Fun with loading extensions

[Greg Ward]

I think I have just figured out a very tricky and subtle detail of
loading extensions.  It boils down to a simple rule: if extension A
wraps localrepo.X() and extension B subclasses repo.__class__ and
overrides X(), then you *MUST* load B before A.  If you do not, B's
override of X() will not be called.

Here's the catch: the authors of extensions A and B can't do a damn
thing about this.  Only the end user who is configuring his extensions
is able to do the right thing, by listing them in the right order in
his hgrc.  But the end user doesn't have a clue how A or B work.  And
either A or B could change their implementation from one version to
the next.

Hmmmm.  Some possible ways to mitigate this problem:

  * advise extension authors against the subclassing trick
  * demand that extension authors document if their extension must be
loaded early because it uses the subclassing trick
  * advise end users to read the source for all the extensions they
are interested in before configuring them

...better ideas?

Greg

(Why yes, I am one of those extension authors who has recently
converted to using the subclassing trick because it avoided another
conflict.  Aieee!)