ghc-lib-0.20190402: The GHC API, decoupled from GHC versions

Safe HaskellNone
LanguageHaskell2010

NameShape

Synopsis

Documentation

data NameShape #

A NameShape is a substitution on Names that can be used to refine the identities of a hole while we are renaming interfaces (see RnModIface). Specifically, a NameShape for ns_module_name A, defines a mapping from {A.T} (for some OccName T) to some arbitrary other Name.

The most intruiging thing about a NameShape, however, is how it's constructed. A NameShape is *implied* by the exported AvailInfos of the implementor of an interface: if an implementor of signature H exports M.T, you implicitly define a substitution from {H.T} to M.T. So a NameShape is computed from the list of AvailInfos that are exported by the implementation of a module, or successively merged together by the export lists of signatures which are joining together.

It's not the most obvious way to go about doing this, but it does seem to work!

NB: Can't boot this and put it in NameShape because then we start pulling in too many DynFlags things.

emptyNameShape :: ModuleName -> NameShape Source #

Create an empty NameShape (i.e., the renaming that would occur with an implementing module with no exports) for a specific hole mod_name.

mkNameShape :: ModuleName -> [AvailInfo] -> NameShape Source #

Create a NameShape corresponding to an implementing module for the hole mod_name that exports a list of AvailInfos.

extendNameShape :: HscEnv -> NameShape -> [AvailInfo] -> IO (Either SDoc NameShape) Source #

Given an existing NameShape, merge it with a list of AvailInfos with Backpack style mix-in linking. This is used solely when merging signatures together: we successively merge the exports of each signature until we have the final, full exports of the merged signature.

What makes this operation nontrivial is what we are supposed to do when we want to merge in an export for M.T when we already have an existing export {H.T}. What should happen in this case is that {H.T} should be unified with M.T: we've determined a more *precise* identity for the export at OccName T.

Note that we don't do unrestricted unification: only name holes from ns_mod_name ns are flexible. This is because we have a much more restricted notion of shaping than in Backpack'14: we do shaping *as* we do type-checking. Thus, once we shape a signature, its exports are *final* and we're not allowed to refine them further,

nameShapeExports :: NameShape -> [AvailInfo] Source #

The export list associated with this NameShape (i.e., what the exports of an implementing module which induces this NameShape would be.)

substNameShape :: NameShape -> Name -> Name Source #

Given a Name, substitute it according to the NameShape implied substitution, i.e. map {A.T} to M.T, if the implementing module exports M.T.

maybeSubstNameShape :: NameShape -> Name -> Maybe Name Source #

Like substNameShape, but returns Nothing if no substitution works.