On Tue, Apr 5, 2016 at 2:08 PM, Rustom Mody <rustompm...@gmail.com> wrote: >> 'import tk.ttk' looks for 'tk' in sys.modules, does not find it, looks >> for a module named 'tk' on disk, does not find it, and says so. > > A well-known quote comes to mind: > > | There are only two hard things in Computer Science: cache invalidation and > | naming things. > > eg. http://martinfowler.com/bliki/TwoHardThings.html > > particularly since this seems to be in both categories :-)
sys.modules isn't really a cache in that sense, though. The "hard problem" of cache invalidation comes from the fundamental assumption that a cache hit should be semantically identical to a cache miss; Python's import system has fundamentally different semantics, specifically that modules can exist independently of anything on the disk, and modules are guaranteed to be singletons - which means that "import decimal; decimal.getcontext().prec = 50" will affect anyone else who uses the decimal module, because there can't be a duplicate. (The one exception to this, where __main__ gets reimported under another name, causes significant confusion despite being extremely uncommon.) If you like, you can look at sys.modules as a mandatory cache that never records negatives and always records positives. Once an import has succeeded, it will always be resolved from the cache, until program termination; failed imports will always be retried. Python dodges the cache invalidation problem by never invalidating anything :) ChrisA -- https://mail.python.org/mailman/listinfo/python-list
