498307 - Various mailnews tests leak nsSocketTransport, nsStringBuffer, nsThread and others (fakeserver related?)

Status: RESOLVED FIXED

(MailNews Core :: Backend, defect)

Description

[Mark Banner (:standard8)]

After part 2 of bug 438922 has landed to clean up the leak logs, there will be
remaining some leaks that are seen across mailnews, I think just when using fakeserver in tests (but not always). Typical log entry:

== BloatView: ALL (cumulative) LEAK STATISTICS

     |<----------------Class--------------->|<-----Bytes------>|<----------------Objects---------------->|<--------------References-------------->|
                                              Per-Inst   Leaked    Total      Rem      Mean       StdDev     Total      Rem      Mean       StdDev
   0 TOTAL                                          26     2380    13429        8 (  114.84 +/-   125.76)    47844       12 (  128.15 +/-   150.36)
  45 nsAStreamCopier                                60       60        4        1 (    2.29 +/-     1.11)      403        1 (    8.38 +/-     1.70)
 176 nsPipe                                        152      152        4        1 (    2.29 +/-     1.11)      260        2 (   23.04 +/-     7.66)
 191 nsRecyclingAllocatorImpl                       48       48        1        1 (    1.00 +/-     0.00)        8        1 (    2.00 +/-     0.93)
 205 nsSocketTransport                             296      296        2        1 (    1.33 +/-     0.58)       98        2 (   12.01 +/-     4.34)
 206 nsSocketTransportService                     1684     1684        1        1 (    1.00 +/-     0.00)      279        3 (   12.12 +/-     2.44)
 212 nsStringBuffer                                  8        8     3198        1 (  200.10 +/-   161.50)     3580        1 (  214.91 +/-   165.30)
 222 nsThread                                       72       72        3        1 (    1.80 +/-     0.84)     1580        1 (  353.47 +/-   198.20)
 224 nsTimerImpl                                    60       60        7        1 (    3.77 +/-     1.96)       44        1 (    8.21 +/-     3.63)

This bug will cover all the leaks of this style in whichever mailnews test file unless we discover that it is an issue in the structure of the test files, at which point we can revisit this decision.

Comment 2

[Mark Banner (:standard8)]

Attachment: The fix

Turns out this is quite simple - we're not closing the output stream object that's attached to the transport - hence it keeps the transport open. Adding a close when we close the transport works nicely.

This cleans up most of the fake server tests. Although this test does make it obvious that we have an intermittent leak of a nsBaseAppShell and a nsRunnable in the mailnews/imap tests which I believe is due to how we are shutting the imap run thread down, so I'll file a separate bug for those.

Comment 3

[Serge Gautherie (:sgautherie)]

(In reply to comment #2)
> obvious that we have an intermittent leak of a nsBaseAppShell and a nsRunnable

MacOSX bug 472773?!

Comment 4

[Serge Gautherie (:sgautherie)]

(In reply to comment #3)
> MacOSX bug 472773?!

Obviously not, as boxes report it on Linux and Windows :->

Comment 5

[Mark Banner (:standard8)]

This bug is not a regression - afaik it would have happened ever since we started the fake servers, we just had no way of finding it.

Comment 6

[Mark Banner (:standard8)]

Filed bug 499644 on the nsRunnable/nsBaseAppShell leak.

Checked in patch: http://hg.mozilla.org/comm-central/rev/b546a5072e6f