bug#59900: 30.0.50; map pattern in pcase causes a byte-compile error

[Michael Heerdegen]

Eli Zaretskii <eliz@gnu.org> writes:

> Thanks, but all this still doesn't answer my questions,
> unfortunately.  You-all are describing something that I cannot wrap my
> head around, because that basic question was not answered yet.

What specific question?

pcase is extensible: It allows to define new pattern types.  "rx.el"
defines a `rx' pcase pattern, "map.el" a `map' pcase pattern.  The
normal way to be able to use them is to `require' the library in which
they are defined.

Since `pcase' is a macro, the library (rx.el or map.el) must be loaded
when a `pcase' form using such a pattern is compiled.

The `rx' pattern definition in rx.el uses autoload cookies so that
compiling works even when rx.el is not loaded - it gets loaded
automatically when compiling.

The map.el `map' pcase pattern doesn't use autoloading.  The OP found a
case where compiling succeeded until some days ago without requiring
map.el explictily, and now a require is needed.  But that was only by
luck: AFAIU, map.el was already loaded in that scenario, and now it is
any more.  But there is no guarantee that the library is always loaded
when compiling arbitrary files, so one should _always_ explicitly
require map.el when the file contains pcase forms with `map' patterns -
even if this worked without in some cases in the past by luck.

The wish of the OP to make the `map' pattern in map.el `autoload'able
like the `rx' pattern in rx.el is reasonable, I can try to create a
patch.

But strictly speaking here is no bug, just the OP relying on something
that in the past worked by luck.

Hope this answers everything - else please ask specific questions.


Michael.