NetworkProcess aborts in WebKit::NetworkLoad::didCompleteWithError at Source/WebKit2/NetworkProcess/NetworkLoad.cpp:423

[Andres Gomez Garcia]

Created attachment 302925 [details]
BT from gdb for the NetworkProcess

I'm using WebKitGtk+ with my own JHBuild setting:
https://github.com/tanty/jhbuild-epiphany/tree/master

Epiphany 3.22.5 and WebKit 2.15.90.

However, the rest of the dependencies are all provided from Debian Testing.

The compilation was done with CMake args:

'-DPORT=GTK -DCMAKE_BUILD_TYPE=Release -DENABLE_MINIBROWSER=ON -DCMAKE_C_FLAGS_RELEASE="-O0 -g -DNDEBUG  -DG_DISABLE_CAST_CHECKS" -DCMAKE_CXX_FLAGS_RELEASE="-O0 -g -DNDEBUG -DG_DISABLE_CAST_CHECKS"'

After visiting several pages, eventually, the NetworkProcess aborts.

This bug is not reproducible in a predictable way.

[Andres Gomez Garcia]

(In reply to comment #0)
> Epiphany 3.22.5

Epiphany 3.22.6

[Carlos Garcia Campos]

I don't think this is specific to GTK+

[Carlos Garcia Campos]

Created attachment 302929 [details]
Patch

I haven't been able to reproduce the crash, but looking at the bt, I think this patch should fix it.

[Andres Gomez Garcia]

This is *very* easy to reproduce ... :(

[Carlos Garcia Campos]

(In reply to comment #4)
> This is *very* easy to reproduce ... :(

Could you try the patch then? It has never happened to me.

[Chris Dumez]

Antti, can you please take a look, you added this code recently.

[Antti Koivisto]

Comment on attachment 302929 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=302929&action=review

> Source/WebKit2/NetworkProcess/cache/NetworkCacheSpeculativeLoad.cpp:86
> -    didComplete();
> -
> -    // This causes call to didFailLoading().
> -    if (load)
> +    if (load) {
> +        // This causes call to didFailLoading() that calls didComplete().
>          load->continueWillSendRequest({ });
> +    } else
> +        didComplete();

This would break the functionality since didFailLoading will null out the m_response before didComplete. I think the right fix is just to remove the continueWillSendRequest and let the load be deleted.

[Carlos Garcia Campos]

Created attachment 303045 [details]
Updated patch

New patch after talking to Antti on IRC. Previous patch was not correct, because I didn't take into account that didFailLoading deletes the cache entry.

[Carlos Garcia Campos]

Committed r213206: <http://trac.webkit.org/changeset/213206>