Re: [Chicken-hackers] incorrect warning during compilation

[Kristian Lein-Mathisen]

Thanks for you feedback Felix,

I talked to Peter on #chicken and he wasn't happy about another foreign-type. I suppose I agree with him, there are quite a few already and maybe it'll just be confusing: having a "c-string" and a "string" foreign type would probably do more harm than gain.

I decided to do a (c-pointer char) and wrap the string argument in (location ...) which isn't all that bad. I couldn't use scheme-pointer because my C-function argument is already declared as char* and not void*.

And I guess string->blob/shared isn't all that useful if no other Schemes support it and we have plenty of other workarounds. I never really got blobs anyway, when do you not just want to use a string instead? They print so much nicer than blobs, even with binary data that's escaped with \x!

K.

On Fri, Jun 27, 2014 at 11:05 AM, Felix Winkelmann <address@hidden> wrote:

> Oh, I see. I though it was meant to be used interchangeably. I can't use
> string->blob, though, because that will copy the data and I need to modify
> it from C. What foreign type should I use for this sort of thing?
>
> - (c-pointer char) with (location "string") works, but is a lot of typing
>
> - scheme-pointer seems to work, but:
>
> requires type cast in C:
>
> (print ((foreign-lambda* c-string ((scheme-pointer x)) "((char*)x)[0] = 65;
> return(x);") "xBC"))
>
> allows all types, I just want blob/string:
>
> (print ((foreign-lambda* c-string ((scheme-pointer x)) "((char*)x)[0] = 65;
> return(x);") 45.0))
>
> is there a "string" foreign type that works like "blob", that provides a
> (char*) argument and checks that argument is a string? Or is there a
> string->blob/shared procedure anywhere?

You could also keep everything as it is and ignore the warning, but I
understand that it isn't the cleanest solution. I'm not aware of any
documented interchangability between strings and blobs, it is an
implementation artefact. "string->blob/shared" would make sense but is
not possible in the current implementation (srfi-4 vectors have an
indirection to the raw data buffer, there this conversion only changes
the wrapper object, but strings and blobs contain the data without the
indirection.)

So, I can't think of a solution that answers all of your questions.
I would go with the scheme-pointer.

We could add another foreign type, that accepts strings but doesn't
copy. I'll look into that.


felix