php-mode executes c-mode hooks

[NateEag]

I recently added cquery to my c-mode setup, which meant adding a c-mode-hook to start cquery.

Doing that broke my ability to open PHP files sanely, because cquery starts itself before my lsp-php instance gets a chance to start, resulting in a broken state. Since they're both lsp-mode plugins, each targeting a specific language, it makes sense that they conflict.

I gather that php-mode is derived from c-mode, and therefore runs c-mode's hooks. I've worked around this problem by only starting cquery when (equal major-mode 'c-mode), but it doesn't seem like I should have to do that.

This StackOverflow question shows a similar problem and workaround.

The answer there points out that c-mode and friends actually derive from prog-mode, and suggests that php-mode should probably do that too. A quick search of cc-mode.el in Emacs 25.3.1 for define-derived-mode shows that is in fact how cc-mode.el currently does things.

I looked for any existing issues or discussion around this point, but I wasn't able to find any. I did find this thread with some discussion about the move to basing php-mode on cc-mode in the first place.

Should we consider extending prog-mode rather than c-mode?

If so, how much work would that involve?

Could you check this branch in my repo? https://github.com/fallchildren/php-mode/tree/prog-mode-derived This seems to do it (2 lines changed as we just need the syntax table from c-mode). But seems to introduce a font-locking bug in emacs 25 regarding comments (The comment test fails). Which is fixed in conjunction with my Pull-Request #401. @syohex Would it be okay to add this as a pull request but only with the pre condition of #401 merged?

I haven't tested this thoroughly though.

[NateEag]

@fallchildren That change does resolve my hook problem. Thanks!

I can see the font-lock bug you're talking about, and I haven't tested it thoroughly either.

I probably can use it for a daily driver for a while if/when #401 lands and see if it causes me any issues.

@zonuexe What's your thought on this?

[zonuexe]

I think it is better to bypass (define-derived-mode cc-mode) as well.
js-mode is a good model for us, it uses the function of cc-mode, but bypassed inheritance.
https://github.com/emacs-mirror/emacs/blob/973d10a34b23f2ce5acc00a90ec9767e6237fefa/lisp/progmodes/js.el#L3858

I think this method is correct, I agree to refactor php-mode as that way.

@fallchildren
I will be able to undertake that change in the next leisure, but it will probably take some time.
Can you make this refactor?

@zonuexe I can try... a very simple approach is in #409. This one does get rid of the derived-mode but not of (c-add-language 'php-mode 'java-mode) and all the calls to c-lang-defconst which js-mode does not use. You would want to get rid of those as well as shown in js-mode, right?

[ejmr]

The command git grep "cc-mode" will be a fast way to find most of what will need careful re-testing after any switch to prog-mode, js-mode, etc.

[zonuexe]

@NateEag
This problem was solved by #492. This change will be released as v1.21.0.

[NateEag]

@zonuexe Awesome. Thanks for the heads-up!