Hello, While writing some horrific code [1] related to the development server’s auto-reloading, I realized that its design is fundamentally flawed and I think that we should reconsider it.
The current implementation walks through the list of all imported Python modules, attempts to figure out which file they come from and builds a list of these files. Then it watches all these files for changes by checking if their mtime changed every second or, since 1.7, with inotify on Linux. I tried to do it with kqueue on BSD (read: OS X) but I failed. This stuff is hard. In addition to the source of imported Python modules, the autoreloader watches .mo files in order to catch translation changes and source files for modules that failed to import — it extracts them from exception tracebacks. This shows the limits of basing the list on imported Python modules. Even with such hacks, the auto-reloader will never handle all cases. Two examples: - It doesn’t survive a syntax error in the settings module. I have reasons to believe that this would be extremely messy to fix. - If a module reads a configuration file on disk at startup and caches it, the auto-reloader doesn’t detect changes to that file. So we’re stuck with cases where the development server doesn’t reload. I can’t say they’re edge cases; who never made a typo in a settings file? People who run tutorials report that it’s a very common cause of frustration for beginners. It's also wasteful. There’s little need to watch every file of every dependency in the virtualenv for changes. I think it would be much better to remove the auto-reloading logic from Django and have an external process watch the current working directory for changes. (When virtualenv is considered appropriate for people with exactly 0 days of Django experience under their belt, I think we can require a dependency for the development server’s auto-reloading.) The behavior would be much easier to describe: if you change something in your source tree, it reloads automatically; if you change something somewhere else, you must reload manually. It would also be quite easy to customize with a standard include / exclude mechanism to select which files should be watched. We aren’t in the business of writing production web servers, neither are we in the business of writing a filesystem watcher :-) Watchman [2] appears to be a robust solution. However I would prefer a pure-Python implementation that could easily be installed in a virtualenv. Does someone have experience in this area? I’m not making a concrete proposal as I haven’t studied the details. It this point I’d just like to hear what you think of the general concept. Thanks! -- Aymeric. [1] https://github.com/django/django/pull/5106 <https://github.com/django/django/pull/5106> [2] https://github.com/facebook/watchman <https://github.com/facebook/watchman> [3] http://tutorial.djangogirls.org/en/django_installation/index.html#virtual-environment <http://tutorial.djangogirls.org/en/django_installation/index.html#virtual-environment> -- 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 [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/django-developers. To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/8E08AB59-2450-40D5-A52B-C8803DA39D5F%40polytechnique.org. For more options, visit https://groups.google.com/d/optout.
