590776 - Make threadlocal Prefixer die with the request

Status: VERIFIED FIXED

(support.mozilla.org :: Code Quality, task, P2)

Description

[Erik Rose [:erik][:erikrose]]

It currently hangs around past the scope of the current request and makes our test success dependent on the order of their running. Our current need to "prime" the Client class by making some arbitrary request before testing is a symptom of this.

Comment 1

[Erik Rose [:erik][:erikrose]]

To clarify, the threadlocal is currently set on the way into the request by a piece of middleware (LocaleURLMiddleware), so symmetry demands that it be unset on the way out. This will expose any incorrect tests with loud, immediate, reliable failures so we can fix them. Said fix is also in scope for this bug.

Comment 2

[James Socol [:jsocol, :james]]

We need to "prime" it because otherwise it is not set at all, not set but incorrect.

Comment 3

[Erik Rose [:erik][:erikrose]]

This bug does *not* fix the random IntegrityError failures: I get one intermittently and another reliably after rebasing this branch. Grrr.

Comment 4

[Erik Rose [:erik][:erikrose]]

Fixed in http://github.com/jsocol/kitsune/commit/c7da7d9728837c09da7b11981640185a8ba7781a

QA: this is mostly a change to our automated tests, so the other devs will be constantly QAing this. However, if you were to see any bad URLs generated (perhaps missing locales or having them where they shouldn't), this should be a suspect.

Comment 5

[Paul Craciunoiu [:paulc]]

I would just mark this [qa-] and have QA file bugs as they come up (and related ones would be regressions).

Comment 6

[Paul Craciunoiu [:paulc]]

Also landed on 2.2.x branch:
http://github.com/jsocol/kitsune/tree/2.2.x

Comment 7

[Rebecca Billings [:rbillings]]

Closed as [qa-]