Re: [GSOC][PATCH] userdiff: add support for Scheme

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Mon, Mar 29 2021, Johannes Sixt wrote:

> Am 29.03.21 um 12:18 schrieb Phillip Wood:
>> It would be nice to include indented define forms but including them
>> means that any change to the body of a function is attributed to the
>> last internal definition rather than the actual function. For example
>> 
>> (define (f arg)
>>   (define (g x)
>>     (+ 1 x))
>> 
>>   (some-func ...)
>>   ;;any change here will have '(define (g x)' in the hunk header, not
>> '(define (f arg)'
>> 
>> I don't think this can be avoided as we rely on regexs rather than
>> parsing the source so it is probably best to only match toplevel defines.
>
> There can be two rules, one that matches '(define-' that is indented,
> and another one that matches all non-indented forms of definitions. If
> that is what you mean.

Yes, but that doesn't help in these sorts of cases because what a rule
like that really wants is some version of "don't match this line, but
only if you can reasonably match this other rule".

We can only do rule precedence on a per-line basis via the inverted
matches.

So for languages like cl/elisp/scheme and others where it's common to
have nested function definitions (then -W would like the top-level) *OR*
similarly looking nested function definitions, but the top-level isn't a
function but a (setq) or whatever we're basically stuck with picking one
or the other.

I've pondered how to get around this problem in my userdiff.c hacking
without resorting to supporting some general-purpose Turing machine, and
have so far come up with nothing.

You can see lots of prior art by grepping Emacs's source code for
beginning-of-defun, it solves this problem by exposing a Turing machine
:)