Question

So I have a tool lints python changes I’ve made and produces errors and warnings. I would like this to be usable in compile mode with Emacs, but I have an issue. The file name is output only once at the beginning, and then only line numbers appear with the errors and warnings. Here’s an example:

Linting file.py
E0602: 37: Undefined variable 'foo'
C6003: 42: Unnecessary parens after 'print' keyword
2 new errors, 2 total errors in file.py.

It’s very similar to pylint, but there’s no output-format=parseable option. I checked the documentation for compilation-error-regexp-alist, and found something promising:

If FILE, LINE or COLUMN are nil or that index didn’t match, that
information is not present on the matched line. In that case the
file name is assumed to be the same as the previous one in the
buffer, line number defaults to 1 and column defaults to
beginning of line’s indentation.

So I tried writing a regexp that would optionally match the file line and pull it out in a group, and then the rest would match the other lines. I assumed that it would first match

Linting file.py
E0602: 37: Undefined variable 'foo'

and be fine. Then it would continue and match

C6003: 42: Unnecessary parens after 'print' keyword

with no file. Since there was no file, it should use the file name from the previous match right? Here’s the regexp I’m using:

(add-to-list 'compilation-error-regexp-alist 'special-lint)
(add-to-list 'compilation-error-regexp-alist-alist
         '(special-lint
           "\\(Linting \\(.*\\)\\n\\)?\\([[:upper:]][[:digit:]]+:\\s-+\\([[:digit:]]\\)+\\).*"
           2 4 nil nil 3))

I’ve checked it with re-builder and manually in the scratch buffer. It behaves as expected. the 2nd group is the file name, the 4th is the line number, and the 3rd is what I want highlighted. Whenever I try this, I get the error:

signal(error ("No match 2 in highlight (2 compilation-error-face)"))

I have a workaround for this that involves transforming the output before the compile module looks at it, but I’d prefer to get rid of that and have a “pure” solution. I would appreciate any advice or pointing out any dumb mistakes I may have made.

EDIT

Thomas’ pseudo code below worked quite well. He mentioned that doing a backwards re search could mess up the match data, and it did. But that was solved by adding the save-match-data special form before save-excursion.

Answer 1

FILE can also have the form (FILE
FORMAT…), where the FORMATs (e.g.
“%s.c”) will be applied in turn to the
recognized file name, until a file of
that name is found. Or FILE can also
be a function that returns (FILENAME)
or (RELATIVE-FILENAME . DIRNAME). In
the former case, FILENAME may be
relative or absolute.

You could try to write a regex that doesn’t match the file name at all, only the column. Then for the file, write a function that searches backwards for the file. Perhaps not as efficient, but it should have the advantage that you can move upwards through the error messages and it will still identify the correct file when you cross file boundaries.

I don’t have the necessary stuff installed to try this out, but take the following pseudo-code as an inspiration:

(add-to-list 'compilation-error-regexp-alist-alist
         '(special-lint
           "^\\S-+\\s-+\\([0-9]+\\):.*" ;; is .* necessary?
           'special-lint-backward-search-filename 1))

(defun special-lint-backward-search-filename ()
  (save-excursion
    (when (re-search-backward "^Linting \\(.*\\)$" (point-min) t) 
      (list (match-string 1)))))

(It could be that using a search function inside special-lint-backward-search-filename will screw up the sub-group matching of the compilation-error-regexp, which would suck.)