bug#28557: 26.0.60; Bugs using (:documentation FORM) in closures

[Stefan Kangas]

Gemini Lasswell <gazally@runbox.com> writes:

> Since Emacs 25, docstrings can be constructed using
> (:documentation FORM) in closures, when lexical-binding is on. This
> isn't documented except in NEWS. Bug #24773 is about Edebug specs
> which have not been updated to recognize :documentation. I'd like to
> fix it, so I started trying to figure out which macros accept
> :documentation forms so I would know which Edebug specs to update, and
> that effort turned into a collection of tests for the feature, see
> attached:

There is a long list of tests here, but it seems like none of this was
ever committed to master.  Should it be?

> While doing this I found these bugs:
>
> 1. A lambda wrapped by cl-function with a :documentation form will
> not get a docstring and the :documentation form will be left in the
> body of the closure.
>
> 2. In code run interactively, a :documentation form in cl-defmacro or
> defmacro will work correctly. However if you byte-compile that code,
> the byte-compiler stops with the following error message:
>     Symbol’s function definition is void: internal-make-closure
>
> 3. A :documentation form in cl-defgeneric only works if its argument
> is a string literal. In cl-defmethod, it works whether the argument is
> a string literal or an expression.
>
> 4. A :documentation form works in cl-iter-defun and iter-defun, but
> not in iter-lambda.
>
> Stefan, here's a list of macros in which it appears to me that
> (:documentation FORM) is supposed to work. Is this correct? Am I
> missing any?
>
> lambda
> pcase-lambda
> defun
> defmacro
> cl-defun
> cl-defmacro
> iter-defun
> cl-iter-defun
> iter-lambda
> function
> cl-function
> cl-defgeneric
> cl-defmethod
> defsubst
> cl-defsubst

Does anyone know the answers to the above questions?

Best regards,
Stefan Kangas