[erlang-questions] Maps branch and disclaimers

[Chris King]

On Thu, 31 Oct 2013 06:58:18 -0400, Anthony Ramine <n.oxyde@REDACTED>  
wrote:

> Write an unary function with 15 (might need more) different clauses  
> f(a1) -> ...; ...; f(a15) -> ....
>
> Dialyzer will infer atom() for the argument type because large type  
> unions are broadened to make the various type unifications happen in a  
> reasonable time, if I understand this correctly.

Yes, Dialyzer necessarily approximates types.  And it, by design, fails on  
the false negative side of things: it can infer that functions accept  
types which actually make them crash.

This has no bearing on the *actual* type signature written by a human to  
communicate to other humans.  These have meaning outside of Dialyzer.   
Unfortunately without a semantic definition of these signatures (which I  
sadly can't find!) we're both grasping at straws.

The fact that Dialyzer is *able* (with -Wunderspecs) to warn that a type  
is too broad tells me at least that the semantics of type signatures is  
ambiguous.  Whether this option is a "here's a more sound analysis we're  
working on" option, or a "this is for weird people who insist that types  
imply success" is unclear.