Re^5: rmtree() is failing after some time

[grinder]

I am calling rmtree() in DESTROY() method of a class

The error is due to the fact that the code is unable to stat the directory in question. Is the path canonical or relative? If it's relative, is your code's working directory where you think it is? Can you prove it (print getcwd)?

How are you generating your temp names? Could two threads create the same name?

In general, putting resource teardown code beyond closing a socket or the like in a DESTROY handler is a bit weird. I think you'll need to solve this with another a level of indirection. You have to go up a level and ask some parent code to clean up afterwards.

[gwadej]

In general, putting resource teardown code beyond closing a socket or the like in a DESTROY handler is a bit weird.

I have to disagree. If I have resource creation in new or other constructor, I often put cleanup of that resource in DESTROY. If the resource is managed by the object, shouldn't the object do the cleanup?

I guess my C++ background shows here. I see putting the cleanup code in DESTROY as serving to hide those implementation details from the caller of my object. It also tends to reduce the need for cleanup code to be repeated everywhere the object is used (DRY principle).