Hello, I've been thinking on and off about how to improve the autoreloader implementation and I wanted to gather some feedback on potential solutions.
For some background, Django uses a fairly basic autoreload implementation that simply polls the last modified time for loaded Python files once a second. While this isn't the most efficient, it does work and has worked quite well for a long time. When running manage.py runserver, the autoreloader will launch a child "manage.py" with the same arguments and this child process actually runs Django and serves requests. To reload, the child process exits with exit code 3 and the parent restarts it. The code is some of the oldest in Django, with a fair bit of it not touched in 9-12 years. While it works (and I'm a believer in "if it isn't broke don't fix it") there are some architectural and performance issues with the current code: - Polling every second is not very efficient - Detecting when the child process has exited during startup (i.e problem in settings.py) is problematic and the code is rather nasty - i18n files are 'reloaded' when they change in a rather hacky way (resetting private attributes in another module) - There is limited support for extending the current implementation, and there are cases during development where the parent autoreloader will terminate. I don't want this email to be too long, so I'm going to summarize what I think would be a good approach to tackling these problems. 1. Refactor the current implementation by removing `pyinotify`, redundant python 2 checks and implement a 'file_changed' signal so other parts of Django can react to file changes (i.e the i18n module can handle resetting it's own state). 2. Add support for the "watchdog" library as a replacement for pyinotify. Watchdog implements file system notifications for all major platforms using a fairly simple API, so we can remove polling and have instant reloading. Also support Watchman, a notification Daemon from Facebook. 3. Add support for more advanced features, like proper handing of startup errors and socket sharing. I've got a merge request that implements all three stages as a proof of concept, but I think it's far too much a change to be done at once and should be done carefully stage by stage. One and two are fairly simple to implement, but three requires see careful consideration as to the best approach (this message is long enough already, I don't want to describe them here). Does anyone have any feedback on these ideas? Is it worth persuing even if the current implementation works ok-ish? -- You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group. To unsubscribe from this group and stop receiving emails from it, send an email to django-developers+unsubscr...@googlegroups.com. To post to this group, send email to django-developers@googlegroups.com. Visit this group at https://groups.google.com/group/django-developers. To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/CAFNZOJMT9qDk-4pKKXSJysEQCmd6CGxMZBYZs_7BQs_WbAqL6g%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.