www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - dub test errors during linking for struct with uninstantiated

reply Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
When a module containing


module nxt.rational;

Rational!(I1) rational(I1, I2)(I1 , I2)
{
     return typeof(return)();
}

struct Rational(Int)
{
     bool opEquals(Rhs)(Rhs _) {}
}

 nogc unittest
{
     auto _ = rational(1, 2);
}


is used as the sole module of a dub project then `dub test` fails 
as

.dub/build/phobos-next-test-library-unittest-linux.posix-x86_64-dmd_2092-06C9DAF60B64114484B3FE9AC9866D5F/phobos-next-test-library.o:src/nxt/rational.d:_D40TypeInfo_S3nxt8rational__T8Ra
ionalTiZQm6__initZ: error: undefined reference to
'_D6object10_xopEqualsFMxPvMxQeZb'

When I remove

     bool opEquals(Rhs)(Rhs _) {}

the error goes away.

I'm suspecting this is a bug in dmd that fails to correctly check 
if a struct has its templated `opEquals` instantiated.

For a complete minimal sample project see

https://github.com/nordlow/phobos-next/tree/master/rational.reduced
Jun 21 2020
next sibling parent Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Sunday, 21 June 2020 at 12:42:11 UTC, Per Nordlöw wrote:

 When a module containing


BTW this fails for dmd stable and dmd nightly but works for ldc 
1.22.0.
Jun 21 2020
prev sibling parent reply kinke <noone nowhere.com> writes:
On Sunday, 21 June 2020 at 12:42:11 UTC, Per Nordlöw wrote:

 I'm suspecting this is a bug in dmd that fails to correctly 
 check if a struct has its templated `opEquals` instantiated.


I'd suggest taking dub out of the equation then, so that you can 
file a DMD issue once you can confirm it. Your sample works with 
`-main -unittest` on run.dlang.io, so it seems to depend on how 
the test executable is built.
Jun 21 2020
parent reply kinke <noone nowhere.com> writes:
On Sunday, 21 June 2020 at 16:58:43 UTC, kinke wrote:

 Your sample works with `-main -unittest` on run.dlang.io


After quickly looking at your list of compiler flags in dub.sdl, 
the problem manifests when adding `-preview=in`. Please file an 
issue.
Jun 21 2020
next sibling parent Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Sunday, 21 June 2020 at 17:05:11 UTC, kinke wrote:

 On Sunday, 21 June 2020 at 16:58:43 UTC, kinke wrote:

 Your sample works with `-main -unittest` on run.dlang.io


 After quickly looking at your list of compiler flags in 
 dub.sdl, the problem manifests when adding `-preview=in`. 
 Please file an issue.


Thanks a lot! I'll do so.
Jun 21 2020
prev sibling parent reply Per =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Sunday, 21 June 2020 at 17:05:11 UTC, kinke wrote:

 Please file an issue.


Done: https://issues.dlang.org/show_bug.cgi?id=20968
Jun 21 2020
parent Boris Carvajal <boris2.9 gmail.com> writes:
On Sunday, 21 June 2020 at 22:48:20 UTC, Per Nordlöw wrote:

 On Sunday, 21 June 2020 at 17:05:11 UTC, kinke wrote:

 Please file an issue.


 Done: https://issues.dlang.org/show_bug.cgi?id=20968


You can get more info compiling with "-verrors=spec"

"(spec:1) Error: template rational.Rational!int.Rational.opEquals 
cannot deduce function from argument types 
!()(const(Rational!int)) const, candidates are:"

 From reading the docs for function opEquals 
(https://dlang.org/spec/operatoroverloading.html#equals).
It shows a templated version with a const modifier.
If you change the function signature to:

bool opEquals(Rhs)(Rhs _) const {return true;} // Note the 
'const', and putting some return back because dustmite removed it

The program links fine.

So it seems a compiler error is gagged that should be exposed and 
also could be improved.
Jun 22 2020