The proposal was approved by TechCom, the code has been merged, and it's
live now on the Beta Cluster. I'm running the maintenance script now.
Please test things there and report any bugs you encounter, either by
replying to this message or by filing it in Phabricator and adding me as a
subscriber. Assuming no major errors turn up that can't be quickly fixed,
I'll probably start running the maintenance script on the production wikis
the week of December 11 (and perhaps on
and testwiki the week
before).
If you're curious as to what the history of an existing imported page might
look like after the maintenance script is run, see
for an example.
On Tue, Oct 31, 2017 at 10:52 AM, Brad Jorsch (Anomie) <
bjorsch(a)wikimedia.org> wrote:
Handling of usernames in imported edits in MediaWiki
has long been weird
(T9240[1] was filed in 2006!).
If the local user doesn't exist, we get a strange row in the revision
table where rev_user_text refers to a valid name while rev_user is 0 which
typically indicates an IP edit. Someone can later create the name, but
rev_user remains 0, so depending on which field a tool looks at the
revision may or may not be considered to actually belong to the
newly-created user.
If the local user does exist when the import is done, the edit is
attributed to that user regardless of whether it's actually the same user.
See T179246[2] for an example where imported edits got attributed to the
wrong account in pre-SUL times.
In Gerrit change 386625[3] I propose to change that.
- If revisions are imported using the "Upload XML data" method, it
will be required to fill in a new field to indicate the source of the
edits, which is intended to be interpreted as an interwiki prefix.
- If revisions are imported using the."Import from another wiki"
method, the specified source wiki will be used as the source.
- During the import, any usernames that don't exist locally (and can't
be auto-created via CentralAuth[4]) will be imported as an
otherwise-invalid name, e.g. an edit by User:Example from source 'en' would
be imported as "en>Example".[5]
- There will be a checkbox on Special:Import to specify whether the
same should be done for usernames that do exist locally (or can be created)
or whether those edits should be attributed to the existing/autocreated
local user.
- On history pages, log pages, and the like, these usernames will be
displayed as interwiki links, much as might be generated by wikitext like "
[[:en:User:Example|en>Example]]". No parenthesized 'tool' links
(talk,
block, and so on) will be generated for these rows.
- On WMF wikis, we'll run a maintenance script to clean up the
existing rows with valid usernames and rev_user = 0. The current plan there
is to attribute these edits to existing SUL users where possible and to
prefix them with a generic prefix otherwise, but we could as easily prefix
them all.
- Unfortunately it's impossible to retroactively determine the
actual source of old imports automatically or to automatically do anything
about imports that were misattributed to a different local user in pre-SUL
times (e.g. T179246[2]).
- The same will be done for CentralAuth's global suppression
blocks. In this case, on WMF wikis we can safely point them all at Meta.
If you have comments on this proposal, please reply here or on
https://gerrit.wikimedia.org/r/#/c/386625/.
Background: The upcoming actor table changes[6] require some change to the
handling of these imported names because we can't have separate attribution
to "Example as a non-registered user" and "Example as a registered
user"
with the new schema. The options we've identified are:
1. This proposal, or something much like it.
2. All the existing rows with rev_user = 0 would have to be attributed
to the existing local user (if any), and in the future when a new user is
created any existing edits attributed to that name will be automatically
attributed to that new account.
3. All the existing rows with rev_user = 0 and an existing local user
would have to be re-attributed to different *valid* usernames,
probably randomly-generated in some manner, and in the future when a new
user is created any existing edits for that name would have to be similarly
re-attributed.
4. Like #2, except the creation (including SUL auto-creation) of the
same-named account would not be allowed. Thus, an import before the local
name exists would forever block that name from being used for an actual
local account.
5. Some less consistent combination of the "all the existing rows" and
"when a new user is created" options from #2–4.
Of these options, this proposal seems like the best one.
[1]:
https://phabricator.wikimedia.org/T9240
[2]:
https://phabricator.wikimedia.org/T179246
[3]:
https://gerrit.wikimedia.org/r/#/c/386625/
[4]:
https://phabricator.wikimedia.org/T111605
[5]: ">" was chosen rather than the more typical ":" because the
former is
already invalid in all usernames (and page titles). While a colon is *now*
disallowed in new usernames, existing names created before that restriction
was added can continue to be used (and there are over 12000 such usernames
in WMF's SUL) and we decided it'd be better not to suddenly break them.
[6]:
https://phabricator.wikimedia.org/T167246
--
Brad Jorsch (Anomie)
Senior Software Engineer
Wikimedia Foundation