More partial application warnings

[stedolan]

As #7221 points out, warning 5 (on maybe-unintended partial applications) would be useful in more situtations, e.g:

# let f a b = a + b
  let g x = x + 1
  let _ = g (f 1);;
Error: This expression has type int -> int
       but an expression was expected of type int

This patch makes it report this instead:

# let f a b = a + b
    let g x = x + 1
    let _ = g (f 1);;
Warning 5: this function application is partial,
maybe some arguments are missing.
Error: This expression has type int -> int
       but an expression was expected of type int

Concretely, whenever a type_expect fails with a Expr_type_clash on an application which is partial, warning 5 gets reported in addition to the error.

[stedolan]

Any takers for review? This is not a complicated change. (@gasche / @lpw25 / @trefis ?)

[gasche]

I reviewed the code and believe it is correct (but see comment below). I am inclined to say that the feature is a good idea, and I trust @stedolan and @whitequark's judgment, but I also know that this is a general area @trefis has given more though about, so he might want to give his opinion as well (if he has time).

[gasche]

I agree that the patch seems correct, but I worry (slightly) about trashing of backtraces that may occur during check_partial_application (it uses Format, etc.). We could use the get_raw_backtrace / raise_with_backtrace dance to avoid this, but it makes the code much less pleasant. We could almost use Misc.try_finally, but unfortunately its exceptionally clause does not take the exception as an input, so this is not quite possible. My perfectionist self would recommend tweaking try_finally in this way and using it here, or defining a new Misc combinator for this.

[stedolan]

There's now a shiny new Misc.reraise_preserving_backtrace

(I consider the behaviour of overwriting the global backtrace to be a bug, hopefully one day we can delete reraise_preserving_backtrace as it will just be how all reraises act)

[trefis]

I'm not sure I have really thought more than anyone else about this. But anyway, I like this proposal :)

[gasche]

Approved, but see my non-trivial code comment.

[gasche]

Thanks! Merging.