897474 - Create or update webrtc import procedure to avoid problems with history blowup

Status: RESOLVED WONTFIX

(Core :: WebRTC, enhancement)

Description

[Randell Jesup [:jesup] (needinfo me)]

The as-designed (in webrtc_update.sh) import mechanism can't be used, because it imports too much history into m-c if merged directly.  We've been doing it "by hand" after running the import process, and also (partly because of this, and partly from lack of knowledge) not been using it to hold patches-for-upstreaming as it was designed to.  (It imports from webrtc.org, processes any deletions or renames, deletes things not wanted, and then applies any to-be-upstreamed patches (by merging from a bookmark), leaving a repo set for merging to m-c -- which we can't do)

We need to revised this to work better, or find a way to replace it.  Especially we need to improve our ability to manage upstreamed patches.

See also bug 688178 and bug 818631 (and some others), and the webrtc_updates bug used to track bugs that have been proposed to upstream.

(This is similar to the issues surrounding all library imports, and the lack of standardization of how we deal with them - even the directory structure or ability for someone (or code) to recognize a directory is an import).

Right now I can do the merge by hand in a day or so (and it's gotten a bit easier since we removed all the peerconnection/jingle-related code).  There are exceptions where the merge part of the process (which has to be done in any case) is trickier because of widespread changes or renames in the upstream.  Hopefully the worst of these is done now that they converted to stdint types in this last update.

In the most recent update I did, I simply replaced media/webrtc/trunk/webrtc, then took an hg diff off all changes since the last base-code import (i.e. all our changes), and re-applied that as one patch, and dealt with the conflicts.  As mentioned, the stdint change made almost all code patches in this diff conflict, which made this merge a bear, but that shouldn't be normal.  It does lose hg blame of what patch a particular change came from, which is unfortunate but not horrible (it's still possible to see the patches applied to the file in the log).

Comment 1

[Randell Jesup [:jesup] (needinfo me)]

overtaken by events