Re: [Caml-list] exception Foo = Bar.Baz

[Gabriel Scherer <gabriel.scherer@gmail.com>]

You can already eliminate one of the two comparisons if both A are
statically known to be the same. Moreover, cross-module optimization
information is likely to give you the actual definition of both A and
B -- if they are just defined with either "exception Foo" or
"exception Bar = Baz" as a structure item of the compilation unit
level, their identity will be known. In the complex cases where it is
not (functor parameters, inclusion of a first-class module, etc.), the
cost corresponds to some extra expressiveness for the user that I
think is worth it.

On Wed, Jul 27, 2016 at 4:07 AM, Alain Frisch <alain.frisch@lexifi.com> wrote:
> On 26/07/2016 18:32, Gabriel Scherer wrote:
>>
>> I would see some use in being able to rename variant constructors, see
>>
>>   MPR#7102: Ability to re-export a variant definition with renamed
>> constructors?
>>   http://caml.inria.fr/mantis/view.php?id=7102
>
>
> The fact that an expression "A = B" can evaluate to true can be confusing
> and really change the interpretation of constructors as atoms.  It's already
> the case for exceptions/extension constructors, but I'd prefer to have less
> of it than more.
>
> Thinking more about it, even for exceptions, not allowing different name for
> the same constructor could be useful to optimize cases such as:
>
>
>   try
>
>     try
>      ....
>      raise A
>      ...
>     with B -> ...
>
>   with A -> ....
>
>
> If the compiler could assume that A and B were different constructors, it
> could compile the "raise" to a static jump to the handler for A, removing
> two dynamic comparisons with exception slots (and their associated memory
> accesses).
>
>
> Alain