Bug 74784 – Add support for 8 bits strings to Document::isValidName()

• Component: WebKit Misc.
• Status: RESOLVED FIXED

[Benjamin Poulain]

We don't have to go through the slow case when we have a 8 bit string.

[Benjamin Poulain]

Created attachment 119719 [details]
Patch

[Andreas Kling]

Comment on attachment 119719 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=119719&action=review

Ahh, these 8-bit string hacks are just great! :)

> Source/WebCore/ChangeLog:9
> +        The valid name has a fast path for ASCII, and a slow path
> +        taking into account Unicode characters.

... a slow path taking Unicode characters into account.

> Source/WebCore/ChangeLog:12
> +        For 8 bits strings, we do not need to test again the slow path
> +        as it should never succeed if the fast path does not succed.

Since the slow path here is really the non-ASCII path, you could be more specific, e.g:
"For 8-bit strings, we don't need to take the non-ASCII path as it could never succeed if the ASCII path didn't."

> Source/WebCore/dom/Document.cpp:3802
> +    return isValidNameASCII(name.characters16(), length) || isValidNameNonASCII(name.characters16(), length);

We should cache the return value from characters16() to avoid doing the String::m_impl null check again.

[Benjamin Poulain]

> We should cache the return value from characters16() to avoid doing the String::m_impl null check again.

Good catch Baron Kling.

[Benjamin Poulain]

Committed r103205: <http://trac.webkit.org/changeset/103205>

[Darin Adler]

Comment on attachment 119719 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=119719&action=review

>> Source/WebCore/ChangeLog:12
>> +        as it should never succeed if the fast path does not succed.
> 
> Since the slow path here is really the non-ASCII path, you could be more specific, e.g:
> "For 8-bit strings, we don't need to take the non-ASCII path as it could never succeed if the ASCII path didn't."

Is that true? Are there no valid name characters in the 0x80-0xFF range that could be in 8-bit strings? That doesn’t sound right to me.

[Benjamin Poulain]

> > Since the slow path here is really the non-ASCII path, you could be more specific, e.g:
> > "For 8-bit strings, we don't need to take the non-ASCII path as it could never succeed if the ASCII path didn't."
> 
> Is that true? Are there no valid name characters in the 0x80-0xFF range that could be in 8-bit strings? That doesn’t sound right to me.

All I had seen in the nonASCII path is characters > 0xFF. Did I miss something?

[Darin Adler]

(In reply to comment #6)
> All I had seen in the nonASCII path is characters > 0xFF. Did I miss something?

Well, I went to read isValidNameStart and isValidNamePart and I immediately spotted these:

- All letters are legal name starts. I believe there are many letters in the U+0080-U+00FF range, such as U+00C0.

- U+00B7 is a legal name part. It’s right there as a constant in isValidNamePart.

So I think there are many non-ASCII valid name characters that fit in 8-bit.

[Benjamin Poulain]

I checked the Unicode categories of the valid characters, and there are some in 0x80-0xFF.

The function should probably be changed isValidNameASCII() -> isValidNameLatin1. Or if not possible, skip the shortcut altogether.

I will revert this. There should be tests for those cases.

[Darin Adler]

(In reply to comment #8)
> The function should probably be changed isValidNameASCII() -> isValidNameLatin1.

That might be OK.

> Or if not possible, skip the shortcut altogether.

Why!? Valid ASCII names are far more common than other kind of name, so it’s great to have a fast case for that.

> There should be tests for those cases.

Absolutely. I agree!

[Benjamin Poulain]

> > Or if not possible, skip the shortcut altogether.
> 
> Why!? Valid ASCII names are far more common than other kind of name, so it’s great to have a fast case for that.

I meant skipping the Unicode test if the string is 8 bits.

The current fast path totally make sense, I have no intention to change that.

[Benjamin Poulain]

Reverted in r103210. Reopening to fix this again with tests.

[Darin Adler]

I suggest logic that uses characters() if the 8-bit string is not all ASCII. I think it’s OK to expand the string to 16-bit to check if it’s valid if the string has non-ASCII characters in it. I believe those cases are rare enough that we don’t need to optimize them.

[Darin Adler]

Here’s the code I suggest:

    UChar* characters;
    if (name.is8Bit()) {
        if (isValidNameASCII(name.characters8(), length))
            return true;
        characters = name.characters();
    } else {
        characters = name.characters16();
        if (isValidNameASCII(characters, length))
            return true;
    }
    return isValidNameNonASCII(characters, length);

You like that?

[Benjamin Poulain]

(In reply to comment #13)
> Here’s the code I suggest:
> 
>     UChar* characters;
>     if (name.is8Bit()) {
>         if (isValidNameASCII(name.characters8(), length))
>             return true;
>         characters = name.characters();
>     } else {
>         characters = name.characters16();
>         if (isValidNameASCII(characters, length))
>             return true;
>     }
>     return isValidNameNonASCII(characters, length);
> 
> You like that?

Sure, I think that is the right thing to do. I just want to add tests before updating and I am busy with a merge at the moment.

[Benjamin Poulain]

Created attachment 119815 [details]
Patch

[WebKit Review Bot]

Comment on attachment 119815 [details]
Patch

Clearing flags on attachment: 119815

Committed r103222: <http://trac.webkit.org/changeset/103222>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.