search

kislyuk / argcomplete

Python and tab completion, better together.
https://kislyuk.github.io/argcomplete/
Apache License 2.0
1.42k stars 134 forks source link

Caching choices for choices completer #136

Open sils opened 8 years ago

sils commented 8 years ago

Hi,

we use argcomplete in coala (coala-analyzer.org) and getting possible values for one of our completers is actually something that takes about .5 seconds on decent systems. It would be nice if argcomplete would have a way to cache completions so that the experience of the user. Possibly you want the ability to mark a completion as not cached if it depends on others e.g.

kislyuk commented 8 years ago

Thanks for the suggestion. How would you envision this cache working?

sils commented 8 years ago

So this seems only relevant to the ChoicesCompleter to me, it would have an additional __init__ parameter that enables caching. When caching is enabled it is considered cheaper to read the cached choices from a cache than actually calculating them (related: https://github.com/kislyuk/argcomplete/pull/138, https://github.com/kislyuk/argcomplete/issues/137 btw.).

Assuming #138 gets merged one could hook up writing and loading to a file (name provided by https://pypi.python.org/pypi/appdirs/ ideally to be platform independent) to the choices setter or the _load_choices function (remove the lambda, put the function in a protected/private member.)

kislyuk commented 8 years ago

What would the cache key and expiration strategy be?

sils commented 8 years ago

Not sure indeed about this one. The user of the library should definitely be able to influence that behaviour because it'll be tied to his needs.

kislyuk commented 8 years ago

OK. You're not the first person who requested this :) I'm going to think about this some more. So far I'm leaning toward building caching into CompletionFinder, not Completer, and using the full command line as the key. There are quite a few design chocies to make in how the cache should behave, especially in making sure it doesn't make the completions slower in pathological cases like slow cache I/O.