sb-alien, cl:boolean, and the C _Bool type

[Attila Lendvai]

1.0.55 x86-64

(define-alien-type-translator boolean (&optional (bits sb!vm:n-word-bits))
  ...)

but on the C side:

typedef int boolean;

so, the C side is 32 bits, while the default size for the type translator is 64 bits.

this is very misleading and results in e.g.:

(setf (sb-alien:extern-alien "lose_on_corruption_p" boolean) t)

for a user who doesn't look twice it looks completely fine, but it silently corrupts memory.

[Attila Lendvai (attila-lendvai)]

err, sorry, scratch that. it's due to something i call, and that used to work, but not anymore.

CL-USER> (sb-alien:extern-alien "lose_on_corruption_p" boolean)
T
CL-USER> (sb-alien:extern-alien "internal_errors_enabled" boolean)
T
CL-USER> (setf (sb-alien:extern-alien "lose_on_corruption_p" boolean) t)
T
CL-USER> (sb-alien:extern-alien "internal_errors_enabled" boolean)
NIL

i guess the way boolean type is given to extern-alien results in a wider setf than what C uses for booleans... any hints on how to do it the right way?

this is on linux x64.

[Paul Khuong (pvk)]

C runtime: typedef int boolean;

boolean alien type definition: (define-alien-type-translator boolean (&optional (bits sb!vm:n-word-bits)) ...)

Use int, or pass the right number of bit (32) on x86-64.

[Attila Lendvai (attila-lendvai)]

any way to add a warning or some other hint when misusing cl:boolean the way I used with aliens?

this misuse was already suggested by someone else to me, seems like it's a prevalent mistake...

[Paul Khuong (pvk)]

This type isn't about the C runtime. If it were, it wouldn't be exported. There is no one boolean type in C. What there are are integer values that are used for their zero/non-zero-ness. In that context, there is no correct width.

It might hit more common cases right if we default the width that for int instead of long, but, either way, I don't see why it would follow any change to the runtime.

[Attila Lendvai (attila-lendvai)]

ok, i see that it can't really be made to Just Work.

but if that's the case, then how about some steps to make it less of a trap? e.g. how about making the size argument of the boolean type translator mandatory instead of defaulting to a value that is mostly wrong?

[Attila Lendvai (attila-lendvai)]

copying over the description of another bug that i once again unwittingly reported 4 years later.

on linux x64:

(sb-alien:alien-size cl:boolean) => 64
sizeof(_Bool) => 1

i would expect sb-alien either not to define a boolean type, or map it to C's _Bool.