James N V Cash writes: > I've attached a patch that addresses the recursion issue with Emacs 28 > and shows eldoc properly with example python. It presumably should act > the same with older versions of Emacs, although I haven't tested.
Thanks for the patch! For information about the expected commit message format, please see <https://orgmode.org/worg/org-contribute.html>. > diff --git a/contrib/lisp/org-eldoc.el b/contrib/lisp/org-eldoc.el > index 3b0999340..ccc23b523 100644 > --- a/contrib/lisp/org-eldoc.el > +++ b/contrib/lisp/org-eldoc.el > @@ -116,9 +116,12 @@ > (when (fboundp mode-func) > (with-temp-buffer > (funcall mode-func) > - (setq doc-func (and eldoc-documentation-function > - (symbol-value > 'eldoc-documentation-function))) > - (puthash lang doc-func org-eldoc-local-functions-cache)) > + (setq doc-func (if (boundp 'eldoc-documentation-functions) > + (car eldoc-documentation-functions) > + (and eldoc-documentation-function > + (symbol-value > 'eldoc-documentation-function)))) > + > + (puthash lang doc-func org-eldoc-local-functions-cache)) > doc-func) > cached-func))) Okay, so when eldoc-documentation-functions is defined (Emacs >=28), we take the first function and go with it. That might not be exactly what you'd see in the native buffer, depending on whether there are other members of eldoc-documentation-functions and how they interact. (I'm being vague, because I don't really know anything about eldoc, but it seems like that must be the case.) Anyway, I'd guess it will be good enough in most cases, and it's certainly better than the recursion error. However, the docstring of eldoc-documentation-functions says Each hook function is called with at least one argument CALLBACK, a function, and decides whether to display a doc short string about the context around point. But downstream in org-eldoc-documentation-function, the function is called with no arguments: (t (let ((doc-fun (org-eldoc-get-mode-local-documentation-function lang))) (when (functionp doc-fun) (funcall doc-fun)))) That happens to work with python-eldoc-function, whose signature is (&rest _ignored), but it may error with other valid eldoc functions. So, at the very least, I think we should wrap the call in a condition-case with a wrong-number-of-arguments handler. Thoughts?
