GitLab

also changed installed-pkg-config to a dotfile.

** I'm not sure why I was encoding the version string into the package
   name for ghc-pkg.  I got rid of this.  Seems to still be happy.

** Had to fix some test cases for this also.

- this caused a problem since the parser doesn't accpet HUnit-1.0 as a
  package name.  That broke make-install.

  - The almost-correct behavior was to fix the version within Configure
    if they specify an exact version dependency.

    - But the question is, why have I encoded the version into the name
      string for ghc-pkg?  Shouldn't I remove this?  Was it to simulate
        version dependencies since ghc-pkg doesn't do this yet?

	- Anyway,  make install works now.

- new Setup.distribution for A
- make clean && make for A in moduleTest
- Setup.hs -> Setup.lhs, just because

Note that while the parser supports description files without library
building and with multiple executable building, the build system does
not (yet!)

I've implemented a mapping[1] between compiler extensions and
command-line flags for Cabal[2] for GHC, NHC98, and Hugs.  Can some
representatives from each take a look at these and let me know if I'm
wrong about how any of them work, whether I missed support for any of
the compilers, or what-have-you?

Some notes & questions:

- I gathered these by the rather error-prone process of copying them
  from the end-user documentation.  If anyone notices anything that's
  missing, or can suggest some way I could discover missing items,
  please do.

- I'm not including any non-standard modules, since maybe those will
  be handled by the packaging system.

- Are the "hood debugging extensions" in NHC and Hugs compatible?
  Should they be included?

- Do any of Hugs, GHC, and NHC disallow Long.Hierarchical.Module.Names
  by default?  Can they all turn them off?

- Should there be a "Force98" option or something which does "-98" in
  NHC, "+98" in Hugs, and forces no other extensions in GHC?
  Presumably, it would disallow other extensions for the other
  compilers as well.

** now does rm -r instead of trying to delete each created file
** added tests for make clean
** got rid of clean module altogether
** added removeFileRecursive to Utils

-- |FIX: Doesn't remove directories created.  Perhaps it should just
-- delete the .hi and .o files?  Perhaps it should just delete the
-- entire build location recursively?

* Implement parseDescription in terms of a splitLines function
* Remove old code (also commented out test cases, sorry...)
* The error monad wrapping Parsec's ParseError could use some work

** Changed 'package' field to be empty identifier
** Implemented multiple field parsing (works for version, name, license)
** implemented version parsing
** STILL have not got parseFreeText to stop correctly
** more testing

** Made PackageDescription derive Eq, for testing
** setPkgName and setPkgVersion convenience functions
** outline (commented out) for parseReqFields
** outline for parseFreeText
** type sig for parseDesc
** string for package description and known correct answer

** Started parser and tests cases
** But fails test cases
** Probably because of only one look-ahead
** reorganize for 1 look-ahead