600301 - Optimize ReplacePrep

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Olli Pettay [:smaug][[e-mail]]]

This should also optimize out some MutatePrep calls.
Patch coming soon.

Comment 1

[Olli Pettay [:smaug][[e-mail]]]

Attachment: v1

This is one way to speed up append/assign.
Speeds up GetInnerHTML 5-10% where append is called a lot.

I pushed the patch to tryserver.

Comment 2

[Olli Pettay [:smaug][[e-mail]]]

Benjamin, any chance for a review?
I'm hoping to get this to FF4 as a part of other optimization patches.

Comment 3

[Benjamin Smedberg]

Comment on attachment 479154 [details] [diff] [review]
v1

Does ReplacePrep actually get inlined in this case? That seems like a fair bit of code to get inlined...

Comment 4

[Olli Pettay [:smaug][[e-mail]]]

ReplacePrep is not used in too many places.

Comment 5

[[e-mail]]

How can you be sure that it's safe to mutate the given string in place?

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

Neil, I don't understand what you mean.

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

Neil, I tried to keep the behavior of ReplacePrep and MutatePrep.
Is there some case I missed?

Comment 8

[[e-mail]]

Comment on attachment 479154 [details] [diff] [review]
v1

>+        if (cutStart == mLength && Capacity() > newTotalLen) {
OK, so I worked that last one out. But perhaps you could have checked for cutStart + cutLength == mLength, like ReplacePrep does, thus allowing you to remove the check in ReplacePrep?

Comment 9

[Olli Pettay [:smaug][[e-mail]]]

http://hg.mozilla.org/mozilla-central/rev/9b79a3d0c554