[plt-dev] coding ideas from JaneStreet

On Tue, Apr 13, 2010 at 1:09 PM, Robby Findler
<robby at eecs.northwestern.edu> wrote:
>>
>> What if we change the contract to:
>>
>> ((list/c any/c any/c) . -> . any/c)
>>
>> Now the error case is unreachable, but `match' doesn't know that.
>> This also suggests that almost every single use of `match-let' should
>> trigger a warning.
>
> We would presumably just follow ml in this case.

I don't know what this means.  For example, the OCaml code:

match (x : (int,int)) with
  (7,z) -> ...
| (_,z) -> ...

is complete, and therefore no warning would be issued.  If you use
general list patterns in OCaml:

match (x : int list) with
  [7,z] -> ...
| [_,z] -> ...

you'll get a warning about pattern incompleteness.  I don't think this
warning is useful, given the contract I specified.

Further, ML checks the completeness of patterns in `let' using the
type system, which `match' can't do.  For example, this code:

(match-let ([(list x ...) (map add1 some-list])
  ...)

clearly doesn't cover the case where `map' doesn't produce a list.
That's impossible, but `match-let' doesn't know that, so it would have
to issue a warning.  Would you find that warning useful?  I don't
think I would.
-- 
sam th
samth at ccs.neu.edu