Open Bug 948641 Opened 11 years ago Updated 2 years ago

Temporary profiles should be named more descriptively (and possibly deleted on abnormal shutdown)


(Testing :: General, defect)



(Not tracked)


(Reporter: cjones, Unassigned)


My use case is running gecko test harnesses, let's say the mochitest harness for concreteness, many many times to debug a problem with an external tool.  The details of the tool aren't important, but the important part is that the problem manifests as a "hang" of the mochitest harness.  I then have to interrupt the harness with SIGINT (Ctrl-c).  When this happens, apparently the temporary profile is left behind.  There are two potential problems with this

 1. The profile is named "/tmp/tmpXXXXXX".  That's not a good name, because to delete all the old profiles, I have to run |rm -rf /tmp/tmp*|.  I expect there are other programs that create temporary files that match the pattern "/tmp/tmp*", so I never know what I'm deleting.

 2. The old profiles accumulate (obviously).  I'm notified that I need to clean up the profiles when my /tmp filesystem fill up.  That will interfere with my host system.

I certainly think (1) should be fixed, with a more descriptive name like "/tmp/gecko-mochitest-profileXXXXX" used.  That seems like it should be relatively easy too.

I would like for (2) to be changed so that the profile is always deleted.  To do so, the profile creation code would need to install the equivalent of a bash EXIT trap handler (I don't know the python-ese translation off hand) that deleted the profile.  I could certainly understand an argument to the contrary, that abnormal shutdown should leave behind the profile for later debugging, so maybe automatic always-deletion could be a command-line option.

Apologies if I filed this in the wrong place.
I'm pretty sure we used to remove the profile in a finally block, maybe that behavior got lost in the great refactoring. Naming the profiles usefully would be nice, but if we fix the cleanup bit it's not really necessary.
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.