Re: \with mods targeting parents

Carl Sorensen <[email protected]> writes:

> On Mon, Oct 27, 2025 at 1:14 PM David Kastrup <[email protected]> wrote:
>>
>> We actually went through all this once already including actual code,
>> and the results were unhelpful.
>>
>> Please check
>>
>> <https://gitlab.com/lilypond/lilypond/-/issues/2982>
>>
>
> Sorry about the noise.  I agree with your conclusion.

I did not actually remember that we already were there before, and I did
not remember whether I merely had an internal discussion with myself on
it.  And looking at this thread right now, it looks an awful lot like I
tried to strongarm myself through that discussion.

I apologize for that.

I agree that not all of the consequences of this design decision are
likeable.  But the problem is that something like

mysettings = \with { ... }

creates a context modification block that has no clue whatsoever to what
kind of context it will be applied, and context modifications do not
contain any context specifier.

So with that kind of statement, we don't have what it would take to act
context-specifically, and we don't retain enough information to act on
it later on.

Apart from that syntax, we have something like

\new Staff \with ...

and here at the time of parsing we only know about this being for
"Staff" but not where Staff is actually sitting in the hierarchy: the
context hierarchy is only established once music is iterated.  So we
cannot really decide whether we are upward or downward or off-side of
Staff with any context specified in an override.

And then we have

\layout {
  \context {
    \Staff
    ...
  }
}

Again, this is before iteration happens, and where the respective
overrides are sitting, it is not even clear whether the final \name of
the context has been established (\Staff just copies the \name and other
settings of "Staff", but it may well be renamed to something else
afterwards in order to create a Staff-like context).

Given all that, I actually wonder in what manner the discriminative
conversion implemented before that change was actually able to work at
all rather than "merely" in a surprising manner.

Warnings, of course, have the advantage that they don't need be
consistent.  But I don't see how we could confine them to a subset of
circumstances that were reasonably certain to be unintentional.

So I don't really see us getting more helpful here.  The logic and
information does not seem to deliver.

-- 
David Kastrup