Bug 231410 – Scroll To Text Fragment directive parsing

• Component: New Bugs
• Status: RESOLVED FIXED

[Megan Gardner]

Scroll To Text Fragment directive parsing

[Megan Gardner]

Created attachment 440570 [details]
Patch

[Tim Horton]

Comment on attachment 440570 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440570&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:62
> +    m_URLFragment = fragmentIdentifier.substring(0, fragmentDirectiveStart);

I would call this like "remaining fragment" or something. Also the comment should have a capital/punctuation.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:84
> +    LOG_WITH_STREAM(Scrolling, stream << *this << " parseFragmentDirective: ");

I think this probably should not use the Scrolling log channel

> Source/WebCore/dom/FragmentDirectiveParser.cpp:88
> +        // FIXME: add decoding for % encoded strings.

This is fairly critical, because it comes up as soon as you have a space in the string :)

Oddly the decoding code that decodeURIComponent is inline in JSGlobalObjectFunctions (and the URL parser's is similarly internal to it), so it's not clear what the right way to do this is.

> Source/WebCore/dom/FragmentDirectiveParser.h:45
> +class FragmentDirectiveParser : public RefCounted<FragmentDirectiveParser> {

Unclear why this is refcounted, probably just a plain class?

> Source/WebCore/page/FrameView.cpp:2224
> +    }

I'm guessing you either want to bail here, or pass the fragment parser's "remaining fragment" on down below, but likely /not/ fall through like this.

[Megan Gardner]

Created attachment 440628 [details]
Patch

[Megan Gardner]

Created attachment 440632 [details]
Patch

[Megan Gardner]

Created attachment 440638 [details]
Patch

[Megan Gardner]

Created attachment 440658 [details]
Patch

[Megan Gardner]

Created attachment 440699 [details]
Patch

[Megan Gardner]

Created attachment 440823 [details]
Patch

[Megan Gardner]

Created attachment 440843 [details]
Patch

[Alex Christensen]

Comment on attachment 440843 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440843&action=review

Initial glance through seems reasonable, needs some polish and tests.

> Source/WebCore/dom/Document.h:1631
> +    String& fragmentDirective() { return m_fragmentDirective; }

This one should probably be removed.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:37
> +    String fragmentDirectiveDelimiter = ":~:"_s;

ASCIILiteral, which would prevent an allocation.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:40
> +    m_isValid = false;

This should probably be done with an initializer list in the header
bool m_isValid { false };

> Source/WebCore/dom/FragmentDirectiveParser.cpp:79
> +    auto directives = fragmentDirective.split("&");

split('&')

> Source/WebCore/dom/FragmentDirectiveParser.h:32
> +class URL;

This should probably be replaced by wtf/Forward.h, which also has "using WTF::URL"

> Source/WebCore/dom/FragmentDirectiveParser.h:52
> +    String fragmentDirective() { return m_fragmentDirective; };
> +    StringView remainingURLFragment() { return m_remainingURLFragment; };

const String& fragmentDirective() const
StringView remainingURLFragment() const

> Source/WebCore/dom/FragmentDirectiveParser.h:57
> +    bool parseFragmentDirective(String fragmentDirective);

Can this take a StringView instead?

[Chris Dumez]

Comment on attachment 440843 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440843&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:68
> +bool FragmentDirectiveParser::parseFragmentDirective(String fragmentDirective)

I agree with Alex that this should take in a StringView, but even if it took at String, we would not pass it by value.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:83
> +    for (auto directive : directives) {

auto&

> Source/WebCore/dom/FragmentDirectiveParser.cpp:90
> +        for (auto token : tokens) {

auto&

> Source/WebCore/dom/FragmentDirectiveParser.cpp:99
> +            tokens[0].truncate(tokens[0].length()-2);

missing spaces around -

What if the token is '-' ? (would result in `1 - 2`)

> Source/WebCore/dom/FragmentDirectiveParser.h:46
> +    WTF_MAKE_FAST_ALLOCATED(FragmentDirectiveParser);

Not needed since you subclass RefCounted.

> Source/WebCore/dom/FragmentDirectiveParser.h:48
> +    FragmentDirectiveParser(const URL&);

explicit?

> Source/WebCore/dom/FragmentDirectiveParser.h:50
> +    Vector<ParsedTextDirective> parsedTextDirectives() { return m_parsedTextDirectives; };

I don't think you mean to copy the vector here? Should probably be const and return a `const Vector<>&`

> Source/WebCore/dom/FragmentDirectiveParser.h:51
> +    String fragmentDirective() { return m_fragmentDirective; };

ditto.

> Source/WebCore/dom/FragmentDirectiveParser.h:53
> +    bool isValid() { return  m_isValid; };

should be const.

> Source/WebCore/page/FrameView.cpp:2218
> +        auto fragmentDirectiveParser = FragmentDirectiveParser(url);

FragmentDirectiveParser fragmentDirectiveParser(url); seems shorter.

[Tim Horton]

Comment on attachment 440843 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440843&action=review

>> Source/WebCore/dom/FragmentDirectiveParser.h:46
>> +    WTF_MAKE_FAST_ALLOCATED(FragmentDirectiveParser);
> 
> Not needed since you subclass RefCounted.

Oh, we discussed earlier not making this thing refcounted (see the public constructor and lack of ::create), but I think Megan missed removing the inheritance from RefCounted.

[Chris Dumez]

Comment on attachment 440843 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440843&action=review

>>> Source/WebCore/dom/FragmentDirectiveParser.h:46
>>> +    WTF_MAKE_FAST_ALLOCATED(FragmentDirectiveParser);
>> 
>> Not needed since you subclass RefCounted.
> 
> Oh, we discussed earlier not making this thing refcounted (see the public constructor and lack of ::create), but I think Megan missed removing the inheritance from RefCounted.

Good point. Having a public constructor AND subclassing RefCounted<> is definitely wrong :)

[Megan Gardner]

Yes, I didn't remove the RefCounted bit, will fix shortly.

[Megan Gardner]

Created attachment 440874 [details]
Patch

[Megan Gardner]

Created attachment 440962 [details]
Patch

[Tim Horton]

Comment on attachment 440874 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440874&action=review

> Source/WebCore/ChangeLog:11
> +        * Sources.txt:

This whole section could use more words.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:29
> +#import "Logging.h"

#include to fix the windows build

> Source/WebCore/dom/FragmentDirectiveParser.cpp:57
> +    // FIXME: this needs to be set on the original URL

Complete sentences with capitals and punctuations! Also I think it is a lie, you can't modify the loaded URL; the real thing you want to do is expose the de-directived fragment to other things in WebCore that use the fragment (e.g. normal scroll-to-fragment)... which it looks like you already do. Should talk to Alex.

> Source/WebCore/page/FrameView.cpp:2225
> +            

Shouldn't this `return` here?

[Chris Dumez]

Comment on attachment 440962 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440962&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:61
> +    parseFragmentDirective(fragmentDirectiveString);

Should pass fragmentDirective directly here. Going StringView -> String -> StringView is wrong.

> Source/WebCore/dom/FragmentDirectiveParser.h:43
> +    FragmentDirectiveParser(const URL&);

explicit

> Source/WebCore/dom/FragmentDirectiveParser.h:52
> +    bool parseFragmentDirective(StringView fragmentDirective);

parameter name can be omitted.

[Wenson Hsieh]

Comment on attachment 440962 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=440962&action=review

Looks reasonable to me! Just a few minor comments.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:41
> +    if (!fragmentIdentifier || fragmentIdentifier.isEmpty()) {

Nit - just `if (fragmentIdentifier.isEmpty())` is sufficient here, since the null string is considered empty.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:42
> +        m_fragmentDirective = String();

Nit - I think you can just omit this line, since `m_fragmentDirective` is already initialized to the null string.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:50
> +        m_fragmentDirective = String();

(Ditto)

> Source/WebCore/dom/FragmentDirectiveParser.cpp:58
> +    m_remainingURLFragment = fragmentIdentifier.substring(0, fragmentDirectiveStart);

Nit - `m_remainingURLFragment = fragmentIdentifier.left(fragmentDirectiveStart);` might be a tiny bit cleaner.

> Source/WebCore/page/FrameView.cpp:2213
> +    auto& document = *frame().document();

Nit - this should be a Ref or RefPtr.

[Megan Gardner]

Created attachment 440975 [details]
Patch

[Megan Gardner]

Created attachment 440989 [details]
Patch

[Megan Gardner]

Created attachment 441016 [details]
Patch

[Chris Dumez]

Comment on attachment 441016 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441016&action=review

r=me with changes.

> Source/WebCore/dom/Document.h:2225
> +    StringView m_fragmentDirective;

It doesn't seem right of have a StringView data member on the Document. What guarantees that the underlying String stays alive?

We probably want to use String here and `const String&` in the getter/setter.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:73
> +    fragmentDirective = fragmentDirective.substring(textDirective.length(), fragmentDirective.length());

Do you really need to specify the second parameter? Seems you want everything after the `text=` prefix.

> Source/WebCore/dom/FragmentDirectiveParser.h:29
> +#include <wtf/IsoMalloc.h>

Should not be needed.

> Source/WebCore/dom/FragmentDirectiveParser.h:41
> +    WTF_MAKE_FAST_ALLOCATED(FragmentDirectiveParser);

Should not be needed since you only stack-allocate it.

> Source/WebCore/dom/FragmentDirectiveParser.h:45
> +    const Vector<ParsedTextDirective>& parsedTextDirectives() { return m_parsedTextDirectives; };

getter should be marked as const.

> Source/WebCore/page/FrameView.cpp:2224
> +            // TODO: Scroll to the range specified by the directive

Per coding style (https://webkit.org/code-style-guidelines/#comments), this should be FIXME, not TODO.

[Alex Christensen]

Comment on attachment 441016 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441016&action=review

>> Source/WebCore/dom/Document.h:2225
>> +    StringView m_fragmentDirective;
> 
> It doesn't seem right of have a StringView data member on the Document. What guarantees that the underlying String stays alive?
> 
> We probably want to use String here and `const String&` in the getter/setter.

Definitely don't store a StringView on the Document.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:81
> +        Vector<String> tokens;
> +        for (auto token : directive.split(','))
> +            tokens.append(token.toString());

Could this be a Vector<StringView>?
nit: I think WTF::map could be used here to be more concise.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:98
> +            tokens.remove(0);

Removing the first element of a Vector isn't efficient.  Could tokens be a Deque instead?

> Source/WebCore/dom/FragmentDirectiveParser.cpp:107
> +        if (tokens.size() != 1 && tokens.size() != 2) {
> +            LOG_WITH_STREAM(TextFragment, stream << " not enough tokens ");

not enough or too many?

[Megan Gardner]

Comment on attachment 441016 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441016&action=review

>> Source/WebCore/dom/FragmentDirectiveParser.cpp:81
>> +            tokens.append(token.toString());
> 
> Could this be a Vector<StringView>?
> nit: I think WTF::map could be used here to be more concise.

I tried to do a vector of StringViews, but in the end, we want strings, and the change was easier to do here than later.

[Megan Gardner]

Created attachment 441141 [details]
Patch

[Chris Dumez]

Comment on attachment 441141 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441141&action=review

> Source/WebCore/dom/Document.h:1635
> +    void setFragmentDirective(String& fragmentDirective) { m_fragmentDirective = fragmentDirective; }

const String&

> Source/WebCore/dom/FragmentDirectiveParser.cpp:92
> +                continue;

What's the point of this continue statement? Seems like a no-op.

Maybe we could do this logging in the `for (auto token : textDirective.split(','))` loop above to avoid iterating over the tokens again?

[Megan Gardner]

Created attachment 441148 [details]
Patch

[Chris Dumez]

Comment on attachment 441148 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441148&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:82
> +        bool emptyToken = false;

The name is a bit unclear for a boolean. I think it would be nice to rename to containsEmptyToken or similar.

> Source/WebCore/dom/FragmentDirectiveParser.cpp:128
> +    

Extra blank line.

[Chris Dumez]

Comment on attachment 441148 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441148&action=review

> Source/WTF/ChangeLog:6
> +        Reviewed by NOBODY (OOPS!).

Feel free to put my name in the changelog since I already r+

[Megan Gardner]

Created attachment 441153 [details]
Patch

[Alex Christensen]

Comment on attachment 441153 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441153&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:58
> +    parseFragmentDirective(fragmentDirective);

The return value of this isn't used.  Do you want to check if it returns false and early return, or make it return void?

[Megan Gardner]

Created attachment 441170 [details]
Patch for landing

[EWS]

Committed r284143 (242963@main): <https://commits.webkit.org/242963@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 441170 [details].

[Radar WebKit Bug Importer]

<rdar://problem/84228621>

[WebKit Commit Bot]

Re-opened since this is blocked by bug 231870

[Tim Horton]

Reverted r284143 for reason:

Caused a series of assertions in layout tests in EWS

Committed r284326 (243121@main): <https://commits.webkit.org/243121@main>

[Tim Horton]

Reverted the in https://trac.webkit.org/changeset/284326/webkit
I wasn't able to repro the issue from https://bugs.webkit.org/show_bug.cgi?id=231752 locally either, so I didn't manage to fix it, we'll just have to try again; hopefully someone else knows what the assertion is about.

[Chris Dumez]

Comment on attachment 441170 [details]
Patch for landing

View in context: https://bugs.webkit.org/attachment.cgi?id=441170&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:60
> +    m_fragmentDirective = fragmentDirective.toString();

This is wrong. You are allocating a new temporary String and then creating a StringView to that temporary String, since m_fragmentDirective is a StringView. This is very likely what’s causing the crashes.

[Chris Dumez]

Comment on attachment 441170 [details]
Patch for landing

View in context: https://bugs.webkit.org/attachment.cgi?id=441170&action=review

> Source/WebCore/dom/FragmentDirectiveParser.cpp:60
> +    m_fragmentDirective = fragmentDirective.toString();

This is wrong. You are allocating a new temporary String and then creating a StringView to that temporary String, since m_fragmentDirective is a StringView. This is very likely what’s causing the crashes.

[Megan Gardner]

Created attachment 443639 [details]
Patch

[Megan Gardner]

Created attachment 443645 [details]
Patch

[Megan Gardner]

Created attachment 443647 [details]
Patch

[EWS]

Committed r285528 (244045@main): <https://commits.webkit.org/244045@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 443647 [details].