Jump to content

Talk:Wikimedia technical search

Add topic
From mediawiki.org
Latest comment: 6 years ago by Quiddity (WMF) in topic "All" section not working properly

Original comments from Nike's blog

[edit]
Copied here for future reference; originally from laxstrom.name/blag

@Nemo, you should place the list of URLs for the custom search engine somewhere public, preferably in source control (github?) so people could contribute additions / tweaks.

Specifically, I’d love if the search included blog posts from Planet Wikimedia and Open Planet Wikimedia. Also, technical village pumps and perhaps template talk pages would be good places to search, also.

Finally, it seems obvious that code itself should be searchable (e.g. single-line comments that don’t make it to doxygen, or even class/function/variable names, etc.

Waldir 14:22, 11 February 2013 (UTC)Reply

@Waldir: I’ve just placed the list on the wiki: there isn’t any way to sync it automatically, is there? Planets can’t be searched, they’re ephemeral resources without archives so you’d be searching only the last few days, is this ok? Technical village pumps and template_talk namespaces are quite a challenge but I guess they may be included if someone made a list. Or you can add them yourself, I’ve added you as admin.
Nemo 17:27, 11 February 2013 (UTC)Reply
I never tried creating a custom Google search engine before, but I doubt there's an automatic way to keep it up to date, which is a shame. I've heard that Yahoo's BOSS is quite powerful, but I don't know whether it allows automatic updating either (e.g. from a text file somewhere public, a git repo, etc.). It might be interesting to try out nevertheless.
As for the planets, that's quite a shame. I wonder if we should then include the urls of the blogs included in the planet(s). The blogs are quite an important resource.
I see that code search is already included. So that leaves, from my suggestions:
  • blog posts from Planet Wikimedia (needs to be regularly updated, from here)
  • Technical village pumps
  • Template talk pages
And a new one I just thought of:
  • Signpost Technology Reports (including possible previous URL formats using the BRION nomenclature)
What else? --Waldir (talk) 21:25, 11 February 2013 (UTC)Reply
Maybe there's a way to create a mirror of the planet which is easier to update. Could be just a feedburner or Google reader or whatever, maybe: do you know some such services? Updating hundreds of entries by hand doesn't look fun, unless you think it's worth adding only a smallish subset of tech-oriented blogs. --Nemo 06:38, 12 February 2013 (UTC)Reply
I do think that a subset would make more sense. Not only most of the blogs aren't exclusively about wiki-stuff (even less so wikitech-stuff), but many of those who are aren't exclusively so (i.e. they use tags or categories). Unfortunately tags/categories can't be filtered through the URLs, so we'll have to add only blogs whose main focus is wikitech, which I believe is a reasonable subset. Then tere are those which can be filtered because they publish posts whose titles follow a pattern, such as Wikimedia blog's tecnical reports. Ther used to be a blog only for tech staff, is the domain still active? we could add it to search the older posts that were published there. In any case, it seems to be manageable. --Waldir (talk) 13:01, 12 February 2013 (UTC)Reply

Gitweb not indexable

[edit]

https://www.google.com/search?q=site:https://gerrit.wikimedia.org/r/gitweb returns a single result, with the following non-description: "A description for this result is not available because of this site's robots.txt". What purpose does that setting serve? --Waldir (talk) 21:29, 11 February 2013 (UTC)Reply

I asked on IRC. Apparently this was for performance reasons (gitweb couldn't cope with crawlers). Possibly the upcoming update to gitblit will make things better. Logs of that conversation should be available here sometime from now; timestamps approx 21:32 --> 21:42. --Waldir (talk) 21:44, 11 February 2013 (UTC)Reply
It's enough to look in ^demon's talk, where I asked the same question. ;) I had added it "just in case". --Nemo 06:38, 12 February 2013 (UTC)Reply
Indeed I re-added it on the list here, so it doesn't raise any eyebrows, but not in the actual search engine definition, by laziness. We can certainly have it tere too, as it's inconsequential. --Waldir (talk) 13:01, 12 February 2013 (UTC)Reply

Bugzilla not indexed

[edit]

See http://www.google.com/search?q=site:bugzilla.wikimedia.org -- it says HTTPS has to be used. http://www.google.com/search?q=site:https://bugzilla.wikimedia.org doesn't do any better. --Waldir (talk) 22:52, 11 February 2013 (UTC)Reply

This is weird, it used to work. If Gmane works, however, it would be a duplicate of wikibugs. --Nemo 06:38, 12 February 2013 (UTC)Reply
Yes, the wikibugs list was only added because bugzilla wouldn't. If we manage to make it work the ideal would be to remove wikibugs-l. --Waldir (talk) 13:01, 12 February 2013 (UTC)Reply

Sandbox

[edit]

Less user-friendly, but immediate for testing purposes, one can use urls such as https://www.google.com/search?q=site:mediawiki.org+OR+site:github.com/wikimedia/ --Waldir (talk) 22:52, 11 February 2013 (UTC)Reply

Duplicates

[edit]
  1. Why were duplicates added, like github.com/wikimedia/* and its subset github.com/wikimedia/*/pull/*? I think the wildcard crosses directories and everything.
  2. Gmane is not crawled, but some of its subdomains indeed are (comments.gmane.org, perhaps permalink, what else?). I'd use only Gmane.
  3. mediawiki-cvs/mediawiki-commits was added to search commit messages and code review comments, doesn't github add duplicates? --Nemo 06:38, 12 February 2013 (UTC)Reply
  1. Those aren't really duplicates. Both will be searched in the generic search (and duplicate results are, I believe, filtered by google itself), but the second one is exclusive to the "commits" category. That is, searching in that category won't return entries such as github.com/wikimedia/mediawiki-core/blob/master/*, etc.
  2. What do you mean, "only use gmane"? There's only one category (Bugs) where a specific list (wikibugs-l) was added through two archivers (gmane and mail-archive), but I did so because I'm not sure google crawls all gmane's archives or presents them in the best way (the page title, for instance, in many cases doesn't seem to be the message's title but the list's description). In this case, if we are to remove one entry, I'd probably vote to keep mail-archive, but I don't think having both is necessarily detrimental. Let's wait and see what people say.
  3. Yes, duplicates are added, but as I mentioned in the previous point, that's not necessarily a bad thing. I would, for instance, prefer clicking github links, while others might be more comfortable wiht the commit format that is sent to the mailing lists (doubtful, but you never know). Again, I think actual usage and feedback is key to make decisions here. In any case, too many results is better than too few.
--Waldir (talk) 13:01, 12 February 2013 (UTC)Reply

Page name

[edit]

Should we think about moving this page? It seems to be stable enough for announcement in mailing lists, signpost, blogs, whatever, and given the permanent nature of these, it would be better to have a better URL going in the archives. I renamed the search engine to "Mediawiki Unified Search", but we can simply call it MediaWiki Search or something like that. What do you think? --Waldir (talk) 13:01, 12 February 2013 (UTC)Reply

After you've helped with it, I agree that it shouldn't be a user subpage any longer (not that this really prevents announcements etc.). The problem is to avoid names which could be confused with search features for MediaWiki installations. Maybe replace "MediaWiki" with "MediaWiki.org" or "MediaWiki project" or something like that? --Nemo 08:51, 8 March 2013 (UTC)Reply
Hmmm... what do you think about "Wikimedia technical search"? --Waldir (talk) 02:33, 10 March 2013 (UTC)Reply
Seems a good option: MediaWiki software is a Wikimedia project, while other uses of MediaWiki are not. It's not a self-evident name and it clashes with some naming headaches, but all the proposals do. :) --Nemo 11:52, 10 March 2013 (UTC)Reply
Yeah, it won't be possible to find the perfect name, but Wikimedia technical seems reasonably short and accurate. Wikitech perhaps would work if as a community we adopt that brand (There's already wikitech-l, so that's a start), but currently Wikimedia tech seems like the lesser of several evils. In any case, having this page linked from relevant places will hopefully improve its findability. I'll perform the move and announce it :) --Waldir (talk) 14:21, 10 March 2013 (UTC)Reply
[edit]

Given the universally positive response, I've added the search to Special:Search by default via a gadget imported from en.wiki (which has been in use by default on it.wiki for many years). Note that the gadget has no localisation, as the CSE itself. No idea how to fix it. --Nemo 14:34, 22 March 2013 (UTC)Reply

yay, awesome :) --Waldir (talk) 20:18, 23 March 2013 (UTC)Reply
Reverted for now, see MediaWiki talk:Gadgets-definition.--Eloquence (talk) 19:42, 3 January 2014 (UTC)Reply

Look & feel

[edit]

Thank you very much for this very useful feature! I wonder if it could have a look & feel integrated to mediawiki.org. In a previous project, this feature was implemented quite nicely: http://talk.maemo.org/msearch.php Everybody could still see that this was a custom Google search, but it didn't feel out of context.--Qgil (talk) 16:25, 9 April 2013 (UTC)Reply

Integrated how? We have it on Special:Search. I'm quite sure that, at least with monobook, some wikis added the same dropdown for the search in sidebar, but I'm unable to find it. We could recycle wikidata:MediaWiki:Gadget-Search.js maybe, but I'd rather not play with JS myself. --Nemo 17:37, 9 April 2013 (UTC)Reply
Sorry, I mean the results page. We get a 100% google.com / 0% mediawiki.org look&feel in the search results. Search anything at http://talk.maemo.org/msearch.php and you will see the difference.--Qgil (talk) 17:57, 9 April 2013 (UTC)Reply
Ah. I doubt that's possible, except by including the results page in a frame or something here, which however is against privacy policy. Maybe there's some way to customise the result page to replicate our skin but that's way beyond my capabilities. --Nemo 18:08, 9 April 2013 (UTC)Reply
We could emulate Wikidata's new version of their search gadget.
For those who don't care about privacy, there is Extension:GoogleSiteSearch (just found it by chance). --Nemo 20:42, 14 April 2013 (UTC)Reply

"All" section not working properly

[edit]

The results from the default "all" tab don't seem to be restricted to the items listed on the main page here. Is something broken and fixable (at our end, or Google's end), or should we add a note of documentation explaining that users need to click the tabs in order to get the filtered results? Ping, Nemo bis. Thanks! Quiddity (WMF) (talk) 00:28, 3 September 2018 (UTC)Reply