1647107 - ClearPoisonIOInterposer is never called

Status: NEW

(Core :: XPCOM, task)

Description

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

Recently in bug 1529610 (not landed yet as I write this) I added a static object in PoisonIOInterposerWin.cpp, with associated clean-up code in ClearPoisonIOInterposer(), which I expected to be called at some point before shutdown.

However ClearPoisonIOInterposer() is never called. This caused some confusing test failures with my new code, before I noticed the missing calls!

Its existence implies that things created at/after InitPoisonIOInterposer() will safely be removed. so I think that either:

• ClearPoisonIOInterposer() should actually be called to match InitPoisonIOInterposer() calls; however since InitPoisonIOInterposer() may be called from multiple places, we would need to implement a kind of reference counting to make sure it's only called at the last possible moment.
• Or ClearPoisonIOInterposer() should just be deleted, to remove the expectation that clean-up happens.

(In the meantime, in bug 1529610 I will add some comments around ClearPoisonIOInterposer(), to notify future users that it's not used yet.)

Comment 1

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

Oh, I've only just noticed there's a MOZ_ASSERT(false) at the top of the Windows implementation! I cannot see an explanation for it though. This was added in bug 902587 in 2013.

Aaron, you last worked on it in 2016(!), any ideas/suggestions?

Comment 2

[(No longer employed by Mozilla) Aaron Klotz]

Ugh, as you can see it has been a loooong time since that code was worked on.

My best guess is that there were issues with cleanly tearing down the PoisonIOInterposer and we hadn't made a final decision as to how (or even whether) to do it. Unfortunately both Jonas and I got pulled onto different projects by the end of 2013 and this stuff never really made it to 100% completion.

I would say that, at this point, it's probably your call whether or not it is worth implementing ClearPoisonIOInterposer. I'm happy to review it, of course, but at this point I don't have a good enough recollection of anything to have strong opinions about it.