581944 - Reorder the HTML5 tokenizer's states to make it easier to optionally partition the tokenizer loop into two methods

Status: RESOLVED FIXED

(Core :: DOM: HTML Parser, defect, P3)

Description

[Henri Sivonen (:hsivonen)]

Attachment: Reorder tokenizer states

Currently, the method that holds the main loop of the HTML5 tokenizer compiles to over 8000 bytes in Java. This means that HotSpot in its default configuration refuses to JIT it.

To work around this, I need to reorder the tokenizer states so that it's easy to have a script that extracts a some states into a separate method so that I get two methods that each compile to less than 8000 bytes.

To keep the whole thing sanely maintainable, I need to apply the same reordering to Gecko--before Gecko 2.0 diverges from trunk.

This patch should not have change any behavior in Gecko. Note that the states will stay as one method in Gecko. I'm planning on partitioning the method into two only in Java releases.

(The patch is unreviewable, of course... but I'm hoping to get an rs on it...)

Comment 1

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 460236 [details] [diff] [review]
Reorder tokenizer states

rs=me based on your description.

Comment 2

[Henri Sivonen (:hsivonen)]

Attachment: Reorder tokenizer states for maintainability (rs=sicking)

Requesting approval for 2.0, because having the states in different order in 2.0 and in post-2.0 trunk development would make maintenance harder (if tokenizer changes to the branch are required).

If this patch gets approved while I'm on vacation, please mark this bug 'checkin-needed'.

Comment 3

[Johnny Stenback (:jst)]

Pushed to mozilla-central.

http://hg.mozilla.org/mozilla-central/rev/19367a75c1b6