Hi,

While working on my openmethods library, I came across a gap that may be worth filling.

I have this kind of code:

// in openmethods.d
string registerMethods() { ... }

// in code using the lib
import openmethods;
mixin(registerMethods);

When I add -H, the interface file takes precedence over the source file, but the body of registerMethods() is removed. I easily found a workaround - make the function a template (see here https://github.com/jll63/openmethods.d/commit/6a7bf93932bebc53fca368e2280c71a0cbbf3c3b) but it's a bit of a hack. I also had to remove a couple of ctors that was there only because of my ignorance at the time.

I tried pragma(inline) but it doesn't preserve the body. Maybe it should, because you cannot inline without the source. Or maybe we need another pragma, I couldn't find a nice name but something along pragma(interface) or pragma(keepBody).

On Friday, 4 August 2017 at 13:55:29 UTC, Jean-Louis Leroy wrote:
> Hi,
>
> While working on my openmethods library, I came across a gap that may be worth filling.
>
> [...]

make the function return auto.
Then you'll have the body.

On Friday, 4 August 2017 at 16:33:06 UTC, Stefan Koch wrote:
> On Friday, 4 August 2017 at 13:55:29 UTC, Jean-Louis Leroy wrote:
>> Hi,
>>
>> While working on my openmethods library, I came across a gap that may be worth filling.
>>
>> [...]
>
> make the function return auto.
> Then you'll have the body.

Thanks. This is, hmmm, surprising ;-)

On Saturday, 5 August 2017 at 00:45:51 UTC, Jean-Louis Leroy wrote:
> On Friday, 4 August 2017 at 16:33:06 UTC, Stefan Koch wrote:
>> On Friday, 4 August 2017 at 13:55:29 UTC, Jean-Louis Leroy wrote:
>>> Hi,
>>>
>>> While working on my openmethods library, I came across a gap that may be worth filling.
>>>
>>> [...]
>>
>> make the function return auto.
>> Then you'll have the body.
>
> Thanks. This is, hmmm, surprising ;-)

for the auto return type to work the compiler needs to infer it.
in order to infer it you need the body ;)

On Saturday, 5 August 2017 at 01:05:41 UTC, Stefan Koch wrote:
> On Saturday, 5 August 2017 at 00:45:51 UTC, Jean-Louis Leroy wrote:
>> On Friday, 4 August 2017 at 16:33:06 UTC, Stefan Koch wrote:
>>> On Friday, 4 August 2017 at 13:55:29 UTC, Jean-Louis Leroy wrote:
>>>> Hi,
>>>>
>>>> While working on my openmethods library, I came across a gap that may be worth filling.
>>>>
>>>> [...]
>>>
>>> make the function return auto.
>>> Then you'll have the body.
>>
>> Thanks. This is, hmmm, surprising ;-)
>
> for the auto return type to work the compiler needs to infer it.
> in order to infer it you need the body ;)

I see...OTOH the compiler could resolve the return type when generating the .di file, then the body would not be needed anymore. I changed my registerMethods back from a template to an auto function but I feel that I am in undocumented territory here.

On 2017-08-04 15:55, Jean-Louis Leroy wrote:

> When I add -H, the interface file takes precedence over the source file,
> but the body of registerMethods() is removed. I easily found a
> workaround - make the function a template (see here
> https://github.com/jll63/openmethods.d/commit/6a7bf93932bebc53fca368e2280c71a0cbbf3c3b)
> but it's a bit of a hack.

I've used that workaround as well. If you still want to go with the template, I recommend making the template argument empty:

string registerMethods()(string moduleName = __MODULE__)

Otherwise you'll get a new template instantiation for each call to the function where the module is different.

-- 
/Jacob Carlborg

On Saturday, 5 August 2017 at 02:12:33 UTC, Jean-Louis Leroy wrote:
>
> I see...OTOH the compiler could resolve the return type when generating the .di file, then the body would not be needed anymore. I changed my registerMethods back from a template to an auto function but I feel that I am in undocumented territory here.

No, the header-generation is done before sema.
Therefore it cannot reslove anything.
That is not going to change.