I have a template in Harmonia (code reduced):

struct array(T)
{
  T[] elements;
  ....
  struct ENUM // bidirectional enumerator
  {
    int opApply(int delegate(inout T) dg)  {  ...  }
  }

  ENUM items(bool forward) { ENUM w; w._forward = forward; w._a = elements;
return w; }
  ENUM forward() { return items(true); }
  ENUM backward() { return items(false); }
}

And its instantiation in other module:

array!(Doc.Brick*) bricks;

Since DMD 0.157(probably 0.156 too) compilation is yelding to:

OPTLINK :
C:\d\build\lib\release\view.obj(view)  Offset 03958H Record Type 0091
Error 1: Previous Definition Different :
__init_8harmonia5types46__T5arrayTPS8harmonia4html8document3Doc5BrickZ5array4ENUM

This is in Release mode, in Debug mode everything is fine.

I cannot understand how to solve this. Instantiation of array!(Doc.Brick*)
is in single place and
I have no idea what "Previous Definition Different" means here.

If pattern looks familiar to somebody - please let me know as Harmonia is broken now.

Andrew.

Working DMD version (for Harmonia) was DMD 0.150
Since then something was changed in the way how it links/compiles.

Andrew.

"Andrew Fedoniouk" <news@terrainformatica.com> wrote in message news:e47jup$21kb$1@digitaldaemon.com...
> Working DMD version (for Harmonia) was DMD 0.150
> Since then something was changed in the way how it links/compiles.
>
> Andrew.

Looks kind of like when the template name mangling algorithm changes, and so the compiler's looking for one thing when the library has another which uses the old name mangling (happened, for example, when Walter added string and float template params).  But if you're not linking to any precompiled libraries, hmm..

"Jarrett Billingsley" <kb3ctd2@yahoo.com> wrote in message news:e491l3$dg9$1@digitaldaemon.com...
> "Andrew Fedoniouk" <news@terrainformatica.com> wrote in message news:e47jup$21kb$1@digitaldaemon.com...
>> Working DMD version (for Harmonia) was DMD 0.150
>> Since then something was changed in the way how it links/compiles.
>>
>> Andrew.
>
> Looks kind of like when the template name mangling algorithm changes, and so the compiler's looking for one thing when the library has another which uses the old name mangling (happened, for example, when Walter added string and float template params).  But if you're not linking to any precompiled libraries, hmm..

Thanks Jarrett.

Probably this information will be useful:

I am able to compile in release but only if I'll remove -inline flag.

Dflags which does not work:
DFLAGS  = -c -inline -O -release

Dflags which does work:
DFLAGS  = -c -O -release

No inlining anymore?

Andrew Fedoniouk.
http://terrainformatica.com

Andrew Fedoniouk wrote:
> I am able to compile in release but only if I'll remove -inline flag.
> 
> Dflags which does not work:
> DFLAGS  = -c -inline -O -release
> 
> Dflags which does work:
> DFLAGS  = -c -O -release
> 
> No inlining anymore?
> 
> Andrew Fedoniouk.
> http://terrainformatica.com
> 

Since DMD 0.151, with -inline -release, inlining is done for all imported modules as well as compiled ones. I don't know how this affects templates, but that might have something to do with it.

"Deewiant" <deewiant.doesnotlike.spam@gmail.com> wrote in message news:e499rk$upp$1@digitaldaemon.com...
> Andrew Fedoniouk wrote:
>> I am able to compile in release but only if I'll remove -inline flag.
>>
>> Dflags which does not work:
>> DFLAGS  = -c -inline -O -release
>>
>> Dflags which does work:
>> DFLAGS  = -c -O -release
>>
>> No inlining anymore?
>>
>>
>
> Since DMD 0.151, with -inline -release, inlining is done for all imported
> modules as well as compiled ones. I don't know how this affects templates,
> but
> that might have something to do with it.

Sounds suspiciously close....

Sideffect of fixing these? http://d.puremagic.com/bugzilla/show_bug.cgi?id=67 http://d.puremagic.com/bugzilla/show_bug.cgi?id=73

Andrew Fedoniouk.
http://terrainformatica.com