Bug 30506 – Resources that the server sent as 304 not modified are not shown in the inspectors resource pane

• Component: Web Inspector (Deprecated)
• Status: RESOLVED FIXED

[Martin Häcker]

Which is unfortunate, as I often want to inspec them - even if Safari pulled them from it's internal cache. Interestingly this behaviour persists even if you deactivate all caches from the "developer" menu.

[Alexey Proskuryakov]

What nightly build revision are you seeing this with? Is this a regression?

I understand you're saying this happens everywhere, but nonetheless, do you have a specific example of a page we could refer to when investigating this bug?

[Martin Häcker]

I've seen it with builds for at least the last several months, the latest with build 49764.

I have also seen it with Safari 4.0.3 so if it is a regression it probably was in 4 already.

For reference, it mostly annoys me when I am debugging our web application that we develop in-house.

Here's the homepage http://www.agile42.com/cms/pages/agilo/ and the accompanying trac instance: https://dev.agile42.com/

It is open source, so you could get it to test with, I do suspect that this error is so widespread that you can test it with any page that sends 304s for objects.

I would advise to try it with javascript files, because it becomes really apparant if you hit a breakpoint in a file that is just displayed as empty or step into a function just to be shown an empty screen.

Something I just noticed: sometimes reloading the page and trying again helps (but its more a 1 out of 10 tries works) scenario and it almost always means that other files will then be hidden.

Still this points to the theory that this could also be a timing issue (which I hope it's not).

Regards,
Martin

[Alexey Proskuryakov]

<rdar://problem/7319845>

[Alexey Proskuryakov]

Looks like the main culprit here is preloading logic (I'm not even sure if this only happens for 304 Not Modified responses). Here is what happens:

1. As a document is loaded, preload scanner is invoked, requesting subresources while HTML parser is blocked waiting for the first script to download.
2. These preloads result in InspectorController::addResource being called, but cache->requestResource() returns null, thus the resource is never added to document resource set in DocLoader.
3. Once a response for preload is received, Inspector asks DocLoader about it (in InspectorResource::updateScriptObject), but DocLoader says that there is no such resource.

It's no good that DocLoader doesn't know about resources that were revalidated by preloader.

[Alexey Proskuryakov]

Seems related to FrameLoader::loadedResourceFromMemoryCache() not being called in this case.

[Alexey Proskuryakov]

Created attachment 41692 [details]
proposed fix

This isn't exactly the kind of fix I was trying to make - as mentioned above, it seems that client callbacks are messed up when preloading meets revalidation, which this doesn't correct. But it is a simple and hopefully safe fix.

[Alexey Proskuryakov]

Committed <http://trac.webkit.org/changeset/49960>.