Re: [PATCH] Improve find-sibling-rules option type

[Mauro Aranda <maurooaranda@gmail.com>]

On 24/9/23 22:17, Paul W. Rankin wrote:
 >
 > On 2023-09-24 19:44, Mauro Aranda wrote:
 >> In particular, AFAICT there could be more than one EXPANSION.  In that
 >> case, :value-type should be a repeat of strings, not just a string.  In
 >> addition, regexp should be used as the :key-type type. (I see that
 >> Philip already spotted this)
 >
 > Just looking at `find-sibling-file-search' and it looks like maybe? 
But if this is the case I think the docstring should be rewritten. 
Looking at it from a user's perspective if the docstring says "a string" 
but there's a repeat of strings, that would be confusing.

My clue was (MATCH EXPANSION...) and then it says EXPANSIONS.  And each
EXPANSION is a string, hence we have a repeat of strings.

 >> In general, when converting from sexp to a more specific type, I think
 >> we have to take extra precaution and check the code for how the
 >> variable it's used (i.e., the docstring might not tell the whole story).
 >> If ever in doubt, then it might be better and safer to offer a choice
 >> with the more specific type first, and keep sexp as a catch-all
 >> alternative.
 >
 > Customize already provides for this! The user can click [State] > 
Show Saved Lisp Expression to edit/override the value with their own sexp.

That was not my point.  My point was that in a "sexp to
more-specific-type" change, there's always a risk to introduce a minor
regression: that the user starts seeing a MISMATCH instead of SET/SAVED
as a message in Customize.  IMO, that needs to be avoided as much as
possible.

But AFAICS, there will be no trouble here, provided that :value-type is
a repeat of strings.

 > I think the main thing is not to have user options that demand sexps.

I agree.  I'm in favor of avoiding sexp as much as possible, be it by
providing a more specific type, or a not-that-complicated user option.