1353659 - Initial-content code for cross-channel localization work

Status: RESOLVED FIXED

(Developer Services :: General, task)

Description

[Axel Hecht [:Pike]]

One module for the x-channel work is the code that back-fills version control data for files that get newly added by configuration changes.

I've written more about the details in the patch I'll attach in bug 1353658.

Right now, I'm replaying lines of blame in a linearized history, which is good for some files, but doesn't reliably replay blame.

I've tried replaying patches, which led to even weirder patches, as I end up removing un-used code in unrelated changesets.

Looking at the full version history of the current files led to a rather complicated graph. It's a graph good to study in the light of incremental updates, but maybe we can do something more straight-forward here.

Comment 1

[Axel Hecht [:Pike]]

Attachment: bug 1353659, initial content for a cross-channel l10n file based on blame,

The generation of a file trying to replay hg blame is just about the text,
and doesn't care about the specific l10n file format.
Thus there are no dependencies on parsers for particular files in this
code and its tests.

Review commit: https://reviewboard.mozilla.org/r/135514/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/135514/

Comment 2

[Gregory Szorc [:gps]]

Comment on attachment 8863733 [details]
bug 1353659, initial content for a cross-channel l10n file based on blame,

Cancelling review per Vidyo conversation. Will wait for rebase and consolidation of all the x-channel patches.

Comment 3

[Axel Hecht [:Pike]]

I've been going back and forth over the various algorithms, and updated the user-story to the one that seems to generate good data and is least weird.

One thing that I still want to do is improve the P1 on merges. In particular the merge of im into c-c seems to have picked the wrong one, creating blame that's rather off. (They had dumps in c-c before they merged in the history of the external chat repo, yay).

The resulting repo is just 14M.

Also still to do is to convert the configuration files in the target repo.

Comment 4

[Axel Hecht [:Pike]]

Update, I've got a rambling repository at https://hg.mozilla.org/users/axel_mozilla.com/cross-channel-experimental/ . It's like the worst github commit rambling and then some. I won't be requesting review on that, no worries.

The most interesting part of that is that it has a growing test suite, https://hg.mozilla.org/users/axel_mozilla.com/cross-channel-experimental/file/tip/tests.

Also, it creates good target repositories, and for the contents of that repository, I think we're down to technical questions around the commit meta data, which is bug 1397736.

I'd like to iron that out before we approach landing the code that generates it, mostly because the merge tactics around 57 really want me to use cross-channel for 57, so I'm splitting things into "blocks 14th", "blocks 20th", etc.

My current idea to actually get the code reviewed is to dump the end of the rambling repo into vct, and then remove one feature at a time, removing/fixing tests as that goes. Then replay that patch queue in reverse order, and request review on those patches.

Also, we're working on the actual content generation in parallel to the repo magic, the content pieces in bug 1384570 are now developed by stas.

Comment 5

[Axel Hecht [:Pike]]

I'm having second thoughts on the vct approach.

I'm working on building a story to land in m-c, so closing this one out. Also, w/in the scope of this bug, this has long been fixed.