Bug 215830 – HashMap<Ref<T>>::take should return RefPtr<T>

• Status: RESOLVED
• Resolution: FIXED
• Component: Web Template Framework

[Ryosuke Niwa]

HashMap<Ref<T>>::take should return RefPtr<T> instead of Optional<Ref<T>>.

[Ryosuke Niwa]

Created attachment 407253 [details]
Patch

[Darin Adler]

Comment on attachment 407253 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=407253&action=review

Looks like an improvement at almost every call site.

> Source/WTF/wtf/HashTraits.h:215
> +    typedef RefPtr<P> TakeType;

Since we’re touching it we could use using instead of typedef.

> Source/WebCore/dom/ScriptRunner.cpp:102
> -        ASSERT(m_pendingAsyncScripts.contains(pendingScript));
> -        m_scriptsToExecuteSoon.append(m_pendingAsyncScripts.take(pendingScript)->ptr());
> +        auto script = m_pendingAsyncScripts.take(pendingScript);
> +        ASSERT(script);
> +        m_scriptsToExecuteSoon.append(script.releaseNonNull());

This is an example where the "after" looks worse because releaseNonNull is such a long function name. But I see that there was reference count churn in the old version!

No need to assert non-null separately before calling releaseNonNull, because that asserts. I would write this as a one-liner, I think.

[Ryosuke Niwa]

(In reply to Darin Adler from comment #2)
> Comment on attachment 407253 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=407253&action=review
> 
> Looks like an improvement at almost every call site.
> 
> > Source/WTF/wtf/HashTraits.h:215
> > +    typedef RefPtr<P> TakeType;
> 
> Since we’re touching it we could use using instead of typedef.

Fixed. I'm gonna also fix PeekType.

> > Source/WebCore/dom/ScriptRunner.cpp:102
> > -        ASSERT(m_pendingAsyncScripts.contains(pendingScript));
> > -        m_scriptsToExecuteSoon.append(m_pendingAsyncScripts.take(pendingScript)->ptr());
> > +        auto script = m_pendingAsyncScripts.take(pendingScript);
> > +        ASSERT(script);
> > +        m_scriptsToExecuteSoon.append(script.releaseNonNull());
> 
> This is an example where the "after" looks worse because releaseNonNull is
> such a long function name. But I see that there was reference count churn in
> the old version!
> 
> No need to assert non-null separately before calling releaseNonNull, because
> that asserts. I would write this as a one-liner, I think.

Sure, I'd use the one liner.

[Ryosuke Niwa]

Created attachment 407256 [details]
Patch for landing

[EWS]

Committed r266157: <https://trac.webkit.org/changeset/266157>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 407256 [details].

[Radar WebKit Bug Importer]

<rdar://problem/67774006>