620164 - nsTheoraState::MaxKeyframeOffset doesn't need to use MulOverflow

Status: RESOLVED FIXED

(Core :: Audio/Video, defect)

Description

[timeless]

# media/libtheora/include/theora/theora.h
    * line 214 -- ogg_uint32_t fps_denominator; /**< frame rate denominator **/ 
# media/libtheora/include/theora/codec.h
    * line 241 -- ogg_uint32_t fps_denominator; 

264 nsTheoraState::MaxKeyframeOffset()
265 {
271   PRInt64 frameDuration;
272   PRInt64 keyframeDiff;
273 
274   PRInt64 shift = mInfo.keyframe_granule_shift;
275 
276   // Max number of frames keyframe could possibly be offset.
277   keyframeDiff = (1 << shift) - 1;
278 
279   // Length of frame in ms.
280   PRInt64 d = 0; // d will be 0 if multiplication overflows.

as fps_denominator is 32bit, there's no need to use MulOverflow:
281   MulOverflow(1000, mInfo.fps_denominator, d);
282   frameDuration = d / mInfo.fps_numerator;
283 
284   // Total time in ms keyframe can be offset from any given frame.
285   return frameDuration * keyframeDiff;
286 }

Comment 1

[Abhishek Bhatnagar]

I'd like to take on this bug if no one else was eyeing it.

Comment 2

[Chris Pearce [:cpearce (GMT+13)]]

Thanks!

Comment 3

[Abhishek Bhatnagar]

This bug has been fixed in the mozilla hg repository by a commit that was unknown to me.

In my opinion it can be closed, however the patch that I had produced for it is available here:

https://github.com/abhishekbh/mozilla-central/commit/5107348bd0fd66a96a9db4587bdc6c05d25d2926

I'm attaching it to the bug just in case it is needed in the future.

Comment 4

[Matthew Gregan [:kinetik]]

Bug 601535 converted a bunch of code, including this, from the hand-rolled *Overflow functions to CheckedInt.  So the code is different, but it looks like it still uses CheckedInt in a place where it's unnecessary (per timeless's analysis in comment 0).

Comment 5

[Abhishek Bhatnagar]

Attachment: Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset

Comment 6

[Abhishek Bhatnagar]

Ah indeed Matthew.

I've updated a patch without CheckedInt.

Comment 7

[Chris Pearce [:cpearce (GMT+13)]]

Comment on attachment 613489 [details] [diff] [review]
Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset

Review of attachment 613489 [details] [diff] [review]:
-----------------------------------------------------------------

::: content/media/ogg/nsOggCodecState.cpp
@@ +355,5 @@
>    PRInt64 keyframeDiff = (1 << mInfo.keyframe_granule_shift) - 1;
>  
>    // Length of frame in usecs.
> +  PRInt64 d = 0;
> +  d = PRInt64(mInfo.fps_denominator) * USECS_PER_S;

Do you need the cast to silence a warning? You don't on Windows at least...

We don't really need the temporary variable "d" anymore, can you instead eliminate "d" and make that:

  frameDuration = (mInfo.fps_denominator * USECS_PER_S) / mInfo.fps_numerator;

That makes the code a little simpler. Thanks!

Comment 8

[Abhishek Bhatnagar]

Attachment: 620164 - Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset; r=reviewers

Comment 9

[Chris Pearce [:cpearce (GMT+13)]]

Comment on attachment 613493 [details] [diff] [review]
620164 - Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset; r=reviewers

Review of attachment 613493 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks!

Comment 10

[Abhishek Bhatnagar]

Attachment: 620164 - Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset

Ah sure.

I thought it was considered better to break up long lines of code, but I suppose extra variable don't help anyone.

Comment 11

[Chris Pearce [:cpearce (GMT+13)]]

Actually, before we set checkin-needed, you need to change the commit message to match the guidelines and re-uploadyour patch:

Guidelines:
http://blog.bonardo.net/2010/06/22/so-youre-about-to-use-checkin-needed

i.e. "Bug 620164 - Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset. r=cpearce"

Comment 12

[Abhishek Bhatnagar]

Attachment: Proposed patch for Bug 620164

Removing CheckedInt and Muloverflow from nsTheoraState::MaxKeyframeOffset

Comment 13

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a5d3abd78272

Thanks for the patch! If you're ever looking for more bugs to work on, feel free to drop by #developers on irc.mozilla.org.

Comment 14

[:ehsan akhgari]

https://hg.mozilla.org/mozilla-central/rev/a5d3abd78272