Re: wayland client in Guile without libwayland-client

[Matt Wette <matt.wette@gmail.com>]

On 11/18/22 8:26 PM, Damien Mattei wrote:
> hello,
>
> it is just a question about semantic and aside the subject of project, and
> related to another subject of internal definitions that Linus is talking
> about:
>
> On Fri, Nov 18, 2022 at 5:55 PM Matt Wette<matt.wette@gmail.com>  wrote:
>
>> Here is an example of an auto-coded event decoder:
>>
>>         (lambda (obj-id bv ix cm)
>>           "event decoder for global"
>>           (let*-values
>>             (((name ix) (dec-u32 bv ix))
>>              ((interface ix) (dec-string bv ix))
>>              ((version ix) (dec-u32 bv ix)))
>>             (values obj-id name interface version)))
>>
>>
>>
> i was just thinking how i would write the procedure above in Scheme+ (
> https://github.com/damien-mattei/Scheme-PLUS-for-Guile  ) that has feature
> for multiple values.
> this code could be rewritten like that:
> (lambda (obj-id bv ix cm)
>           "event decoder for global"
>           {(name ix) <+ (dec-u32 bv ix)}
>           {(interface ix) <+ (dec-string bv ix)}
>           {(version ix) <+ (dec-u32 bv ix)}
>             (values obj-id name interface version))
>
> note that <+ is an "operator" that allows multiple values definitions.
>
> but it would fails to compile because of multiple redefinition of ix
> variable:
>
> scheme@(guile-user)> (use-modules (Scheme+))
> scheme@(guile-user)> (lambda (obj-id bv ix cm)
>           "event decoder for global"
>           {(name ix) <+ (dec-u32 bv ix)}
>           {(interface ix) <+ (dec-string bv ix)}
>           {(version ix) <+ (dec-u32 bv ix)}
>             (values obj-id name interface version))
> While compiling expression:
> Syntax error:
> unknownfile:7:0: invalid or duplicate identifier in definition in form
> (lambda (obj-id bv ix cm) "event decoder for global" (<+ (name ix) (dec-u32
> bv ix)) (<+ (interface ix) (dec-string bv ix)) (<+ (version ix) (dec-u32 bv
> ix)) (values obj-id name interface version))
>
> a longer and a bit unsightly solution could be:
>
> (lambda (obj-id bv ix cm)
>           "event decoder for global"
>           (declare name interface version)
>           {(name ix) <v (dec-u32 bv ix)}
>           {(interface ix) <v (dec-string bv ix)}
>           {(version ix) <v (dec-u32 bv ix)}
>             (values obj-id name interface version))
>
> that would be correct:
>
> (note <v instead of <+ is a setter! for already defined values and declare
> just define variables initialised by default with '() in Scheme+)
>
> scheme@(guile-user)> (lambda (obj-id bv ix cm)
>           "event decoder for global"
>           (declare name interface version)
>           {(name ix) <v (dec-u32 bv ix)}
>           {(interface ix) <v (dec-string bv ix)}
>           {(version ix) <v (dec-u32 bv ix)}
>             (values obj-id name interface version))
> ;;; <stdin>:16:24: warning: possibly unbound variable `dec-u32'
> ;;; <stdin>:17:29: warning: possibly unbound variable `dec-string'
> $3 = #<procedure 149009320 at <unknown port>:13:0 (obj-id bv ix cm)>
>
> the first example fails not because of my Scheme+ language extension but
> because in scheme redefinition is only allowed at toplevel or in the REPL.
>
> Would it not be a good idea to allow redefinitions in Scheme and Guile not
> only at toplevel or REPL but in procedure definitions and lambda, in their
> bodies?
>
> personaly i can not see any "bad" side effect or consequences... of course
> it relax the possibility of code error by unchecking a redefinition of
> variable but it is like in C++ where you can define private or protected
> variables but even with protective language feature one can wrote bad
> code...
>
> Best regards,
>
> Damien

Another approach might be a state monad (guix has code for this).   I am not sure
about other implementations, but I am reasonably confident that the Guile compiler
can turn my values-heavy Scheme into efficient bytecode.  I'm also working to
keep allocations down also (e.g., reusing bytevector buffers).

Thanks for your interest.
Matt