Re: command-separator char

[Andreas Röhler]

Am 23.06.2015 um 19:00 schrieb Stefan Monnier:
> > Seems still no idea or definition what a sexp should be.
> Nope.  It's only a UI issue.
>
> Sexp navigation is basically navigation within the AST, but a given
> buffer position corresponds to several different spots in the AST, so
> there's an ambiguity.  Take your example line:
>
>      echo "foo"; echo "asdf";
>
> the AST looks like
>
>             echo
>           /
>       cmd
>      /    \
>     /       "foo"
>    ;
>     \       echo
>      \     /
>        cmd
>           \
>             "asdf"

Thanks a lot for your care!

> The position at BOL can be take to be right before the "echo" node, or
> right before the first "cmd" node or right before the toplevel composite
> instruction that comprises the whole line.
>
> You can mostly tell SMIE which one you want if you call smie-forward-sexp,
> by passing it a token explaining the level you care about.
>
> But forward-sexp is called by the user without any extra info, so it has
> to make an arbitrary choice.  For compatibility with the usual
> forward-sexp semantics, the choice it makes is to consider the "deepest"
> position in the tree.

That would be plausible. Unfortunatly can't see that.
From end of first string jumps into second string. While there is an 
expression of three minor expressions:

echo _ "foo"_ ;

WRT to deepest it should keep the level and end after ";"


> But as explained earlier, if you put point *before* a semi-colon, and then use
> M-C-f, then SMIE's forward-sexp will take this as a clue that you want
> to skip over whatever belongs to the right-hand of this semi-colon.
> So with point right after "foo", C-M-f will jump to right after "asdf".
>
> Along the same lines, you say:
>
>    With cursor inside first string --> end of first string
>
> meaning that this is inconsistent,

No, didn't say that. That behaviour is expected - inconsistence was 
shown by the relation of examples.


> but it's perfectly consistent, since
> forward-sexp does not jump to "the end of the currently enclosing
> entity" but instead to "the end of the entity that starts right after
> point".
>
> SMIE's forward-sexp could very well check syntax-ppss to see we're
> inside a string (or comment) and move outside of that string, but then
> it would be a different command.

IMHO a much more usefull one :)


>          Stefan
>
>
> PS: That doesn't mean that it always works right either, of course.
> M-C-f with point right after the second "o" of "foo" (i.e. right before
> the closing double quotes) should signal an error

Think navigation shouldn't signal an error. Rather return nil - as at EOB.


> but will instead jump
> to right after the opening double quotes of "asdf".
> PPS: And `up-list' doesn't use SMIE quite right either.

IMO grammar oriented commands are needed - starting with token maybe.

Languages commonly know expressions, compositions of them - and so upwards.

Thanks,

Andreas