Re: Relative imports and GNU linker... ?

[Trojany <troianor at trash-mail dot com>]

Thanks for help!

Ian Lance Taylor-3 wrote:
> 
> Both these approaches assume that at compile time there is some way to
> discover the naming convention used by an imported package.  I don't
> know if that is true of your language.
> 
Packages can be shared without source, so if a file compiles, the compiler
can't know which other packages will import this one. So it can't prevent
being imported to a project where its symbols are already in use.
The solution could be when that importing package (source_both in this case)
is compiled:
Maybe I can get the compiler to change the symbol tables in the already
compiled files (source_a and _b) to have unique names. This would require an
additional build-step between compiling and linking -> "preparing" the
object files, by making their symbol names unique.
I don't like my idea, though, sounds like "hacking" the object files.


Ian Lance Taylor-3 wrote:
> 
> In general, you will need a naming trick of this sort to make your
> import statements work with the GNU linker.  The GNU linker does not
> rename symbols on the fly.
> 
Renaming symbols sounds good, but I can't find unique names at compile time.

I just "carefully skimmed through" the ELF specification: it looks like
there's no way to specify the origin (the file) of an imported symbol.
I'd be very happy if someone told me my last statement was a wrong
conclusion!

I assume, if ELF can't do that for me, COFF can't either?

Maybe I could use linker scripts? I haven't read through the linker's
manual, but it has a large section about scripting.

Trojany
-- 
View this message in context: http://old.nabble.com/Relative-imports-and-GNU-linker...---tp28712872p28749749.html
Sent from the gcc - Help mailing list archive at Nabble.com.