[GHC] #8809: Prettier error messages?

[GHC]

#8809: Prettier error messages?
-------------------------------------+-------------------------------------
        Reporter:  joelteon          |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.9
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
  #8809,#10073,#10179                |
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by Rufflewind):

 > Display one line of code extracted from the original source code, along
 with a marker indicating the location of the error.

 So this (“caret diagnostic” is what Clang calls them) is the one I'm a bit
 stuck on.  In order to print a line in the original source file, I would
 need to read a file, i.e. perform `IO`.  This is not possible with
 `ErrUtils.mkLocMessageAnn`, AFAICT.

 I thought of a workaround, but it would require a small tweak to `Doc`: I
 need to add an atom for `CaretDiagnostic` containing the `RealSrcInfo`,
 and then read the original source file in `printDoc`.  However, I feel
 that this is a bit hacky.

 IMO, there should really be two layers in the system:

   - A `Message` type stores each individual diagnostic message unit along
 with an associated `SrcLoc` and any other useful information in their
 original format (e.g. fragments of the AST).
   - `Doc` should remain as a simple, textual output device with fancy
 formatting (no support for embedded ASTs).  (It does need to understand
 colors and bold though, as their widths must be treated as zero.)

 The rendering function `[Message] -> IO Doc` would support `IO`, and this
 is where the caret diagnostics can be constructed, and it would also
 provide all the necessary information for `Doc` deduce alignment
 correctly.  Hence, IMO the debate over how to represent ASTs should focus
 on the `Message` type, not `Doc`.

 That being said, even if this idea does work, I'm not sure I have the time
 to work on a major refactor right now, so I might go with the hacky
 solution for the time being.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8809#comment:27>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler