152695 - cookie leak could lead to a cookie loss

Status: RESOLVED FIXED

(Core :: Networking: Cookies, defect)

Description

[Doug Turner (:dougt)]

The cookie service to be a xpcom shutdown observer.  In this way, if the cookie
service is leaked there is no chance to lose a cookie.

Comment 1

[Stephen P. Morse]

Doug, could you please try to rephrase that.  I don't understand what you are 
trying to say.

Comment 2

[Michiel van Leeuwen (email: mvl+moz@)]

Cookies now listens to profile-before-change (And writes when that occurs). Is
that enough?

Comment 3

[[e-mail]]

hmm, we don't listen to xpcom-shutdown anymore, and nsCookieService's dtor does
_not_ write the cookiefile. we're relying on the timer to write cookies after
every pageload.

dougt: is it possible now, to lose a cookie if the user starts loading a page
and quits the browser immediately? do we care about this edge case?

Comment 4

[Doug Turner (:dougt)]

i do not know if it is a problem any more.  If a user decides to quit in the
middle of a page load, i believe that page load does complete, with an error
code.  If your flushing on these events, any cookies would be flushed.

feel free to close this as invalid.

Comment 5

[[e-mail]]

oh, I meant to cc danm to ask him about this. he's the one who wrote the timer code.

danm: what do you think about comment #3?

from what dougt says, it sounds like we should be ok.

Comment 6

[Dan M]

nsCookieService once explicitly listened to the xpcom-shutdown topic but
actually only during rev 1.71. Regardless, cookies are flushed at application
shutdown because nsCookieService flushes when it sees the profile-before-change
topic, broadcast as the profile is shut down. The additional flushing that
happens on a timer after pageload is only to reduce the damage should some odd
crash take Mozilla down unexpectedly.