LibWeb: Fix GC leaks in Fetch::Infrastructure::Body::fully_read()

By making this function accept the success and error steps as
HeapFunction rather than SafeFunction, we break a bunch of strong
GC cycles.
This commit is contained in:
Andreas Kling 2024-04-23 10:25:20 +02:00
commit 184368285c
Notes: sideshowbarker 2024-07-17 05:09:48 +09:00
9 changed files with 67 additions and 67 deletions

View file

@ -986,7 +986,7 @@ WebIDL::ExceptionOr<void> HTMLMediaElement::fetch_resource(URL::URL const& url_r
// 2. Let updateMedia be to queue a media element task given the media element to run the first appropriate steps from the media data processing
// steps list below. (A new task is used for this so that the work described below occurs relative to the appropriate media element event task
// source rather than using the networking task source.)
auto update_media = [this, failure_callback = move(failure_callback)](auto media_data) mutable {
auto update_media = JS::create_heap_function(heap(), [this, failure_callback = move(failure_callback)](ByteBuffer media_data) mutable {
// 6. Update the media data with the contents of response's unsafe response obtained in this fashion. response can be CORS-same-origin or
// CORS-cross-origin; this affects whether subtitles referenced in the media data are exposed in the API and, for video elements, whether
// a canvas gets tainted when the video is drawn on it.
@ -1005,7 +1005,7 @@ WebIDL::ExceptionOr<void> HTMLMediaElement::fetch_resource(URL::URL const& url_r
if (m_ready_state == ReadyState::HaveMetadata)
set_ready_state(ReadyState::HaveEnoughData);
});
};
});
// FIXME: 3. Let processEndOfMedia be the following step: If the fetching process has completes without errors, including decoding the media data,
// and if all of the data is available to the user agent without network access, then, the user agent must move on to the final step below.
@ -1015,12 +1015,12 @@ WebIDL::ExceptionOr<void> HTMLMediaElement::fetch_resource(URL::URL const& url_r
// 5. Otherwise, incrementally read response's body given updateMedia, processEndOfMedia, an empty algorithm, and global.
VERIFY(response->body());
auto empty_algorithm = [](auto) {};
auto empty_algorithm = JS::create_heap_function(heap(), [](JS::GCPtr<WebIDL::DOMException>) {});
// FIXME: We are "fully" reading the response here, rather than "incrementally". Memory concerns aside, this should be okay for now as we are
// always setting byteRange to "entire resource". However, we should switch to incremental reads when that is implemented, and then
// implement the processEndOfMedia step.
response->body()->fully_read(realm, move(update_media), move(empty_algorithm), JS::NonnullGCPtr { global }).release_value_but_fixme_should_propagate_errors();
response->body()->fully_read(realm, update_media, empty_algorithm, JS::NonnullGCPtr { global }).release_value_but_fixme_should_propagate_errors();
};
m_fetch_controller = TRY(Fetch::Fetching::fetch(realm, request, Fetch::Infrastructure::FetchAlgorithms::create(vm, move(fetch_algorithms_input))));