Re: official orgmode parser

[Tom Gillespie <tgbugs@gmail.com>]

I started writing down Org's grammar as an EBNF (with Racket's #lang
brag) on Saturday. There is indeed a layer of Org grammar that can be
implemented via EBNF, but it is fairly minimal. You can identify
headlines, but you can't identify nesting level; the arbitrary nesting
depth means that you have to have a stack to keep track. There is a
similar issue with the indentation level in order to correctly
interpret plain lists. If the canonical representation of an org
document was required to used org-adapt-indentation: nil;
org-edit-src-content-indentation: 0 and there was a canonical
normalization function some of these issues would go away, but not all
of them, and I'm fairly certain that it is not possible to implement a
safe normalization function that won't mangle someones formatting.
Another example of something that requires a stack is the greater
blocks, where you have #+begin_{name} and #+end_{name}, and the names
must match. If there was a closed set of names you could sort of do it
by hand, but since name can be any string that does not contain
whitespace, you have to have a stack to track which block you are in.
So, you can identify things that are heads, you can identify things
that are block start lines and block end lines, but you need stacks to
keep track of heading level, indentation, plain list level, and block
name. I might be missing a few other places where stacks are required,
but those are the big ones. Best,
Tom

On Mon, Oct 26, 2020 at 12:48 PM Ken Mankoff <mankoff@gmail.com> wrote:
>
>
> On 2020-10-26 at 09:24 -07, Nicolas Goaziou <mail@nicolasgoaziou.fr> wrote...
> >     # This is a comment (1)
> >
> >     #+begin_example
> >     # This is not a comment (2)
> >     #+end_example
> >
> > AFAICT, you cannot distinguish between lines (1) and (2) with EBNF.
>
> I agree. I think this is a better (correct?) example than the footnotes on Org Syntax page.
>
>   -k.
>
>