Last modified: 2013-08-03 20:30:19 UTC
There are two separate ways to serialize, and one of them is incorrect for ranges. Search serializes the entire rows array at once, which includes all annotation contents. However, Read explicitly lists certain fields. Read has a bug. When the database has: {"ranges":[{"start":"/p[1]","startOffset":118,"end":"/p[1]","endOffset":134}],"quote":"The local legend","text":"Is the legend accurate?"} it is serialized by ApiAnnotatorRead as: {"id":14,"text":"Is the legend accurate?","quote":"The local legend","ranges":{"start":"/p[1]","startOffset":118,"end":"/p[1]","endOffset":134}} Note how the array of ranges is stripped of its brackets. It would be good to filter the annotation (only allowing valid fields). However, I don't think that's a blocker, since FormatJson can only serialize JSON, not arbitrary objects. So to simply serialize the entire object in Read, you can loop like: https://git.wikimedia.org/blob/mediawiki%2Fextensions%2FEventLogging.git/3985fbf7d20c9c2ef47fb561b418a62618ff1177/includes%2FApiJsonSchema.php#L112 You still need to explicitly add the ID, but that will handle the ranges correctly, and everything else.
This is also necessary to handle multiple ranges in a single annotation, which is allowed; that's why it's an array.
The best way to avoid Search and Read being inconsistent is to use a base class or helper class to process the annotations (adding id, and probably user, from column data into the JSON).
I recommend you use a simple version of the repository pattern for this. I'll comment further at https://gerrit.wikimedia.org/r/#/c/75106/
Change 75106 had a related patch set uploaded by Mattflaschen: Added the update class https://gerrit.wikimedia.org/r/75106
Change 75106 merged by jenkins-bot: Added the update class https://gerrit.wikimedia.org/r/75106
