Reputation: 709
I have a third-party C library that defines a struct similar to:
struct myStruct {
int a;
int b;
char str1[32];
char str2[32];
};
And a function that takes a pointer to this struct and populates it. I need my Perl6 native call to provide that struct, and then read the results.
So far I've got the struct defined in Perl6 as:
class myStruct is repr('CStruct') {
has int32 $.a;
has int32 $.b;
has Str $.str1; # Option A: This won't work as Perl won't know what length to allocate
has CArray[uint8] $.str2; # Option B: This makes more sense, but again how to define length?
# Also, would this allocate the array in place, or
# reference an array that is separately allocated (and therefore not valid)?
}
And a native call like:
sub fillStruct(myStruct) is native('test_lib') { ... } my $struct = myStruct.new(); fillStruct($struct); # Gives a seg fault with any variation I've tried so far
How can I make this work?
Upvotes: 7
Views: 383
Reputation: 324
As of July 2020, you should be able to something like this:
sub setCharArray($a, $s, $l, $c is rw) {
die 'Too many chars!' unless $s.chars <= $l;;
$c = $s if $c;
my $chars = $a.encode;
$a[$_] = $chars[$_] for ^$chars.elems;
$a[$chars.elems] = 0 unless $s.elems == 128;
}
class A repr<CStruct> is export {
HAS uint8 @!myString[128] is CArray;
state Str $cached;
method myString is rw {
Proxy.new:
FETCH => sub($) {
return $cached if $cached;
$cached = Buf.new(@!myString).decode
},
STORE => $, Str $s is raw {
setCharArray(@!myString, $s, 128, $cached);
}
}
}
Let me explain:
The "HAS" declarator to define static size elements has been in NativeCall for a bit, so that's not the experimental portion. It's the "method myString" that is the tricky part. It allows the consumer of class A to set and get from the @!myString attribute as if it were a proper attribute, rather than an array.
If the data in @!myString is first being read from C, then the $cache state variable will be empty. The Str object is then created via a decoded Buf and returned. It is the hope that the complexity seen in the object implementation is then hidden from the user so that:
my $a = c_function_that_returns_A();
$a.myString;
...will work as one would expect, and that similarly:
$a.myString = 'Crab';
...will also work with no problem.
It is unfortunate that a helper function like setCharArray() needs to iterate to set @!myString, but hopefully that will change, in the future.
IMPORTANT CAVEAT -- This implementation assumes that changes to @!myString are limited to the Raku side once the object is allocated, otherwise once set, the $cached value will mask them. I don't really see a way around that, at the moment, unless you want to spend the cycles to create a new Str a-fresh every time you need to access @!myString.
UPDATE Heh - Code had a minor but in the return statement. It's been fixed.
Upvotes: 1
Reputation: 709
As others have said, there does not appear to be any way to achieve this at present.
I've resorted to defining a new C function(s) as a workaround. The function effectively acts as an accessor method returning just the fields that I need as discrete NativeCall-friendly pointers.
Hopefully the community will get to implementing proper support for this case at some point.
Upvotes: 2
Reputation: 3888
At the time of writing, this doesn't seem to be handled.
As a workaround, my take would be to let a macro generate 32 int8
, adequately placing field names.
Upvotes: 1