In the scenario where a ServiceWorker returns a pass-through fetch via
`evt.respondWith(fetch("underlying"))`, in order for the "underlying"
HTTP channel to be canceled when the outer HTTP channel is canceled,
FetchDriver's OnDataAvailable method needs to return an error when
the output pipe experiences an error.
Unfortunately, the contract for ReadSegments is effectively that it
returns NS_OK regardless of what the rv of the write handler returned,
so relying on the returned rv is insufficient. And because various
Write*() methods will all fast-path to returning NS_OK if a count of 0
is passed, it's necessary to infer a closed/broken pipe by noticing
that we tried to write more than 0 bytes of data but 0 bytes were
written. (This is safe because the pipe we write into was created
by FetchDriver::OnStartRequest which explicitly creates an infinite
pipe, so it's not possible for the write to fail due to lack of space
in the pipe.)
Create a new class AlternativeDataStreamListener for alternative data and
main data overlap loading.
AlternativeDataStreamListener coorperates with FetchDriver to handle
following situations
1. There is no preferred alternative data type in InternalRequest
Directly using FetchDriver to listen on the opened channel
2. If preferred alternative data type exists in InternalRequest, but no
saved data in cache.
AlternativeDataStreamListener is constructed to listen on the channel,
but its status would be set as FALLBACK and redirect callbacks to
FetchDriver.
3. If preferred alternative data type exists in InternalRequest, and the
data also exists in the cache.
AlternativeDataStreamListener is constructed to listen on the channel
for loading the alternative data. And also open a channel listened by
FetchDriver for loading the main data when AlternativeDataStreamListener::
OnStartRequest is called.
If the cacheEntryId is different between main data channel and
alternative data channel, we will cancel the alternative data loading.
The included test crashes without the included fix if run with --disable-e10s.
e10s doesn't crash because HttpChannelChild examines the return value of the
call to the listener's OnStartRequest method and invokes Cancel() if it
NS_FAILED. This is a divergence between e10s and non-e10s. See the bug for
more details and discussion.
Raise the urgent-start flag in the ClassOfService when the Fetch and XHR are
triggered by user input events. The urgent-start classification will tell the
network requests scheduler to perform it with the highest priority and also
ignoring any parallelism or overall connection limits.
MozReview-Commit-ID: 2YavWbuFaln
This change avoids lots of false positives for Coverity's CHECKED_RETURN
warning, caused by NS_WARN_IF's current use in both statement-style and
expression-style.
In the case where the code within the NS_WARN_IF has side-effects, I made the
following change.
> NS_WARN_IF(NS_FAILED(FunctionWithSideEffects()));
> -->
> Unused << NS_WARN_IF(NS_FAILED(FunctionWithSideEffects()));
In the case where the code within the NS_WARN_IF lacks side-effects, I made the
following change.
> NS_WARN_IF(!condWithoutSideEffects);
> -->
> NS_WARNING_ASSERTION(condWithoutSideEffects, "msg");
This has two improvements.
- The condition is not evaluated in non-debug builds.
- The sense of the condition is inverted to the familiar "this condition should
be true" sense used in assertions.
A common variation on the side-effect-free case is the following.
> nsresult rv = Fn();
> NS_WARN_IF_(NS_FAILED(rv));
> -->
> DebugOnly<nsresult rv> = Fn();
> NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "Fn failed");
