idea: tool to suggest adding imports

[John Williams]

I have an idea for a tool I'd like to implement, and I'm looking for advice
on the best way to do it.

Ideally, I want to write an Emacs extension where, if I'm editing Haskell
code and I try to use a symbol that's not defined or imported, it will try
to automatically add an appropriate import for the symbol. If instance, if
I have "import Data.Maybe (isNothing)" in my module, and I try to call
"isJust", the extension would automatically change the import to "import
Data.Maybe (isJust, isNothing)".

The Emacs part is easy, but the Haskell part has me kind of lost. Basically
I want to figure out how to heuristically resolve a name, using an existing
set of imports as hints and constraints. The main heuristic I'd like to
implement is that, if some symbols are imported from a module M, consider
importing additional symbols from M. A more advanced heuristic might
suggest that if a symbol is exported from a module M in a visible package
P, the symbol should be imported from M. Finally, if a symbol is exported
by a module in the Haskell platform, I'd like to suggest adding the
relevant package as a dependency in the .cabal and/or stack.yaml file, and
adding an import for it in the .hs file.

Here are some implementation options I'm considering:

1. Add a ghci command to implement my heuristics directly, since ghc
already understands modules, packages and import statements.
2. Load a modified version of the source file into ghci where imports like
"import M (...)" are replaced with "import M", and parse the error messages
about ambiguous symbols.
3. Write a separate tool that reads Haskell imports and duplicates ghc and
cabal's name resolution mechanisms.
4. Write a tool that reads Haskell imports and suggests imports from a list
of commonly imported symbols, ignoring which packages are actually visible.

Right now the options that look best to me are 2 and 4, because the don't
require me to understand or duplicate big parts of ghc, but if modifying
ghc isn't actually that hard, then maybe 1 is the way to go. Option 3 might
be a good way to go if there are libraries I can use to do the hard work
for me.

Any thoughts?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/glasgow-haskell-users/attachments/20160318/77936e9b/attachment.html>