714286 - Cycle collection log should be written to a place where the process can write to

Status: VERIFIED FIXED

(Core :: XPCOM, defect)

Description

[(no longer active)]

Also, it's a good idea to print the full path of the log file to the error console so that the user wouldn't need to search their entire file system looking for the log file.

I have a patch for this.

Comment 1

[(no longer active)]

Attachment: Patch (v1)

Comment 2

[Andrew McCreight [:mccr8]]

Could you add some kind of #ifdef or something that lets you easily revert to the old behavior of the file name generation with a recompile? I think this is an improvement for people doing a one-off CC dump, but when you are dumping every CC, and doing multiple runs, it is helpful to have them in a predictable location.  The console logging is fine either way.

Comment 3

[(no longer active)]

I'm not sure what you mean.  The location is predictable, it's the temp directory.  Arguably it's more predictable than before, since at least on Windows the current directory changes when you invoke things like the file picker dialog.  :-)

Comment 4

[Andrew McCreight [:mccr8]]

The directory part sounds reasonable.  How does tmpnam affect the file name itself?

Comment 5

[(no longer active)]

tmpnam() returns the full path name to a temp file, something like "/tmp/mygarbage.tmp".  I can walk the string back from the end to find the first directory separator, and then append cc-edges to that, so that we would end up with "/tmp/cc-edges.N.MMMM.log" instead of "/tmp/mygarbage.tmp-cc-edges.N.MMMM.log".  Is that what you want?

Comment 6

[(no longer active)]

Attachment: Patch (v2)

Implements comment 5.

Comment 7

[Andrew McCreight [:mccr8]]

Comment on attachment 585044 [details] [diff] [review]
Patch (v2)

Review of attachment 585044 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks, that's better than the current behavior for my purposes!  Because you can end up with the logs being dumped in some weird subdirectory if you invoke Firefox using Mochitest.

Comment 8

[(no longer active)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/c2c6616eea53

Comment 9

[Olli Pettay [:smaug][[e-mail]]]

Could you update https://wiki.mozilla.org/Performance:Leak_Tools#Cycle_collector_heap_dump

Comment 10

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/mozilla-central/rev/c2c6616eea53

Comment 11

[Nickolay_Ponomarev]

(In reply to Olli Pettay [:smaug] from comment #9)
> Could you update
> https://wiki.mozilla.org/Performance:Leak_Tools#Cycle_collector_heap_dump
https://wiki.mozilla.org/index.php?title=Performance%3ALeak_Tools&action=historysubmit&diff=382891&oldid=382832

Comment 12

[(no longer active)]

tmpnam() is broken on Windows. :(  I have a patch for this.

Comment 13

[(no longer active)]

Attachment: Windows fix followup

Comment 14

[Andrew McCreight [:mccr8]]

Comment on attachment 585589 [details] [diff] [review]
Windows fix followup

Review of attachment 585589 [details] [diff] [review]:
-----------------------------------------------------------------

Looking at MS's documentation, GetTempPath returns a path that ends in \, whereas I think the other code path removes the trailing directory separator.  On Windows, it looks like the path will end up being something like C:\temp\\cc-edges.txt .  Is that a problem?

Comment 15

[(no longer active)]

No.  (I actually tested the patch on Windows this time.  ;-))

Comment 16

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/06de54d7d7b4

Comment 17

[Jim Jeffery not reading bug-mail 1/2/11]

Verified Fixed using latest hourly m-c win32 build on Win7 x64 
cset: https://hg.mozilla.org/mozilla-central/rev/0eec6ba6a87a 

today's m-i -> m-c merge