Why does filling in AUCTeX behave so badly?

Question

I'm using AUCTeX to edit LaTeX files. The behavior of the fill command M-q (also C-c C-q C-p, LaTeX-fill-paragraph) is very erratic. For example, if I have the following text:

test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$

filling it yields:

test $1 + 1 = 2$
test $1 + 1 = 2$
test $1 + 1 = 2$
test $1 + 1 = 2$
test $1 + 1 = 2$
test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$

which is clearly not ideal. However, I have auto-fill-mode enabled, and if I take the text from the beginning and manually add a space at the end of the line, the auto-fill does this:

test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$
test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$ test $1 + 1 = 2$
test $1 + 1 = 2$ 

which is infinitely better. Is there a way I can configure the fill-command to work correctly? I'm using Emacs 24.5.1 on Windows and AUCTeX 11.88.9.

Answer 1

The behavior of filling of math switches is controlled by the LaTeX-fill-break-at-separators variable and there are currently discussions about its best default value.

I recently came up with with setting

(setq LaTeX-fill-break-at-separators '(\\\( \\\[))

which fulfills the goal of keeping math formulae in the the same line (this is useful for example for preview-latex) and to provide a somewhat pleasing filling. With the above setting, your code is filled as you expected (fill-column set to 70). As suggested by Omar Antolín-Camarena in comments, it should be noted that the presence of the opening \( in that list affects also the behavior of $ as AUCTeX treats them in the same way.

---

For the record, the above value LaTeX-fill-break-at-separators is now the default in the just released AUCTeX 11.89.