new lint: unicode_not_nfc

[llogiq]

This closes #85 – I just renamed the lint to make it more clear. Wiki entry will follow once pulled.

I also want to note that once this gets pulled, we will have reached the 50 lints mark! Yay! 😄

@Manishearth r?

[llogiq]

Let this be a testament to my ineptitude for working with anything more complex than SVN. 😄

[llogiq]

Since Manish seems to be awfully busy, @birkenfeld r?

[Manishearth]

(I wasn't awfully busy, my github emails were being sent to spam for some reason. But now I am busy 😛 )

[llogiq]

I was referring to your work week 😜

[birkenfeld]

Well, it seems to work as advertised, however I wonder if it's a good idea to emit a warning for each such sequence.

(Similar for the "non-ASCII char" lint, I think that should be restricted to at most once per string literal).

If you just do a pass that checks for the first mismatch between string and string.nfc().collect() that might be less intrusive, and need less convoluted control flow. If you want to keep the current code, I would request better names for "last" and "ustart" :)

[Manishearth]

Once per stringlit sounds good to me (emit the list of chars if you want)

[llogiq]

I'd really like to show where in the string literal the problematic characters are. Can we report multiple spans in one line?

@birkenfeld Thanks, I'll think about the variable names. I'd really like to keep the convoluted control flow, because I believe showing the locations where things go wrong (which btw. is a bit broken because zero-width whitespace throw off the formatting) is a really useful feature.

[Manishearth]

No, but multispan reporting is a desired feature for rustc (e.g. for unused imports)

[birkenfeld]

BTW, does "showing the locations" do useful things in the case of multiline string literals?

[llogiq]

It should. Until now I had no problems with reporting on multiline snippets.

Since we don't have the means to make the report uncluttered, I'll report the whole string, along with pairs of sequences to search/replace.

[Manishearth]

Make sure to add special handling for the language when you have a single bad char!

[llogiq]

Will do. Alas, it'll have to wait – I'm a bit busy ATM.

[llogiq]

Now I've special-cased the one-range case so that the respective part of the string is spanned. In the other cases, a newline-joined list of ranges + suggested replacements is reported with the whole string as span.

@Manishearth @birkenfeld r?

[llogiq]

Since the solution to the problem does not involve seeing the actual ranges, I created a far less complex version that directly lints the strings and suggests replacements. I'm closing this in favor of the other solution.