Extension talk:LocalisationUpdate/LUv2

From MediaWiki.org
Jump to navigation Jump to search

Getting feedback[edit]

I suggest to ask all our committed users we know of: [1]. You're interested in the wikis without "wmf" in their MediaWiki version name. --Nemo 12:24, 13 March 2014 (UTC)

Thanks for the link Nemo! --kondi 14:03, 13 March 2014 (UTC)

Hashes[edit]

Small note, in "a list of message_id->hash pairs" the hashes may end up being longer than the content, take into account that most messages are very short. Is this a problem? --Nemo 16:39, 13 March 2014 (UTC)

I looked up some of the more complete translation files from mediawiki core, and most of the messages are indeed very short. It'd make more sense to use the messages as is, no hashing. --kondi 10:09, 21 March 2014 (UTC)
My question is superseded by the more detailed one by Nikerabbit in #API. He proposed to hash whole files instead. --Nemo 13:55, 3 April 2014 (UTC)

Some feedback[edit]

Kondi, I'm sorry I was so irritated earlier in IRC. Ordinarily I'm happy to give early feedback on a draft proposal, *if* I know what (specifically) the author wants feedback on. (The basic concept of the project? Prose style? Implementation plans?) But since you don't yet have a schedule and you don't yet provide any evidence that you can do this project (in three months, or at all), there's not much I can say beyond "work on your capitalization" and "when you make assertions, back them up". Sharihareswara (WMF) (talk) 15:01, 20 March 2014 (UTC)

I should've clarified that it wasn't complete before asking. Anyway I've added most of the required details now. -- kondi 12:02, 21 March 2014 (UTC)

Sandbox[edit]

Thanks for drafting a timeline. What's the sandbox you have in mind? Do you plan to set up a test instance of translatewiki.net? Or would that depend on you getting shell access on its server (which is not trivial)? --Nemo 07:09, 21 March 2014 (UTC)

There'll be two standalone servers (even a single server would suffice) in which the current scenario will be replicated. One would host mediawiki/jquery.uls/rails-port applications, the other will host the LUv2 service. Shell access to the translatewiki.net servers is not required. One option for updating the service database is to get the updates directly from translatewiki.net database (instead of depending on git hooks). For this I'll need read access to the translatewiki.net database (primarily the translate_tms table). --kondi 09:53, 21 March 2014 (UTC)
Quick correction, you do not want the translate_tms table, it's for translation memory. For that we use Solr (soon ElasticSearch) so that table is not in use. Git hooks (via GitHub or Gerrit?) on repositories sounds okay, but we could also think whether we can push information (not contents) about exports to your service directly form translatewiki.net. --Nikerabbit (talk) 16:11, 21 March 2014 (UTC)
Thanks for the correction. It'd be good if we can manage to send such information from twn, it'll make the system more independent (i.e. no need to configure hooks). -- kondi 10:47, 22 March 2014 (UTC)

Microtasks[edit]

Hi, at Google Summer of Code 2014 you say you have complete microtasks. However, there is no link in the table, and I could not find any link to them in your proposal either. Please fix this, so we can review the tasks you have completed. Thank you.--Qgil (talk) 19:03, 31 March 2014 (UTC)

Ah, I added a link before seeing this. Still, the parsoid bug you mentioned and the other core bug you chose are not the areas where you're most likely to interact with the primary mentor for this project i.e. Kartik. You could choose a bug in his team's extensions or one of the easy ones or one in LU. --Nemo 19:53, 31 March 2014 (UTC)
Thanks Nemo for adding the link. Quim, I'm working on couple of bugs at the moment. Will link them when done. --kondi 17:07, 1 April 2014 (UTC)

API[edit]

Hi. Would you be able to add more detail to the API section? I'd like to see some thoughts on how to make it as efficient as possible. Have you for example considered a timestamp based approach over sending all messages to the server? Or sha1s of the files? We could easily map the file checksums to timestamps by cross-referencing with git repositories, and only ask to send the messages as fallback. Also, some example communications between the client and server would be useful. --Nikerabbit (talk) 13:38, 3 April 2014 (UTC)

I have updated the proposal page with some details. I'll add the sample communication very soon. --kondi 06:56, 5 April 2014 (UTC)

Repository[edit]

Hello, the updates page (which I just found, linked from another page) mentions some code being written, but I don't find any gerrit commit by you. The proposal text says you'd use gerrit or another public git repo, can you please link the repo you're using? Thanks, Nemo 12:19, 9 July 2014 (UTC) P.s.: I now found [2] but it's empty, I'm probably looking in the wrong place.