Rust: Use Unicode line drawing characters for spans

[kmcallister]

See discussion on old PR #21406. @Manishearth expressed interest in reviving this.

[sinistersnare]

+1 sounds nice.

Would this work on any windows terminals, such as the shell that comes with mingw?

Oh just read through the thread again, WGL4. cool!

[liigo]

Always prefer ASCII characters.

[tbelaire]

The plan, from before was to have both a flag --output-plain-ascii, as well as environment variables which can be set to always, auto or never, for each of color, having underlines at all, and having unicode underlines.

[ftxqxd]

Alternatively, we could use U+203E OVERLINE (‾, also part of WGL4) to create a simpler underline effect:

This is different from using an ANSI escape code such as \x1b[4m because it allows the underline colour to be different from the text’s colour.

[shinglyu]

I'm a noob, can I give it a try? I could try writing a unicode version, then implement the --output-plain-ascii flag.

[Manishearth]

@shinglyu sure! You can borrow a lot of stuff from Keegan's impl here, feel free to ask me if you have more questions!

[shinglyu]

@Manishearth : nice to see you again :)

[shinglyu]

@Manishearth : I tried to read the patch https://github.com/rust-lang/rust/pull/21406, but I had a hard time understanding the codebase. Is there any document/blog post/diagram/etc that explains the folder structure and runtime behavior? Most of the docs I found are explaining the Rust language syntax.

Or should I start by reading the llvm tutorial?

[Manishearth]

src/librustc/session/config.rs contains the stuff for sorting out command line arguments

src/libsyntax/diagnostic.rs has the stuff for printing out a span.

You don't need to understand the rest at this stage, really :)

[shinglyu]

@Manishearth : thanks! I'll take a look.

[shinglyu]

@Manishearth @kmcallister : Changing ascii to UTF-8 is pretty straight forward. But I want to clarify the spec.

From what I understand in the closed PR,

• We have 3 environment variables for output capability
  
  
  
  • RUSTC_OUTPUT_COLOR = always | auto | never
  
  
  • RUSTC_OUTPUT_CURSOR_POS = always | auto | never
  
  
  • RUSTC_OUTPUT_UTF8 = always | auto | never
  
  
  • If any of the three is un-set or is invalid value, use auto
  
  
• There are 2 way to set the above three as never:
  
  
  
  • --output-plain-ascii flag
  
  
  • RUSTC_OUTPUT_PLAIN_ASCII =
  
  

I have some questions:

• If cursor positioning is not enabled, what do we do?
• What are all those flag stability and warning thing?

Thanks!

[bluss]

Most programs will check against LC_CTYPE (i.e. the locale's encoding setting) to see what they can emit. If it's not a UTF-8 locale, just emit ascii.

[christopherdumas]

Is this going anywhere?

[christopherdumas]

I like the idea, but Rust still doesn't do this.

[pmatos]

@shinglyu how is this going?

[shinglyu]

Sorry I claimed it but didn't finish it on time. Give me one week to work on this. If I can't make it, anyone if free to take it.

[pmatos]

OK, @shinglyu , if you see you won't make it, can you please post your progress so someone can pick it up where you left off instead of starting from scratch? Thanks!

[pnkfelix]

Uh @shinglyu @pmatos do note there is https://github.com/rust-lang/rust/pull/28902

[shinglyu]

Oh ok, so look like the old PR is being revived. Let's go with #28902 then.

[brson]

With the recent churn in error reporting, this will probably need to start from scratch. Still a cool idea.

[brson]

cc @jonathandturner maybe you have ideas about how to beautify new error reporting with unicode.

[jonathandturner]

We've got an RFC going here: https://github.com/rust-lang/rfcs/pull/1644

We opted not to use Unicode because we did a look at what was generally available across platforms and Windows doesn't use Unicode by default in the terminal. I know that they're working on a better terminal experience, so things might change there.

[steveklabnik]

The error format RFC ended up explicitly deciding against this: https://github.com/rust-lang/rfcs/blob/master/text/1644-default-and-expanded-rustc-errors.md#format-of-error-messages So I'm going to give it a close.