• Jason House (12/12) Mar 25 2007 In C/C++, the linker will refuse to include a translation unit if
• Daniel Keep (6/22) Mar 25 2007 If I remember correctly, OPTLINK (the linker DMD uses under Windows) doe...
• Thomas Kuehne (11/32) Mar 26 2007 -----BEGIN PGP SIGNED MESSAGE-----

Jason House <jason.james.house gmail.com> writes:

In C/C++, the linker will refuse to include a translation unit if 
there's no call from another translation unit that's already been 
accepted into the executable.

Can the same thing occur in D?  I know this is an obscure kind of 
problem.  Here's some code that could be affected by this kind of issue.

import myRegistrationMechanism.register

static this(){
   register(functionNotUsedAnywhereElse);
}

void functionNotUsedAnywhereElse(inout int x){
   x += 17;
}

Daniel Keep <daniel.keep.lists gmail.com> writes:

Jason House Wrote:

 In C/C++, the linker will refuse to include a translation unit if 
 there's no call from another translation unit that's already been 
 accepted into the executable.
 
 Can the same thing occur in D?  I know this is an obscure kind of 
 problem.  Here's some code that could be affected by this kind of issue.
 
 import myRegistrationMechanism.register
 
 static this(){
    register(functionNotUsedAnywhereElse);
 }
 
 void functionNotUsedAnywhereElse(inout int x){
    x += 17;
 }

If I remember correctly, OPTLINK (the linker DMD uses under Windows) does this,
and gnu link has an option to do it.

Now, using the gnu link option (--gc-sections, I think) *does* cause problems. 
However, OPTLINK doesn't.

Also, with that example, all the functions *do* get called.  Specifically,
static this() gets called by the _moduleCtor function, which I believe is
generated by the compiler at compile time.

So, long and short of it: yes. :)

        -- Daniel

Thomas Kuehne <thomas-dloop kuehne.cn> writes:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Daniel Keep schrieb am 2007-03-26:
 Jason House Wrote:

 In C/C++, the linker will refuse to include a translation unit if 
 there's no call from another translation unit that's already been 
 accepted into the executable.
 
 Can the same thing occur in D?  I know this is an obscure kind of 
 problem.  Here's some code that could be affected by this kind of issue.
 
 import myRegistrationMechanism.register
 
 static this(){
    register(functionNotUsedAnywhereElse);
 }
 
 void functionNotUsedAnywhereElse(inout int x){
    x += 17;
 }

 If I remember correctly, OPTLINK (the linker DMD uses under Windows) does this,
 and gnu link has an option to do it.

 Now, using the gnu link option (--gc-sections, I think) *does* cause problems.
 However, OPTLINK doesn't.

That isn't due to the linker used but how DMD generates the exception
handling code: http://d.puremagic.com/issues/show_bug.cgi?id=879

Thomas


-----BEGIN PGP SIGNATURE-----

iD8DBQFGB4jKLK5blCcjpWoRAoCDAJ0br2GiliMD3fbA+V5aG69SxAbf4wCglS6k
6eTv1E79Aj/E4N5VNyBasaI=
=LR3b
-----END PGP SIGNATURE-----