bug#36434: 26.2; smie-rule-next-p needs smie--after to be defined

bug#36434: 26.2; smie-rule-next-p needs smie--after to be defined

[Sam Halliday]

When writing SMIE indentation rules, as per `smie-rules-function`,
when an :elem is received (confirmed at least with 'empty-line-token)
it is not possible to call the function smie-rule-parent-p, because
the smie--after variable has not been defined.

A workaround is to define the variable locally, e.g.

        (defvar smie--after)
        (setq smie--after (point))

bug#36434: additional information

[Sam Halliday]

Furthermore, and probably for the same reasons, it is not possible to
use `smie-rule-parent-p` in an :elem because `smie--parent` is not
defined.

FWIW I'm able to use these functions ok in a :before or :after

bug#36434: 26.2; smie-rule-next-p needs smie--after to be defined

[Stefan Monnier]

> When writing SMIE indentation rules, as per `smie-rules-function`,
> when an :elem is received (confirmed at least with 'empty-line-token)
> it is not possible to call the function smie-rule-parent-p, because
> the smie--after variable has not been defined.

[ Hmm... this is related to the smie-rule--parent mess.  ]

But I'm not completely sure how we can define "parent" in a really
unambiguous way, here: in the `empty-line-token` case we're presumably
on an empty line, so there's no token either immediately before or
immediately after.

Basically, I think this is a documentation bug: the smie-rule-*
functions are meant to be used for the :after and :before cases, but not
for :elem where there isn't necessarily such a clear "current token"
(tho I think it would be good to be more careful and try to
define such a "current token" when we can).

>         (defvar smie--after)
>         (setq smie--after (point))

Or use smie-backward-sexp rather than smie-rule-parent, which doesn't
depend on internals.


        Stefan

bug#36434: 26.2; smie-rule-next-p needs smie--after to be defined

[Lars Ingebrigtsen]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> Basically, I think this is a documentation bug: the smie-rule-*
> functions are meant to be used for the :after and :before cases, but not
> for :elem where there isn't necessarily such a clear "current token"
> (tho I think it would be good to be more careful and try to
> define such a "current token" when we can).

Looking at the smie-rules-function doc string, this doesn't seem to have
been added?  My understanding of smie is...   er...  let's be
charitable and call it "not funny encompassing", so I don't feel
qualified to take a stab at it.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no