92908 - prefs inconsistency causes cache session creation to fail

Status: RESOLVED FIXED

(Core :: Networking: Cache, defect, P3)

Description

[cls]

This is a spinoff of bug 92099 in which I reported that beos had problems making
ftp connections.  With bbaetz's help, we determined that problem is with the
checks being made against the cache prefs.

This failure is caused by 3 distinct conditions:

1) both disk & mem cache are disabled for this profile
2) nsCacheService::CreateSession() errors out if disk cache & mem cache are disabled
3) nsFtpProtocolHandler::NewChannel() only checks for the pref
browser.cache.enable (which always seems to be on) and exits silently if the
call to nsCacheService::CreateSession() fails.

Also, libpr0n suffers from the same problem except that it will spit out a ton
of warnings about the cache session failure in a debug build.

Comment 1

[gordon]

Hmmm.  Originally the enable cache prefs were only read at startup, so having
CreateSession() fail if no cache devices were available made sense.  If we want
to support a more dynamic response to enabling/disabling the cache devices, it
would make sense to go ahead and return a nsCacheSession even if nothing could
be stored according to the specified storage policy due to a lack of enabled
devices (they might get enabled later).

In any case, cache clients should not check the "browser.enable.cache" pref
because it is ambiguous and should be discontinued.  I expect they should also
handle failure of CreateSession() in a reasonable way.

Comment 2

[Jaime Rodriguez, Jr.]

Gordon/Gagan - Is this worthy of a nsbranch+ nomination (e.g. No keywords).

Pls help . . . we are trying to get a good picture of what work is left for the
next release.

Comment 3

[gordon]

I wouldn't hold up 0.9.4 for this one.

Comment 4

[gordon]

Retargeting for 0.9.6

Comment 5

[gordon]

Fixed with checkin for bug 101857. Marking FIXED.