Is this a bug? Should the DMD compiler ever give an internal error?

import std.stdio;
import std.boxer;

void main() {
	Object[] l;
	l ~= cast(Object) box(1);
	int a = unbox!(int)(cast(Box) l[0]);
	writefln("value %d", a);
}

>dmd tmp.d c:\dmd\src\phobos\std\boxer.d
tmp.d(6): e2ir: cannot cast from Box to object.Object
tmp.d(7): e2ir: cannot cast from object.Object to Box
Internal error: e2ir.c 267


Thanks,
  Bradley

"Bradley Smith" <digitalmars-com@baysmith.com> wrote in message news:e7hvil$1ffi$1@digitaldaemon.com...
> Is this a bug? Should the DMD compiler ever give an internal error?

Considering it's in alpha, the compiler is allowed to have some slack ;) But no, it's not _supposed_ to have internal errors.  Any internal error is a bug.

> import std.stdio;
> import std.boxer;
>
> void main() {
> Object[] l;
> l ~= cast(Object) box(1);

You can't cast a box to Object; box is a struct, not a class.  If you want an array of boxes, hava a look at boxArray().

> int a = unbox!(int)(cast(Box) l[0]);
> writefln("value %d", a);
> }
>
> >dmd tmp.d c:\dmd\src\phobos\std\boxer.d
> tmp.d(6): e2ir: cannot cast from Box to object.Object
> tmp.d(7): e2ir: cannot cast from object.Object to Box
> Internal error: e2ir.c 267

I didn't mean to be overly critical of the compiler. I wasn't sure how to interpret the "Internal error" message, and I just wanted to confirm that the objective was to not have any internal errors.

What is the proper procedure for reporting a bug?

Thanks for the pointer about Box and Object. I'm still learning how D handles data.

Thanks,
  Bradley


Jarrett Billingsley wrote:
> "Bradley Smith" <digitalmars-com@baysmith.com> wrote in message news:e7hvil$1ffi$1@digitaldaemon.com...
>> Is this a bug? Should the DMD compiler ever give an internal error?
> 
> Considering it's in alpha, the compiler is allowed to have some slack ;) But no, it's not _supposed_ to have internal errors.  Any internal error is a bug.
> 
>> import std.stdio;
>> import std.boxer;
>>
>> void main() {
>> Object[] l;
>> l ~= cast(Object) box(1);
> 
> You can't cast a box to Object; box is a struct, not a class.  If you want an array of boxes, hava a look at boxArray().
> 
>> int a = unbox!(int)(cast(Box) l[0]);
>> writefln("value %d", a);
>> }
>>
>>> dmd tmp.d c:\dmd\src\phobos\std\boxer.d
>> tmp.d(6): e2ir: cannot cast from Box to object.Object
>> tmp.d(7): e2ir: cannot cast from object.Object to Box
>> Internal error: e2ir.c 267 
> 
> 

"Bradley Smith" <digitalmars-com@baysmith.com> wrote in message news:449CF0E7.3020107@baysmith.com...
>I didn't mean to be overly critical of the compiler. I wasn't sure how to interpret the "Internal error" message, and I just wanted to confirm that the objective was to not have any internal errors.

Oh no, I didn't take it that way.  Don't worry about it.

> What is the proper procedure for reporting a bug?

The most proper is through the D bugzilla: http://d.puremagic.com/bugzilla/index.cgi

You'll set up an account and then you can add the bug using the "Enter a new issue" link.  As far as I can find, this issue hasn't been added to bugzilla yet.

> Thanks for the pointer about Box and Object. I'm still learning how D handles data.

Sure.  Just wondering what you're trying to store?  There might be a simpler way to store it, whatever it is; I've personally never had to use boxes, though maybe you're coming from a dynamically typed background.

Jarrett Billingsley wrote:

> Sure.  Just wondering what you're trying to store?  There might be a simpler way to store it, whatever it is; I've personally never had to use boxes, though maybe you're coming from a dynamically typed background. 
> 

I'm experimenting with converting some Java code to D, and the code stores int, boolean, and String data in a list through the Integer and Boolean wrapper objects. Using the D boxer is the only way I've found to  create a heterogeneous array which includes primitive data elements.

Perhaps the need for boxing could be eliminated by making the code more D-like, but at this point, I'm only trying a straight translation.

  Bradley

Bradley Smith wrote:
> Jarrett Billingsley wrote:
> 
>> Sure.  Just wondering what you're trying to store?  There might be a simpler way to store it, whatever it is; I've personally never had to use boxes, though maybe you're coming from a dynamically typed background.
> 
> 
> I'm experimenting with converting some Java code to D, and the code stores int, boolean, and String data in a list through the Integer and Boolean wrapper objects. Using the D boxer is the only way I've found to  create a heterogeneous array which includes primitive data elements.
> 
> Perhaps the need for boxing could be eliminated by making the code more D-like, but at this point, I'm only trying a straight translation.
> 
>   Bradley

Well, if it only ever works with those three types, you could go with a Var struct.

# enum VarType { Int, Bool, String }
#
# struct Var {
#   VarType type ;
#
#   union {
#     bool   b ;
#     int    i ;
#     char[] s ;
#   }
# }
#
# Var[] array;

You can simplify creation of Var structs with a set of static call operators, like such:

# struct Var {
#   // ...
#   static Var opCall (bool value) {
#     Var var ;
#
#     var.type = VarType.Bool ;
#     var.b    = value        ;
#
#     return var;
#   }
# }

While it isn't /the/ most elegant solution possible, its tried and true, and pretty effective for definite sets of types.

-- Chris Nicholson-Sauls