1353044 - No loop of PChildToParentStream and PParentToChildStream

Status: RESOLVED INVALID

(Core :: DOM: Core & HTML, enhancement)

Description

[Andrea Marchesini [:baku]]

When a stream is sent via IPC, it can be a PChildToParentStream or a PParentToChildStream (it depends if it goes from parent to child or the other way around). If that stream is sent it back, we create a new PParentToChildStream (or PChildToParentStream).

In this bug, I want to avoid this, and make gecko smart enough to use the origin inputStream instead the IPC one if a cycle is detected.

Comment 1

[Andrea Marchesini [:baku]]

Attachment: part 1 - ID in IPCStream

qdot, are you interested in nsIInputStream and IPC stream serialization?

Comment 2

[Andrea Marchesini [:baku]]

Attachment: part 2 - use of the ID in P*Stream protocols

Comment 3

[Kyle Machulis [:qdot] [:kmachulis] (INACTIVE)]

So the use of IDs and the map makes sense for not duplicating edges, and things look good afaict with the code, but do we have a way to test this easily or at all? Not quite sure if creating a loop outside of adding bogus classes to do so would be possible.

Comment 4

[Ben Kelly [:bkelly, not reviewing]]

Do you have metrics to show that this is worth it or is this purely speculative?  I agree avoiding copying is useful, but minting UUIDs for every stream sent across IPC seems a bit heavyweight.

Comment 5

[Andrea Marchesini [:baku]]

This is not needed for the PBlob refactoring. Sorry for that.