26 mars 2020 kl. 04.42 skrev Stefan Monnier : > Then let's make it into a simple `defvar` rather than a defcustom and > make the docstring clarify that it's only useful for temporary > backward compatibility. Right; patch attached. > In sml-mode.el (GNU ELPA) I set `compilation-error-screen-columns` > buffer-locally but it really should be rule-local. In tuareg-mode (not > in GNU ELPA, sadly), I set both `compilation-first-column` and > `compilation-error-screen-columns` buffer-locally for the same reason. And here I've been staring at OCaml type errors for years and never noticed... * Gcc and Clang also count characters rather than columns. (Bytes, in fact. Perhaps we need a setting for that as well?) * Instead of making compilation-error-regexp-alist entries even more complex and overloaded, perhaps we should use plists? For example, each entry could be (REGEXP :file FILE :line LINE ...) Adding :first-column and :screen-columns would then be more a more natural way of doing it. * There could be multiple tools with different column-number semantics sharing a single pattern. * Rule-specific column-number semantics can be done today by specifying a function as COLUMN parameter, so perhaps no extension is necessary.