The compiler complains about `cannot form tuple of tuples` whenever I try to put an AliasSeq in a UDA and try to use it.

You would expect the compiler to expand it. Is this a bug?

---

import std.meta;

enum A; enum B;

@AliasSeq!(A,B) // <-- Error: cannot form tuple of tuples
struct Foo {}

pragma(msg, __traits(getAttributes, Foo)); // <- must be present (error happens only when queried)

void main() {}

---

link to run.dlang.io: https://run.dlang.io/gist/cd5cd94d8ce5327e8a7d5ad77f1d15b8

On Sunday, 10 March 2019 at 13:20:12 UTC, Sebastiaan Koppe wrote:
> The compiler complains about `cannot form tuple of tuples` whenever I try to put an AliasSeq in a UDA and try to use it.
>
> You would expect the compiler to expand it. Is this a bug?
>
> ---
>
> import std.meta;
>
> enum A; enum B;
>
> @AliasSeq!(A,B) // <-- Error: cannot form tuple of tuples
> struct Foo {}
>
> pragma(msg, __traits(getAttributes, Foo)); // <- must be present (error happens only when queried)
>
> void main() {}
>
> ---
>
> link to run.dlang.io: https://run.dlang.io/gist/cd5cd94d8ce5327e8a7d5ad77f1d15b8

It looks like a bug, a "reject-valid" one.
Try the same code with

    enum A = 0;

and it work, despite of B being still opaque. The problem may be related to the fact the enum A in the orginal code is opaque and has not type.

I've seen a related issue lately.

On Sunday, 10 March 2019 at 13:41:32 UTC, Basile B. wrote:
> On Sunday, 10 March 2019 at 13:20:12 UTC, Sebastiaan Koppe wrote:
>> The compiler complains about `cannot form tuple of tuples` whenever I try to put an AliasSeq in a UDA and try to use it.
>>
>> You would expect the compiler to expand it. Is this a bug?
>>
>> ---
>>
>> import std.meta;
>>
>> enum A; enum B;
>>
>> @AliasSeq!(A,B) // <-- Error: cannot form tuple of tuples
>> struct Foo {}
>>
>> pragma(msg, __traits(getAttributes, Foo)); // <- must be present (error happens only when queried)
>>
>> void main() {}
>>
>> ---
>>
>> link to run.dlang.io: https://run.dlang.io/gist/cd5cd94d8ce5327e8a7d5ad77f1d15b8
>
> It looks like a bug, a "reject-valid" one.
> Try the same code with
>
>     enum A = 0;
>
> and it work, despite of B being still opaque. The problem may be related to the fact the enum A in the orginal code is opaque and has not type.
>
> I've seen a related issue lately.

It was 19605. But the relationship is not so obvious...

On Sunday, 10 March 2019 at 13:41:32 UTC, Basile B. wrote:
> It looks like a bug, a "reject-valid" one.
> Try the same code with
>
>     enum A = 0;
>
> and it work, despite of B being still opaque. The problem may be related to the fact the enum A in the orginal code is opaque and has not type.

Thanks. The example I posted was a reduced case, but I can still use your trick to fool the compiler.

The following is more real:

---

import std.meta;

enum A; enum B; enum Dummy = 0; // <- Dummy needs to be 0

struct Param(alias T) {}

@AliasSeq!(Dummy,Param!A,Param!B) // <- Here Dummy is needed
struct Foo {}

pragma(msg, __traits(getAttributes, Foo));

void main() {}

---

I have created an issue: https://issues.dlang.org/show_bug.cgi?id=19728

On Sunday, 10 March 2019 at 16:05:19 UTC, Sebastiaan Koppe wrote:
> On Sunday, 10 March 2019 at 13:41:32 UTC, Basile B. wrote:
>> It looks like a bug, a "reject-valid" one.
>> Try the same code with
>>
>>     enum A = 0;
>>
>> and it work, despite of B being still opaque. The problem may be related to the fact the enum A in the orginal code is opaque and has not type.
>
> Thanks. The example I posted was a reduced case, but I can still use your trick to fool the compiler.
>
> The following is more real:
>
> ---
>
> import std.meta;
>
> enum A; enum B; enum Dummy = 0; // <- Dummy needs to be 0
>
> struct Param(alias T) {}
>
> @AliasSeq!(Dummy,Param!A,Param!B) // <- Here Dummy is needed
> struct Foo {}
>
> pragma(msg, __traits(getAttributes, Foo));
>
> void main() {}
>
> ---
>
> I have created an issue: https://issues.dlang.org/show_bug.cgi?id=19728

Yes I see. I've refined a bit the test case and maybe I'll took a look this week.

On Sunday, 10 March 2019 at 16:46:43 UTC, Basile B. wrote:
> Yes I see. I've refined a bit the test case and maybe I'll took a look this week.

Cool. Is it normal to create a testcase that doesn't depend on phobos? I suppose it is needed for it to be included in dmd's testcases.

On Sunday, 10 March 2019 at 17:04:20 UTC, Sebastiaan Koppe wrote:
> On Sunday, 10 March 2019 at 16:46:43 UTC, Basile B. wrote:
>> Yes I see. I've refined a bit the test case and maybe I'll took a look this week.
>
> Cool. Is it normal to create a testcase that doesn't depend on phobos? I suppose it is needed for it to be included in dmd's testcases.

Yes.