When I try to compile the following two files, I get a working program, but if the unused function "quux" is removed, the compile fails with a symbol error at link time.  It looks like the writefln() in quux brings in the necessary symbols but the similar writefln() in the template does not do so.

This does not seem to be related to IFTI -- if I change that the error still seems to happen.

NOTE: this is on a 64 bit machine in Linux, so I need to manually link with gcc using a slight variant of the "gcc" command that dmd uses.  I can provide more data on this if needed.

The compiler is DMD 0.149 for Linux.

The errors look like this:

/home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad' /home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad'

>> autotmp.d
:
: import std.stdio;
:
: template foo(f1)
: {
:     void foo(f1 x)
:     {
:         writefln("no dice, chicago, %s.", x);
:     }
: }
:
:
>> trydmd1.d
:
: import std.stdio;
: import autotmp;
:
: // removal of the quux function breaks the compile
: //void quux(int x)
: //{
: //   double foo = x + 4;
: //    writefln("no dice %s.", foo);
: //}
:
: int main(char[][] args)
: {
:     foo(-29.0);
:
:     if (args.length > 1) {
:         writefln("%s", args[1]);
:     }
:     return 0;
: }
:

Kevin Bealer

Kevin Bealer schrieb am 2006-03-13:
> NOTE: this is on a 64 bit machine in Linux, so I need to manually link with gcc using a slight variant of the "gcc" command that dmd uses.  I can provide more data on this if needed.

Hint: add "-L-m32" do the dmd command line.

If your system needs gcc32:
1) create a new dir
2) add the file "gcc" with the the following content to the new dir

#!/bin/bash
gcc32 $*

3) prepend the new dir to your path when you call dmd

PATH=/the_new_dir:$PATH dmd ...

Thomas

In article <agqfe3-uh8.ln1@birke.kuehne.cn>, Thomas Kuehne says...
>
>-----BEGIN PGP SIGNED MESSAGE-----
>Hash: SHA1
>
>Kevin Bealer schrieb am 2006-03-13:
>> NOTE: this is on a 64 bit machine in Linux, so I need to manually link with gcc using a slight variant of the "gcc" command that dmd uses.  I can provide more data on this if needed.
>
>Hint: add "-L-m32" do the dmd command line.
>
>If your system needs gcc32:
>1) create a new dir
>2) add the file "gcc" with the the following content to the new dir
>
>#!/bin/bash
>gcc32 $*
>
>3) prepend the new dir to your path when you call dmd
>
>PATH=/the_new_dir:$PATH dmd ...
>
>Thomas
>

Thank you; I have tried this, and it didn't work.  Although, I don't know whether I need gcc32 or indeed what it is for.  But the following Makefile is what I am using to build instead, and it seems to work fine.

>> Makefile
: # Makefile
:
: # Project configuration
:
: OBJ=hello.o
: D_OPTS=-debug -g
: BINARY=hello
:
:
: # Machinery
:
: D_LIBS=-lphobos -lpthread -lm
: C_OPTS=-m32
:
: TARGET: ${BINARY}
:
: ${BINARY}: ${OBJ}
: 	gcc ${OBJ} -o $@ ${C_OPTS} ${D_LIBS}
:
: clean:
: 	rm -vf ${BINARY} *.o
:
: %.o: %.d ${SRC}
: 	dmd -c ${D_OPTS} $^
:

This works fine for most of my D projects, but unless there is link time magic in DMD for template instantiations, I don't think this is directly connected to the linking issue I was reporting.  Some C++ compilers have such a thing but I don't know if DMD does.

Thanks,
Kevin

Kevin Bealer schrieb am 2006-03-13:
>
> When I try to compile the following two files, I get a working program, but if the unused function "quux" is removed, the compile fails with a symbol error at link time.  It looks like the writefln() in quux brings in the necessary symbols but the similar writefln() in the template does not do so.
>
> This does not seem to be related to IFTI -- if I change that the error still seems to happen.
>
> NOTE: this is on a 64 bit machine in Linux, so I need to manually link with gcc using a slight variant of the "gcc" command that dmd uses.  I can provide more data on this if needed.
>
> The compiler is DMD 0.149 for Linux.
>
> The errors look like this:
>
> /home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad' /home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad'
>
>>> autotmp.d
>:
>: import std.stdio;
>:
>: template foo(f1)
>: {
>:     void foo(f1 x)
>:     {
>:         writefln("no dice, chicago, %s.", x);
>:     }
>: }
>:
>:
>>> trydmd1.d
>:
>: import std.stdio;
>: import autotmp;
>:
>: // removal of the quux function breaks the compile
>: //void quux(int x)
>: //{
>: //   double foo = x + 4;
>: //    writefln("no dice %s.", foo);
>: //}
>:
>: int main(char[][] args)
>: {
>:     foo(-29.0);
>:
>:     if (args.length > 1) {
>:         writefln("%s", args[1]);
>:     }
>:     return 0;
>: }
>:

FAIL:
dmd -c autotmp.d
dmd -c trydmd1.d
dmd autotmp.o trydmd1.o

PASS:
dmd autotmp.d trydmd1.d

Added to DStress as http://dstress.kuehne.cn/complex/arguments

(It's a "complex" target due to its multistage build process.)

Thomas

In article <8gpje3-v6b.ln1@birke.kuehne.cn>, Thomas Kuehne says...
>
>-----BEGIN PGP SIGNED MESSAGE-----
>Hash: SHA1
>
>Kevin Bealer schrieb am 2006-03-13:
>>
>> When I try to compile the following two files, I get a working program, but if the unused function "quux" is removed, the compile fails with a symbol error at link time.  It looks like the writefln() in quux brings in the necessary symbols but the similar writefln() in the template does not do so.
>>
>> This does not seem to be related to IFTI -- if I change that the error still seems to happen.
>>
>> NOTE: this is on a 64 bit machine in Linux, so I need to manually link with gcc using a slight variant of the "gcc" command that dmd uses.  I can provide more data on this if needed.
>>
>> The compiler is DMD 0.149 for Linux.
>>
>> The errors look like this:
>>
>> /home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad' /home/marlon/src/dscript/trydmd1.d:8: undefined reference to `_arguments_Aad'
>>
>>>> autotmp.d
>>:
>>: import std.stdio;
>>:
>>: template foo(f1)
>>: {
>>:     void foo(f1 x)
>>:     {
>>:         writefln("no dice, chicago, %s.", x);
>>:     }
>>: }
>>:
>>:
>>>> trydmd1.d
>>:
>>: import std.stdio;
>>: import autotmp;
>>:
>>: // removal of the quux function breaks the compile
>>: //void quux(int x)
>>: //{
>>: //   double foo = x + 4;
>>: //    writefln("no dice %s.", foo);
>>: //}
>>:
>>: int main(char[][] args)
>>: {
>>:     foo(-29.0);
>>:
>>:     if (args.length > 1) {
>>:         writefln("%s", args[1]);
>>:     }
>>:     return 0;
>>: }
>>:
>
>FAIL:
>dmd -c autotmp.d
>dmd -c trydmd1.d
>dmd autotmp.o trydmd1.o
>
>PASS:
>dmd autotmp.d trydmd1.d
>
>Added to DStress as http://dstress.kuehne.cn/complex/arguments
>
>(It's a "complex" target due to its multistage build process.)
>
>Thomas
>
>
>-----BEGIN PGP SIGNATURE-----
>
>iD8DBQFEFzbn3w+/yD4P9tIRAnQ2AJ4zzHnyhJVk4YrsPltMELI0JWDy1QCfaQKC
>4QtePXTq+7sodj+t6+YMq7I=
>=cSmb
>-----END PGP SIGNATURE-----

Thanks.  I didn't realize that multistage compilation was the trigger.

Kevin

Couldn't you do this with dmd.conf easier?  Or does dmd not respect that when calling gcc?

-[Unknown]


> PATH=/the_new_dir:$PATH dmd ...
> 
> Thomas

Unknown W. Brackets schrieb am 2006-03-15:
> Couldn't you do this with dmd.conf easier?  Or does dmd not respect that when calling gcc?
>
> -[Unknown]
>
>
>> PATH=/the_new_dir:$PATH dmd ...
>> 
>> Thomas

I never tried that - dmd only looks for /etc/dmd.conf not for ~/.dmd.conf or dmd.conf in the same dir as the binary.

(having multiple dmd versions or a non-root installation in that
setting is a pitta)

This doesn't work:
/etc/dmd.conf: PATH=/opt/dmd/cage:$PATH

This does work
/etc/dmd.conf: PATH=/opt/dmd/cage:%PATH%

A mixed Linux-Windows notation in a Linux config file :D

Thomas