[erlang-questions] Cannot make PropEr bitstring(B, U) generator to work

[Manolis Papadakis]

On 09/11/2012 06:36 PM, Motiejus Jakštys wrote:
> On Tue, Sep 11, 2012 at 4:40 PM, Manolis Papadakis <manopapad@REDACTED> wrote:
>> PropEr first tries to resolve the expression "bitstring(8, 1)" to a
>> generator, but that fails, because bitstring/2 is not a predefined
>> generator, and there's no custom generator with that signature declared
>> inside the module.
>
> But... According to the Book of Wisdom[1], bitstring(B, U) is there.

You're right, that should be there. Thanks for pointing it out. Crap, 
now I'll have to find where I've put that code.

> And binary(S) is not. I haven't seen it before.

Yeah, there's lots of other stuff on the API, mostly convenience 
functions and the like. You can find them all on the API documentation 
page [1].

>> PropEr then tries to find a type with that name, but that fails as well:
>> bitstring/2 is not a predifined type (the way to express the same thing in
>> the type language is <<_:8,_:_*1>>), and there's no local type with that
>> signature.
>
> Eh.. That would kinda do, but it raises a syntax error when used in
> ?FORALL context.

Unfortunately, many expressions that are valid in a type expression 
context are invalid Erlang terms, so there's no straightforward way to 
use them in ?FORALLs directly (that's why my example below uses a custom 
type as a thin wrapper for the binary type expression).

The types-in-place-of-generators feature is definitely a leaky 
abstraction [2].

>> Quick fix: If you're OK with using only unit values of either 8 or 1, then
>> you can use the predefined generator binary(Base) or bitstring(Base)
>> respectively. If you need other unit values, you'll have to write a custom
>> generator.
>
> Actually, I want binaries with length 0..1MiB-1:
>
>      ?LET(
>          Len,
>          ?LET(Len, integer(0, 1 bsl 20 - 1), Len),
>          {binary(Len), Len}
>      )
>
> So for this, binary(Number) is sufficient.
>
>> One might have expected something like the following to work:
>> -type my_bitstring() :: <<_:16, _:_*3>>.
>> prop_foo() -> ?FORALL(B, my_bitstring(), true).
>> But PropEr will currently reject this for unit values other than 1 and 8,
>> because I suck at predicting what features people will want to use, and
>> thought this was a safe one to skip.
>
> As said, I am only interested in full-byte binaries. So it was indeed
> safe for this case.
>
> Until 15 minutes ago how PropEr creates type instances was truly magic
> for me. Now Eureka! For ones that are curious how it works:
>
> x() ->
>      ?FORALL(S, integer(0, 1023), S < 100).
>
> translates to:
> x() ->
>      proper:forall(S, fun(S) -> integer(0, 1023) end)).
>
> So we have a closure which tries to execute the nonsense (that's what
> I thought for more than a year).
>
> After a lot of tracing I found out that this closure is executed in
> context of `proper_types` moduke, which *has* a function integer(Low,
> High). That's it. That function returnsa a nothing-magic tuple, and is
> able to proceed.
>
> And now answering my question. binary/1 is defined in proper_types. :)
> All makes sense now.
>
> [1]: http://proper.softlab.ntua.gr/papers/proper_types.pdf
>
> Regards,
> Motiejus Jakštys

[1] http://proper.softlab.ntua.gr/doc/
[2] http://www.joelonsoftware.com/articles/LeakyAbstractions.html