Bug 107317 – Make CompactHTMLToken a little more compact

• Component: New Bugs
• Status: RESOLVED FIXED

[Tony Gentilcore]

Make CompactHTMLToken a little more compact

[Tony Gentilcore]

Created attachment 183533 [details]
Patch

[Eric Seidel (no email)]

Comment on attachment 183533 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=183533&action=review

> Source/WebCore/html/parser/BackgroundHTMLParser.cpp:129
> +        if (m_token.type() == HTMLTokenTypes::DOCTYPE)
> +            m_pendingTokens.append(DocTypeCompactHTMLToken(m_token));

How can this work?  Aren't we appending these by value, not pointers?

[Tony Gentilcore]

(In reply to comment #2)
> (From update of attachment 183533 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=183533&action=review
> 
> > Source/WebCore/html/parser/BackgroundHTMLParser.cpp:129
> > +        if (m_token.type() == HTMLTokenTypes::DOCTYPE)
> > +            m_pendingTokens.append(DocTypeCompactHTMLToken(m_token));
> 
> How can this work?  Aren't we appending these by value, not pointers?

Oops, I'll have to think of another way to go about this.

[Adam Barth]

Comment on attachment 183533 [details]
Patch

Maybe an OwnPtr to hold the DocType specific values?  Normally that will be 0, which is smaller and won't try to ref/deref twice.

[Tony Gentilcore]

Created attachment 183584 [details]
Patch

[Eric Seidel (no email)]

Created attachment 183619 [details]
An alternate approach

[Eric Seidel (no email)]

I hope you don't mind me hijacking your bug.  Please feel free to obsolete my patch and re-state yours for review if you like!

[Adam Barth]

Comment on attachment 183619 [details]
An alternate approach

That's awesome.

[Adam Barth]

Comment on attachment 183619 [details]
An alternate approach

View in context: https://bugs.webkit.org/attachment.cgi?id=183619&action=review

> Source/WebCore/html/parser/CompactHTMLToken.h:71
> +    const String& publicIdentifier() const { return m_attributes[0].name(); }
> +    const String& systemIdentifier() const { return m_attributes[0].value(); }

Can we ASSERT that m_attributes has at least size one?  Maybe Vector does that for us?

> Source/WebCore/html/parser/CompactHTMLToken.h:74
> +    unsigned m_type : 4;

Can we ASSERT that we have enough bits somehow?

[Eric Seidel (no email)]

Comment on attachment 183619 [details]
An alternate approach

View in context: https://bugs.webkit.org/attachment.cgi?id=183619&action=review

>> Source/WebCore/html/parser/CompactHTMLToken.h:71
>> +    const String& systemIdentifier() const { return m_attributes[0].value(); }
> 
> Can we ASSERT that m_attributes has at least size one?  Maybe Vector does that for us?

Yup:
http://trac.webkit.org/browser/trunk/Source/WTF/wtf/Vector.h#L553

>> Source/WebCore/html/parser/CompactHTMLToken.h:74
>> +    unsigned m_type : 4;
> 
> Can we ASSERT that we have enough bits somehow?

I believe the compiler does this for us (at least GCC) when we assign HTMLTokenTypes::Type to this value.  if it were too small it would warn.  At least that's my recollection.

[Eric Seidel (no email)]

Created attachment 183930 [details]
Updated to TOT

[Darin Adler]

Comment on attachment 183930 [details]
Updated to TOT

View in context: https://bugs.webkit.org/attachment.cgi?id=183930&action=review

I am not an expert on the parser, but the patch looks fine to me and probably does not require parser expertise.

> Source/WebCore/ChangeLog:13
> +        however when I tried to assert that it failed.  Presumably because
> +        Vector<T> is more than void* in size.

Yes, Vector<T> is much more than void* in size!

> Source/WebCore/html/parser/CompactHTMLToken.h:78
>      Vector<CompactAttribute> m_attributes;

What’s the typical size of this vector? If it’s typically empty, it could be OwnPtr<Vector>. If it typically has a small number of attributes, maybe it should have inline capacity.

[Darin Adler]

Comment on attachment 183930 [details]
Updated to TOT

View in context: https://bugs.webkit.org/attachment.cgi?id=183930&action=review

>> Source/WebCore/ChangeLog:13
>> +        Vector<T> is more than void* in size.
> 
> Yes, Vector<T> is much more than void* in size!

Roughly speaking, it’s got a pointer for the storage, a size, and a capacity. So we pay for the flexibility of being able to efficiently resize. For specialized uses we can use a different data structure that is more compact.

[WebKit Review Bot]

Comment on attachment 183930 [details]
Updated to TOT

Rejecting attachment 183930 [details] from commit-queue.

Failed to run "['/mnt/git/webkit-commit-queue/Tools/Scripts/webkit-patch', '--status-host=queues.webkit.org', '-..." exit_code: 2 cwd: /mnt/git/webkit-commit-queue

Last 500 characters of output:
-fno-rtti -fno-threadsafe-statics -fvisibility-inlines-hidden -Wsign-compare  -c ../../Source/WebCore/html/parser/CompactHTMLToken.cpp -o obj/Source/WebCore/html/parser/webcore_html.CompactHTMLToken.o
../../Source/WebCore/html/parser/CompactHTMLToken.cpp: In constructor 'WebCore::CompactHTMLToken::CompactHTMLToken(const WebCore::HTMLToken&)':
../../Source/WebCore/html/parser/CompactHTMLToken.cpp:55: error: 'stringFromToken' was not declared in this scope
ninja: build stopped: subcommand failed.

Full output: http://queues.webkit.org/results/16039638

[Eric Seidel (no email)]

Comment on attachment 183930 [details]
Updated to TOT

View in context: https://bugs.webkit.org/attachment.cgi?id=183930&action=review

>>> Source/WebCore/ChangeLog:13
>>> +        Vector<T> is more than void* in size.
>> 
>> Yes, Vector<T> is much more than void* in size!
> 
> Roughly speaking, it’s got a pointer for the storage, a size, and a capacity. So we pay for the flexibility of being able to efficiently resize. For specialized uses we can use a different data structure that is more compact.

Yeah, I forgot about the necessary size member when writing that. :)  I didn't realize we had another one for capacity as well.

[Eric Seidel (no email)]

Comment on attachment 183930 [details]
Updated to TOT

View in context: https://bugs.webkit.org/attachment.cgi?id=183930&action=review

>> Source/WebCore/html/parser/CompactHTMLToken.h:78
>>      Vector<CompactAttribute> m_attributes;
> 
> What’s the typical size of this vector? If it’s typically empty, it could be OwnPtr<Vector>. If it typically has a small number of attributes, maybe it should have inline capacity.

It's only used for attribute tokens, which are less common than character tokens for sure.  We probably should make it OwnPtr<Vector>

[Eric Seidel (no email)]

Created attachment 184026 [details]
Patch for landing

[WebKit Review Bot]

Comment on attachment 184026 [details]
Patch for landing

Attachment 184026 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/16042706

[Eric Seidel (no email)]

Created attachment 184028 [details]
Patch for landing

[Eric Seidel (no email)]

Sorry, this was originally part of a fancier patch series.   An earlier patch never landed.

[WebKit Review Bot]

Comment on attachment 184028 [details]
Patch for landing

Rejecting attachment 184028 [details] from commit-queue.

Failed to run "['/mnt/git/webkit-commit-queue/Tools/Scripts/webkit-patch', '--status-host=queues.webkit.org', '-..." exit_code: 2 cwd: /mnt/git/webkit-commit-queue

Last 500 characters of output:
r'
../../Source/WebCore/html/parser/CompactHTMLToken.cpp:55: error:   crosses initialization of 'WTF::String publicIdentifier'
../../Source/WebCore/html/parser/CompactHTMLToken.cpp:76: error: jump to case label
../../Source/WebCore/html/parser/CompactHTMLToken.cpp:56: error:   crosses initialization of 'WTF::String systemIdentifier'
../../Source/WebCore/html/parser/CompactHTMLToken.cpp:55: error:   crosses initialization of 'WTF::String publicIdentifier'
ninja: build stopped: subcommand failed.

Full output: http://queues.webkit.org/results/16036725

[Eric Seidel (no email)]

Created attachment 184035 [details]
Patch for landing

[WebKit Review Bot]

Comment on attachment 184035 [details]
Patch for landing

Clearing flags on attachment: 184035

Committed r140453: <http://trac.webkit.org/changeset/140453>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Eric Seidel (no email)]

This broke the Windows builder because MSVC won't compact adjacent bitfields of differing types. I have to change "bool" to "unsigned" and it will work.

[Eric Seidel (no email)]

http://build.chromium.org/p/chromium.webkit/builders/WebKit%20Win%20Builder/builds/35188 is the log from the break.

[Eric Seidel (no email)]

Reopening to attach new patch.

[Eric Seidel (no email)]

Created attachment 184046 [details]
Patch for landing

[Hin-Chung Lam]

Comment on attachment 184046 [details]
Patch for landing

Clearing flags on attachment: 184046

Committed r140473: <http://trac.webkit.org/changeset/140473>

[Hin-Chung Lam]

All reviewed patches have been landed.  Closing bug.