walkero
/
harfbuzz
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[buffer] Improve shift_forward() 

"Improve" is a strong word in this case though, I understand.
This commit is contained in:
Behdad Esfahbod 2018-10-07 20:35:06 -04:00
parent f62f6e90ad
commit 14ebf8af0c
1 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
src/hb-buffer.cc
View File

@ -182,7 +182,11 @@ hb_buffer_t::shift_forward (unsigned int count)
if (idx + count > len) if (idx + count > len)
{ {
/* Under memory failure we might expose this area. At least /* Under memory failure we might expose this area. At least
* clean it up. Oh well... */ * clean it up. Oh well...
*
* Ideally, we should at least set Default_Ignorable bits on
* these, as well as consistent cluster values. But the former
* is layering violation... */
memset (info + len, 0, (idx + count - len) * sizeof (info[0])); memset (info + len, 0, (idx + count - len) * sizeof (info[0]));
} }
len += count; len += count;
@ -399,8 +403,14 @@ hb_buffer_t::move_to (unsigned int i)
unsigned int count = out_len - i; unsigned int count = out_len - i;
/* This will blow in our face if memory allocation fails later /* This will blow in our face if memory allocation fails later
* in this same lookup... */ * in this same lookup...
if (unlikely (idx < count && !shift_forward (count + 32))) return false; *
* We used to shift with extra 32 items, instead of the 0 below.
* But that would leave empty slots in the buffer in case of allocation
* failures. Setting to zero for now to avoid other problems (see
* comments in shift_forward(). This can cause O(N^2) behavior more
* severely than adding 32 empty slots can... */
if (unlikely (idx < count && !shift_forward (count + 0))) return false;
assert (idx >= count); assert (idx >= count);