Re: orgalist-mode: wrong indentation in message mode after recent change in emacs

[Stefan Monnier <monnier@IRO.UMontreal.CA>]

>>> The first is whether orgalist-mode couldn't use a custom
>>> indent-line-function instead of advising what may or may not be set to
>>> indent-relative by the user.
>> I don't know how that would work in practice.
> Me neither.
>> But a minor mode taking control over `indent-line-function' sounds
>> wrong.
> Well, orgalist already "takes control" over indent-line-function and
> indent-according-to-mode via advice, and the latter advice seems to
> assume that indent-line-function is set to the default indent-relative
> or indent-relative-first-indent-point.

I haven't bothered to look at the advice to have an opinion here, so
I'll let you guys figure out this part.

>>> The second is why advice--buffer-local does what it does.  Stefan, why
>>> does it behave differently depending on local-variable-p?  Why can't it
>>> simply call make-local-variable before returning the symbol-value?

Normally a hook runs both its local part and its global part, where the
global part is represented in the local list by the special element `t`.

When you do `make-local-variable` you're basically *copying* the global
part to the local part, with the usual implications: when the global
part is later modified those modifications won't be properly reflected
in the local copy.  That's why we had `make-local-hook` which is now
automatically performed by `add-hook` depending on the `local` arg.

The exact same thing goes for `add-function` when applied to a variable.

In the current case, I think calling `make-local-variable` is likely
harmless because *we* know the global value should likely never change,
but that's not something that `add-function` knows to be true in
general.  So instead, `add-function` sets the local value to a function
that looks up the global value and runs it, which is the moral
equivalent of the `t` element on normal hooks.

>>> The third is why indent-according-to-mode hard-codes the check for
>>> indent-relative and indent-relative-first-indent-point.

History.  Comparing functions is always a bad idea.
But I couldn't and still can't see how to avoid it here without
introducing worse problems.

>>> Wouldn't it be nice if this check instead looked up some variable
>>> akin to electric-indent-functions-without-reindent, that can be more
>>> easily customised?

Yes, tho it probably wouldn't help much here: not only we'd still be
comparing functions, but we'd also need for orgalist to go through the
trouble of manually adding the closure dynamically created by
`add-function` to this list.


        Stefan