Hi!

Let me begin by saying, I'm sorry if this is caused of some obvious error but since I am new to D, I am not aware of all the tricks and treats it offers.

I am working with the WindowsAPI binding at dsource.org (though I do not believe this is related to the binding itself). However, in my code I call the function LOWORD (win32 specific). This function is defined in the win32.windef module. Although it is defined there (without any encapsulations in version statements or anything similar) I receive this error:

Error 42: Symbol Undefined _D5win326windef6LOWORDFkZt (LOWORD undefined)

To solve this I had to copy-paste the exact function directly into my own file/module instead. So my question is the following; why do I have to copy and paste the function directly in my code when it is clearly defined in one of the files I import?

Windef.d attached

Am 13.06.2011 23:18, schrieb Loopback:
> Hi!
>
> Let me begin by saying, I'm sorry if this is caused of some obvious
> error but since I am new to D, I am not aware of all the tricks and
> treats it offers.
>
> I am working with the WindowsAPI binding at dsource.org (though I do not
> believe this is related to the binding itself). However, in my code I
> call the function LOWORD (win32 specific). This function is defined in
> the win32.windef module. Although it is defined there (without any
> encapsulations in version statements or anything similar) I receive this
> error:
>
> Error 42: Symbol Undefined _D5win326windef6LOWORDFkZt (LOWORD undefined)
>
> To solve this I had to copy-paste the exact function directly into my
> own file/module instead. So my question is the following; why do I have
> to copy and paste the function directly in my code when it is clearly
> defined in one of the files I import?
>
> Windef.d attached

Your error is an linker error.  This means the last step of creating an exceutable couldn't be taken: finding all names you promised to be there.
When you, import a module, you promise: "Compiler, I know that definitions I'm using in this module aren't declared in there, but look at this module. I promise that all these functions in there will be there during the linking process". The compiler gives no error because you promised, but now the linker has problem: It hasn't got the defintions you said to be there. You have to say the linker where to find them.
One way is to give your .d/.lib file to dmd so he can give it to the linker:
dmd main.d path/to/windef.d

Mafi

Importing it means dmd knows about the function and emits a call but doesn't automatically generate the function code.
This is only done if you also pass the file containing it to dmd.

On 2011-06-13 14:18, Loopback wrote:
> Hi!
> 
> Let me begin by saying, I'm sorry if this is caused of some obvious error but since I am new to D, I am not aware of all the tricks and treats it offers.
> 
> I am working with the WindowsAPI binding at dsource.org (though I do not believe this is related to the binding itself). However, in my code I call the function LOWORD (win32 specific). This function is defined in the win32.windef module. Although it is defined there (without any encapsulations in version statements or anything similar) I receive this error:
> 
> Error 42: Symbol Undefined _D5win326windef6LOWORDFkZt (LOWORD undefined)
> 
> To solve this I had to copy-paste the exact function directly into my own file/module instead. So my question is the following; why do I have to copy and paste the function directly in my code when it is clearly defined in one of the files I import?
> 
> Windef.d attached

Importing the module with the function in it means that the compilation of the current module can see that module and use its symbols. But the compiled functions in the imported module are in the object file generated when compiling that module, not the current module. In the case of binding to a C function, you're telling D that a function with that signature exists, but it doesn't have the source to it. In either case, the actual function definition is going to be needed when linking (at least with static linking - it's a bit more complicated with dynamic linking, but on Windows at least, the .lib file still needs to be linked in). In this particular case, one of two things is likely happening.

1. You never linked in the lib file for the library with the function.

2. The declaration given for the C function in question isn't using the right linking. That is, it needs to be extern(C) and/or extern(Windows) (I'm not quite sure what extern(Windows) does differently than extern(C), but I gather that it's necessary at least some of the time when dealing with Windows system functions).

The symbol that it's complaining about looks rather mangled to me (though I don't look at pure C function symbols very often, so I don't know exactly what they look like unmangled - both C++ and D mangle functions to allow for function overloading). So, it's likely that the function's declaration in D isn't properly marked with extern(C) or extern(Windows). So, the function's name gets mangled, and it can't find the unmangled C function in the .lib file.

Regardless, that particular error indicates that the linker failed to locate that particular function, and you have to make whatever changes are necessary - either in code or in your compilation process - to make it so that the linker is able to find the definition of the function in question.

- Jonathan M Davis

Thanks for your answer!

Seems like supplying with the file location solved the problem, though I still wonder about one thing. The imported module is located in "c:/d/dmd2/import/win32/windef.d" and I have used this command line to the DMD compiler: -I"c:/d/dmd2/import/". Shouldn't the linker/compiler be able to solve this on its own then?

On 2011-06-13 14:44, Loopback wrote:
> Thanks for your answer!
> 
> Seems like supplying with the file location solved the problem, though I still wonder about one thing. The imported module is located in "c:/d/dmd2/import/win32/windef.d" and I have used this command line to the DMD compiler: -I"c:/d/dmd2/import/". Shouldn't the linker/compiler be able to solve this on its own then?

-I just tells it where to look for imports, not what to link. The linker never links in anything unless you explicitly tell it to - either by having it directly on the command line or in DFLAGS (usually set by dmd.conf on Linux and sc.ini or Windows). Phobos is listed in dmd.conf/sc.ini, which is why it gets linked in. If it wasn't there, you'd have to list it explicitly too.

- Jonathan M Davis

On 2011-06-13 23:51, Jonathan M Davis wrote:
> On 2011-06-13 14:44, Loopback wrote:
>> Thanks for your answer!
>>
>> Seems like supplying with the file location solved the problem, though I
>> still wonder about one thing. The imported module is located in
>> "c:/d/dmd2/import/win32/windef.d" and I have used this command line to
>> the DMD compiler: -I"c:/d/dmd2/import/". Shouldn't the linker/compiler
>> be able to solve this on its own then?
>
> -I just tells it where to look for imports, not what to link. The linker never
> links in anything unless you explicitly tell it to - either by having it
> directly on the command line or in DFLAGS (usually set by dmd.conf on Linux
> and sc.ini or Windows). Phobos is listed in dmd.conf/sc.ini, which is why it
> gets linked in. If it wasn't there, you'd have to list it explicitly too.
>
> - Jonathan M Davis
Can you modify the sc.ini so the linker automatically checks a specific
directory for imports? Let's say for example this I have my imports in
c:/d/dmd2/import/, can I implement this in sc.ini so I don't have to
worry about these specific linker errors again?

I tried adding "-I%@P%\..\..\import" (custom import directory) in sc.ini
without success.

Another question I'm wondering is; if I include a module and tell its
location explicitly to the linker, do I then also have to tell the
linker the location of all local imports this file have as well?

sc.ini is going to get overwritten when you upgrade DMD so changing it is a bad idea. An alternative is to copy sc.ini to your projects local directory, this way DMD will use that one instead of its own. But this is all working around the issue that you should be passing import directories via a build script, makefile, or some type of build system (that's a bit difficult given the current state of build tools..), and not hardcoding paths to a global config file.

Btw, there are some notes on how to use DMD/Optlink and static libs here: http://prowiki.org/wiki4d/wiki.cgi?D__Tutorial/CompilingLinkingD

I can't tell if it's any good until someone bothers to try things out and see if things work like the page describes. If something's wrong or that page doesn't explain what you need to know, let me know and I'll try to update it with more information.

> Shouldn't the linker/compiler be able to solve this on its own then?

Use rdmd or xfBuild to automatically compile all needed modules.

Thanks for all the answers! Seems like rdmd did the trick.
I don't see why this isn't built in to dmd though, or does it cause
overhead when you are using rdmd? Benefits, Drawbacks?

I've also stumbled upon an additional error with the win32 DirectX
bindings, but this seems D related actually. When I compile any code
at all (with rdmd) which imports win32.directx.d3d9 and uses the
function Direct3DCreate9, the linker issues this warning:

Error 42: Symbol Undefined _Direct3DCreate9@4

It is shown even though I'm using rdmd and I have linked to d3d9.lib
(other functions like CreateDevice and Present still works though).
To solve this error, I did some searching and it seems like I'm not
the first one with this problem. I got 2 hits:

http://www.digitalmars.com/d/archives/digitalmars/D/learn/1475.html
http://www.dsource.org/forums/viewtopic.php?p=13971&sid=d43d05620f0c3a30758a326394ac2e26

Both of these links says that the problem is solved by adding
_Direct3DCreate9 to imports. I've tried to do this by linking to a
.def file with this content (without success, same error is still
issued).

EXETYPE NT
SUBSYSTEM WINDOWS

IMPORTS
_Direct3DCreate9@4=d3d9.Direct3DCreate9

I also tried this code but pragma build_def seems deprecated:

pragma(build_def, "IMPORTS");
pragma(build_def, "_Direct3DCreate9@4=d3d9.Direct3DCreate9");

So am I doing something wrong with the def file or am I perhaps
supposed to use another pragma?

Thanks in advance!