backend_pgf: improve handling of temporary directories #1286
Conversation
|
@cgohlke Does this fix the problem on Windows? |
|
Hmpf, sometimes pyhton can be a real pain. There is no way of designing a reliable destructor. Using atexit now for cleaning up the remaining temporary directories. I also removed the use of chdir when working with temp dirs, I think @cgohlke had some concern regarding multithreading. |
|
I'm officially going insane. This new handling of temporary directories should be perfectly fine, but the situation on Windows just got a lot worse on my VM test system. I found an application called "ProcessExplorer" that can find and show the open file handles for each running process. The problem on windows is that a spawned child process automatically inherits all open file descriptors and keeps them open as long as the subprocess is running. There is an option called This is a really bad limitation, almost looks like a "won't fix". I don't even dare to imagine what this limitation means for multithreading support on Windows. |
|
Ok, this is the best I can do for now :). Unfortunately on Windows the only real solution is not to inherit fds when spawning a subprocess and then to develop another way of inter-process communication. I don't see how this can be done with stdin/stdout on Windows... |
|
@pwuertz Are you happy with this pull request? I'm not a Windows guy, but this is milestoned for 1.2.x. Should it be merged? |
|
@dmcdougall I'm not really happy, but I don't see another solution right now. I'm not a Windows guy either, pretty demoralized by the limitations I encountered and hoped for some feedback that magically solves the problem. Let me just add some warning in the documentation and then this PR is good to go. |
|
Ok, forwarding the discussion / solution to #1324. This PR should be good for 1.2 now. |
backend_pgf: improve handling of temporary directories
Addresses #1281 where deleting temporary directories fails on Windows. Also, for multithread safety, stop using os.chdir.