Issue #150: Review behaviour of options and plugins directives in included files (blais/beancount)

[Martin Blais]

New issue 150: Review behaviour of options and plugins directives in included files
https://bitbucket.org/blais/beancount/issues/150/review-behaviour-of-options-and-plugins

Martin Blais:

Minimum: There should be a warning if ignored.
Better: Consider alternative semantics.

See thread below.



On Sun, Dec 25, 2016 at 02:35:08PM -0500, Martin Blais wrote:
> Tell me, how would you expect / like that the options and plugins
> directives be interpreted over multiple files?
> What seems to be a good behavior in your view?

JFTR: I use a single "big ledger" file, so I don't personally suffer
from the problem of repeating plugins in multiple files. Still, from a
"ledger engineering" point of view, I'd love to have a single "include
plugins.beancount" directive somewhere.

Answering your question, as a user I'd love "include" to be completely
transparent. That is: the behavior of beancount on a file that contains
one or more "include" lines should be exactly the same of beancount
called on a different file where all includes have been (transitively)
resolved inlining the content(s) of the referenced file(s).

I'd be totally fine if, to achieve this, you'll end up imposing
draconian restrictions such as: "all 'plugin' and/or 'option' directives
must appear at the beginning of the file obtained after resolving all
'include' directives".

I do realize that this might entail implementing a two-stage parsing of
sort, which might be annoying. But you asked about my favorite behavior,
and that would be it :)


Responsible: blais