Prevent crash in Lua bindings.

Prevent crash in Lua bindings.

SkPaint::getTypeface() can return NULL; paint:getTypeface() would
attempt to refcount that value before storing it on the Lua stack
and crash. This is a minimal workaround that fixes the crash.

R=reed@google.com

Committed: http://code.google.com/p/skia/source/detail?r=12706

[tomhudson, 2013-12-17 14:10:26]

Mike, you wrote the guts of the bindings code - PTAL?

Ravi, I need a fix like this on the bots in order to run my next round of
scripts; I assume that once it lands the bots will auto-update to head?

[rmistry, 2013-12-17 14:12:27]

On 2013/12/17 14:10:26, tomhudson wrote:
> Mike, you wrote the guts of the bindings code - PTAL?
> 
> Ravi, I need a fix like this on the bots in order to run my next round of
> scripts; I assume that once it lands the bots will auto-update to head?

Yes the bots auto-update to head, make clean and make tools at every run.

[robertphillips, 2013-12-17 14:16:59]

lgtm - Mike is OOO.

[commit-bot: I haz the power, 2013-12-17 14:19:47]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/tomhudson@chromium.org/109793010/1

[commit-bot: I haz the power, 2013-12-17 14:28:21]

Change committed as 12706

[reed1, 2013-12-17 16:31:50]

This CL also added Annotations, right? Did you not need to peer into the
annotation, to see its key or value?

https://codereview.chromium.org/109793010/diff/20001/src/utils/SkLua.cpp
File src/utils/SkLua.cpp (right):

https://codereview.chromium.org/109793010/diff/20001/src/utils/SkLua.cpp#newc...
src/utils/SkLua.cpp:990: SkSafeUnref(get_ref<SkTypeface>(L, 1));
I believe you, but this just seems wrong. If we had no typeface, we should have
pushed a NULL up to lua, instead of a lua object wrapping null.

[tomhudson, 2013-12-17 16:36:58]

On 2013/12/17 16:31:50, reed1 wrote:
> This CL also added Annotations, right? Did you not need to peer into the
> annotation, to see its key or value?

Patch-set 1 was (or at least should have been the one) landed.

Patch-set 2 was me accidentally re-triggering gcl upload in my shell. I'm trying
to measure the size of paints; if I'm going to do it exactly, I'd need to know
how big SkAnnotations are, which would require exposing more from their public
interface, but in practice we only see them when we're debugging, right? So
they're not performance sensitive and I can assume 1B for the flag that always
gets written.

> https://codereview.chromium.org/109793010/diff/20001/src/utils/SkLua.cpp
> File src/utils/SkLua.cpp (right):
> 
>
https://codereview.chromium.org/109793010/diff/20001/src/utils/SkLua.cpp#newc...
> src/utils/SkLua.cpp:990: SkSafeUnref(get_ref<SkTypeface>(L, 1));
> I believe you, but this just seems wrong. If we had no typeface, we should
have
> pushed a NULL up to lua, instead of a lua object wrapping null.

All I know is that every time I called getTypeface() on a paint with none set,
we returned NULL. As I said in the text this is a minimal workaround; I'm not
surprised if there's a better way to do it.