migrateComments.php fails on numerical comment
Closed, ResolvedPublicBUG REPORT
Actions

Description

Migrating a wiki from 1.31.6 to 1.35.0 failed for me with this error from the database log (PostgreSQL):

ERROR:  operator does not exist: text = integer at character 1055
HINT:  No operator matches the given name and argument types. You might need to add explicit type casts.
STATEMENT:  SELECT /* MigrateComments::loadCommentIDs sysadmin@wiki-p... */  comment_id,comment_text  FROM "comment"    WHERE ((comment_hash = 663705717 AND comment_text = '/* TOPs */') OR (comment_hash = -29472803 AND comment_text = '/* Themenspeicher */') ...

The problematic part around character 1055 was

OR (comment_hash = -1801163141 AND comment_text = 2020)

i.e. the value "2020" was not correctly quoted, it should have been

OR (comment_hash = -1801163141 AND comment_text = '2020')

Table "recentchanges" does indeed contain an entry "rc_comment" of just the value "2020".

I assume (without further proof, I don't speak PHP), that some code generating the SQL statement erroneously decides not to quote "2020", because it does not look like a string but a numeral.

Details

	Project	Branch	Lines +/-	Subject
	mediawiki/core	master	+1 -0	migrateComments: Cast array keys back to string before passing to the DB
	mediawiki/core	REL1_35	+1 -0	migrateComments: Cast array keys back to string before passing to the DB

Customize query in gerrit

Event Timeline

joernc_unibi created this task.Nov 27 2020, 12:46 PM

Restricted Application added a subscriber: Aklapper. · View Herald TranscriptNov 27 2020, 12:46 PM

Aklapper added a project: MediaWiki-Maintenance-system.Nov 27 2020, 1:06 PM

Actually the problem is not in table "recentchanges", but in "revision". A quick and dirty fix is to change "2020" to "2020x". After changing the entry in "recentchanges", the upgrade still failed. I only then saw, that "revision" contained the same entries. After fixing that entry, the upgrade succeeded.

Reedy added projects: Platform Engineering, MW-1.35-release.Nov 29 2020, 12:28 AM

Reedy added projects: MediaWiki-Comment-store, PostgreSQL.

• Clarakosi triaged this task as Medium priority.Dec 1 2020, 9:06 PM

• Clarakosi edited projects, added Platform Team Workboards (Clinic Duty Team); removed Platform Engineering.

• Clarakosi moved this task from Inbox to Later on the Platform Team Workboards (Clinic Duty Team) board.

The problem is that putting a numeric string into an array key causes it to be converted to an integer:

php > $a = [ '2020' => 1, '2020x' => 1, '000' => 1 ];
php > foreach ( $a as $k => $v ) { var_dump($k); }
int(2020)
string(5) "2020x"
string(3) "000"

The solution is to cast it back to string. No information is lost because ZEND_HANDLE_NUMERIC() goes to some trouble to avoid casting non-canonical numeric representations to integer.

Change 644634 had a related patch set uploaded (by Tim Starling; owner: Tim Starling):
[mediawiki/core@master] migrateComments: Cast array keys back to string before passing to the DB

https://gerrit.wikimedia.org/r/644634

gerritbot added a project: Patch-For-Review.Dec 1 2020, 11:38 PM

Change 644634 merged by jenkins-bot:
[mediawiki/core@master] migrateComments: Cast array keys back to string before passing to the DB

https://gerrit.wikimedia.org/r/644634

Maintenance_bot removed a project: Patch-For-Review.Dec 2 2020, 1:10 AM

Change 644647 had a related patch set uploaded (by Reedy; owner: Tim Starling):
[mediawiki/core@REL1_35] migrateComments: Cast array keys back to string before passing to the DB

https://gerrit.wikimedia.org/r/644647

gerritbot added a project: Patch-For-Review.Dec 2 2020, 1:14 AM

Change 644647 merged by jenkins-bot:
[mediawiki/core@REL1_35] migrateComments: Cast array keys back to string before passing to the DB

https://gerrit.wikimedia.org/r/644647

Reedy closed this task as Resolved.Dec 2 2020, 1:29 AM

Reedy assigned this task to tstarling.

Reedy removed a project: Patch-For-Review.

ReleaseTaggerBot added projects: MW-1.36-notes (1.36.0-wmf.21; 2020-12-08), MW-1.35-notes.Dec 2 2020, 2:00 AM