Haskell-prime Digest, Vol 2, Issue 58

[John Hughes]

    From: "Claus Reinke" <claus.reinke at talk21.com>

    let's go through 5.2 "Export Lists" to see what would be missing
    if we tried to replace the export list with a separation of a module
    into a public (exported) and a private (local) part:
    ...
    any other issues I missed here?

I feel unkeen.

One of the nice things about Haskell is that definitions can appear in any order. That makes it possible to gather a group of logically related definitions together, within a module. With your proposal, exported definitions and non-exported ones would have to be separated.

What would that mean in practice? Suppose I have a few exported functions and a collection of auxiliary functions that are used to implement them. I have two choices: either I put the exported definitions in the public section, and the remaining ones elsewhere in the private section, or I put everything in the private section with appropriate redeclarations in the public part -- exportedName = localExportedName or whatever. The first alternative has the disadvantages that logically related code is separated, and that the public section of the module may itself become quite large (since it contains full function definitions), making it hard to see at a glance what the exported names are. The second alternative has the disadvantage of introducing an indirection---finding the actual definition of an exported function becomes more difficult, because one must both scan the module for it, and first look up the public section to see what the private version is called. Neither alternative feels really attractive to me.

Today's export lists at least have the advantage that it is easy to see what is exported, even if changing the status of a definition from private to public is a little awkward (with edits in two places). With a tool like Haddock installed too, once gets a pretty good view of the interface---arguably better than one can get from a public module section. Perhaps, given that Haddock is available, a public modifier makes more sense than an explicit export list---although code browsing would then require running Haddock during development much more often than today, and potentially on syntactically incorrect or ill-typed modules.

Incidentally, the Erlang equivalent of Haddock, edoc, is distributed with the compiler, so that all Erlang installations include the tool, no matter what platform they're running on. Surely that's a prerequisite for adapting the language design on the assumption that tools of various kinds are available?

John