Bugzilla

(In reply to ISHIKAWA, Chiaki from comment #13)
> That requires the proper ordering of subsystem shutdown, which has plagued TB and FF for many years.
> 
> Also, at least in one other case, I remember that TB code is written in such a manner that destruction of initiated by
> GC during shutdown (that is the destruction timing is NOT programmer control at all), and in one such case, by the time, initiating shutdown triggers the GC, and eventually this destruction of certain object is done, but the function called from the destructor of the object is no longer callable because the XPCOM has shutdown. 
>
> If I recall correctly, it is about the deletion of a temporary file which may contain a piece of mail message (private information), but TB code simply discards the object that refers to the file, and let its automatic destruction by GC calls the destructor automatically, and the destructor calls the file system removal API  after obtaining the file path name from the stored file object (can't recall which class). 
>
> Unfortunately, mapping of the object to  the file system pathname fails because the service to provide the mapping has already be shutdown. 
> Thus the temporary file is left around. Tough.

I think the bug I was referring to is
Bug 1595343 C-C TB: nsMsgAttachment::DeleteAttachment fails to delete attachment file occasionally during shutdown, possibily leaving orphaned files in temp/tmp directory