>>>>>    ALLOCATE_ARRAY (const HB_UChar16, item.string, chars_len);
>>>>> +  ALLOCATE_ARRAY (unsigned short, item.log_clusters, chars_len + 2);
>>> Why is the length "chars_len + 2"?
>>
>> 1. I *think* old harfbuzz also adds a final entry to that array,
>> 2. It's widely reported (from Chromium people for example) that some of the
>> old HarfBuzz backends (Tibetan IIRC) make an off-by-one access past that array.
>>
>> In short: to be on the safe side.
> 
> OK. Also when you calculate num_glyphs that will fit in the scratch
> space:
> 
>   ALLOCATE_ARRAY (const HB_UChar16, item.string, chars_len);
>   ALLOCATE_ARRAY (unsigned short, item.log_clusters, chars_len + 2);
> 
>  unsigned int num_glyphs = scratch_size  / (sizeof (HB_Glyph) +
>                                             sizeof (HB_GlyphAttributes) +
>                                             sizeof (HB_Fixed) +
>                                             sizeof (HB_FixedPoint) +
>                                             sizeof (uint32_t));
> 
>   ALLOCATE_ARRAY (HB_Glyph, item.glyphs, num_glyphs);
>   ALLOCATE_ARRAY (HB_GlyphAttributes, item.attributes, num_glyphs);
>   ALLOCATE_ARRAY (HB_Fixed, item.advances, num_glyphs);
>   ALLOCATE_ARRAY (HB_FixedPoint, item.offsets, num_glyphs);
>   ALLOCATE_ARRAY (uint32_t, vis_clusters, num_glyphs);
> 
> shouldn't you first subtract the space consumed by the already allocated
> arrays for item.string and item.log_clusters? For example:

No.  The ALLOCATE_ARRAY does that already.

b