Clean symbol before doing look up

[tanzoniteblack]

If you attempt to run cider-jump-to-var to find the definition of an
atom dereferenced with @, cider can't find the source location. If you remove
the @ cider has no issues finding the location. This allows for cleaning
non-variable components out of a symbol before doing the look up to prevent this
issue.

Possible fix for #886

[bbatsov]

I think it will be better to do the cleanup in cider-symbol-at-point.

[tanzoniteblack]

@bbatsov agreed, that is a much better location for this logic. Updated the pull request to stick the new cider--clean-symbol into cider-interaction.el and to have this function called in cider-symbol-at-point instead.

[tanzoniteblack]

Also, I'm using let* as a hacky way to allow conditional piping of changes (ala clojure's cond->), incase we find other string additions that prevent this from working in the future. If you have a better elisp suggestion for this, or think it's an unnecessary at this time and just doing the basic if statement would be better, let me know.

[bbatsov]

I think you can simply use a regexp to remove anything that's not a letter at the beginning of the symbol name.

[bbatsov]

P.S. You should also add a changelog entry and a couple of unit tests.

[tanzoniteblack]

I'll add some tests and update the change log in a bit (and make sure all existing tests pass...which they don't).

Do we really want to strip anything that's not a letter from the beginning? That would cause a valid symbol like -var or _var to break. According to http://clojure.org/reader a symbol can start with any letter or a character from the list [*, +, !, -, _, ?]. I'll try to get the function to clean it up to consist of that, and make sure we don't break on unicode characters.

[bbatsov]

You're totally right; I should have thought a bit more about the problem. Your proposal is sound. Another alternative would be to strip just things we know we should strip (@, #', ', etc)

[tanzoniteblack]

I prefer the stripping things we know we should strip specifically, rather than attempting to strip anything that doesn't look valid. It's an easier to problem to get right without odd bugs creeping up.

[tanzoniteblack]

I've fixed the failing test (by accounting for empty string causing substring to throw errors) and added a new one to specifically test for @item.

I can include #' and ' in the stripper if you want, but 1.) I'm not sure if it's invalid to send them to our symbol look up, and 2.) the way we get the symbol at point in Emacs causes these to not be counted as part of the word currently anyways.

[bbatsov]

the way we get the symbol at point in Emacs causes these to not be counted as part of the word currently anyways.

This leads me to be believe that clojure-mode's syntax table might need adjusting. Guess @ is treated as a word character and ' is not. Updating the syntax table should solve this without the need to do anything in cider itself.