Re: Modifying self

[Philip Herron <philip.herron@embecosm.com>]

[-- Attachment #1.1: Type: text/plain, Size: 2992 bytes --]


On 05/08/2021 00:58, Mark Wielaard wrote:
> Hi,
>
> I am trying to get this program working:
>
> extern "C" { fn abort (); }
>
> pub struct H
> {
>   l: u32,
> }
>
> impl H
> {
>   fn p (&mut self) -> u32
>   {
>     self.l -= 1;
>     self.l
>   }
> }
>
> fn main ()
> {
>   let mut h = H { l: 11 }; 
>   let eleven = h.l; 
>   let ten = h.p (); 
>   if ten + 1 != eleven { unsafe { abort (); } } 
>   let h2 = H { l: ten }; 
>   if h.l != h2.l { unsafe { abort (); } } 
> }
>
> This doesn't currently compile:
>
> $ gcc/gccrs -Bgcc -g p.rs 
> p.rs:12:5: error: invalid left-hand side of assignment
>    12 |     self.l -= 1;
>       |     ^
>
> But this isn't too hard to solve:
>
> diff --git a/gcc/rust/resolve/rust-ast-verify-assignee.h b/gcc/rust/resolve/rust-ast-verify-assignee.h
> index aed01196f81..1e8988d47df 100644
> --- a/gcc/rust/resolve/rust-ast-verify-assignee.h
> +++ b/gcc/rust/resolve/rust-ast-verify-assignee.h
> @@ -75,6 +75,13 @@ public:
>        }
>    }
>  
> +  void visit (AST::PathInExpression &path) override
> +  {
> +    /* XXX do we need to check self is mutable? How?  */
> +    if (path.as_string () == "self")
> +      ok = true;
> +  }
> +
>  private:
>    VerifyAsignee (NodeId parent) : ResolverBase (parent), ok (false) {}
>  
> I am not sure whether this is a good implementation of the
> VerifyAsignee visitor for a PathInExpression. What exactly is the goal
> of this visitor?
>
> But with the above simple fix, it compiles! And it actually seems to
> mostly work. The implementation method is called, it gets its own
> field, substracts the value and correctly returns it!
>
> But... then it still aborts on the second check.  The method was
> supposed to adjust the given self (H), but it is not given a mutable
> reference, it gets a copy...
>
> On irc Philip suggested this is probably
> https://github.com/Rust-GCC/gccrs/issues/241
>
> But I must admit I don't fully understand what really needs to be done
> here or if the fact that this is a &mut self makes it different from a
> &self argument.
>
> Cheers,
>
> Mark
>
This is a fantastic test case. I think we should add this to the xfail
section. In terms of getting the mutable self working on methods, its a
little more complex since the compiler is very permissive here, where
the type system is meant to be able to coerce the reciever with a system
called autoderef which is not documented very well. This is something
that must be fixed as part of traits since you can end up with many
candidates for a path or method call and choosing the correct one is
very important to get right.

This missing feature is causing a deficiency in constant folding from
this issue: https://github.com/Rust-GCC/gccrs/issues/547

The verify lvalue looks good, this validation could really do with some
work down the line. I will reply to this thread when some movement is
made here on this.

--Phil



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 665 bytes --]