State and caching in completion tables

[Spencer Baugh <sbaugh@janestreet.com>]

If I have a completion table which wants to store internal state - in my
case, for caching data used to generate completions - is there a good
place to do that?

Ideally, the state (in my case, a cache) would only be preserved for a
given call to completing-read.  If a new call to completing-read starts,
I'd like the state to be reset.

Usually, I would do this by having a function which creates a fresh
completion table with fresh state, and call that each time to generate
the completion table to pass to completing-read.  So state would
naturally be preserved only for a given completing-read, because each
completing-read would use a new completion table.

But more specifically, I'm adding a new file-name-handler.  As a
file-name-handler, I don't get the chance to return a fresh completion
table for each read-file-name invocation; I only get to handle
file-name-all-completions and file-name-completion.  Given that, I don't
know where my file-name-all-completions and file-name-completion
functions could store some state which is specific to a read-file-name
call.

I want to store state for a read-file-name call to avoid making
duplicate network requests each time file-name-all-completions and
file-name-completion are called; instead I'd like to store the result I
got the first time.  I'm happy with any way of achieving that.

Is there a natural place for a completion table to store state which is
specific to a given invocation of completing-read?

Or failing that more general solution, a way for a file-name-handler to
save state for a given read-file-name call?