yas--all-parents can fail to find all parents if called before snippet directory is loaded

[nverno]

Relating to ae36504

Sometimes yas--all-parents attempts to get a mode's parents before the
snippet directory containing the parent has been loaded by yas-load-directory.
When this happens, yas--parents doesnt yet contain entries for the parents
who havent been loaded, so yas--all-parents doesnt recurse on the parent,
thus failing to add all the parents to the yas--all-parents prop.

This sequence of events can happen, for example, if yas-minor-mode is called
in prog-mode-hook and a snippet directory is loaded in the subsequent major
mode's hook.

[monnier]

This sequence of events can happen, for example, if `yas-minor-mode` is called in `prog-mode-hook` and a snippet directory is loaded in the subsequent major mode's hook.

Thanks. I pushed a patch for this in `master`. Can you confirm that it fixes it for you?

[stsquad]

Something has broken with git-commit-mode because now we get:

(yas--modes-to-activate)
(text-mode fundamental-mode fundamental-mode)

I'm not sure if it was these changes or something has changed with git-commit-mode. How should minor modes be handled?

[monnier]

Something has broken with git-commit-mode because now we get: ``` (yas--modes-to-activate) (text-mode fundamental-mode fundamental-mode) ```

Hmm... Could you be more specific? The duplicate `fundamental-mode` is a known problem (which I believe should be mostly harmless) for which I have a fix pending. Should `git-commit-mode` be included above? IF so wherewho is this code run? What did you mean by "now" (i.e. which commit or which GNU-devel/MELPA version are you using)? Also on which Emacs version? I have not been able to reproduce a situation where `yas--modes-to-activate` doesn't return something of the form `(git-commit-mode ...)` when called without args in a `git-commit-mode` buffer.

[stsquad]

It works now so possibly fixed with d124280?

(yas--mode-to-activate) ->
  (text-mode fundamental-mode git-commit-mode)

[monnier]

It works now so possibly fixed with d124280?

Might be. I must admit that I don't know yet how `git-commit-mode` finds its way there :-)