crossplatform linking?

I don't know why there should be different format object files. Couldn't they be much the same. And the linker finally decide how to link them to platform specific binary? MSVC perhaps has already get some idea of this kind, the link time code generation. I'm just curious why on the X86 arch , we get different object file formats just for platform specific.

davidl wrote: > > I don't know why there should be different format object files. Couldn't they be much the same. And the linker finally decide how to link them to platform specific binary? > > MSVC perhaps has already get some idea of this kind, the link time code generation. > > I'm just curious why on the X86 arch , we get different object file formats just for platform specific. I would imagine it's because OPTLINK requires OMF, linux wants ELF and Mac OSX wants mach-o (or something). -- Daniel

在 Mon, 23 Mar 2009 13:11:39 +0800，davidl <davidl@126.com> 写道: > > I don't know why there should be different format object files. Couldn't they be much the same. And the linker finally decide how to link them to platform specific binary? > > MSVC perhaps has already get some idea of this kind, the link time code generation. > > I'm just curious why on the X86 arch , we get different object file formats just for platform specific. minid's bind is suffering from optlink issue(seems related to obj feed to the linker, when n files feed to linker at a time would make it crash, n-1 won't, but error messages). Can ming32 ld accept dmd cross compilation elf objs? I think it's a possible wordaround of making dmd generating elf obj to bypass optlink.

在 Mon, 23 Mar 2009 13:30:48 +0800，Daniel Keep <daniel.keep.lists@gmail.com> 写道: > > > davidl wrote: >> >> I don't know why there should be different format object files. Couldn't >> they be much the same. And the linker finally decide how to link them to >> platform specific binary? >> >> MSVC perhaps has already get some idea of this kind, the link time code >> generation. >> >> I'm just curious why on the X86 arch , we get different object file >> formats just for platform specific. > > I would imagine it's because OPTLINK requires OMF, linux wants ELF and > Mac OSX wants mach-o (or something). > > -- Daniel One nice thing of .net is your .net assemblies will not requrie further recompilation, and they can run directly by mono on linux.

davidl wrote: > 在 Mon, 23 Mar 2009 13:30:48 +0800，Daniel Keep <daniel.keep.lists@gmail.com> 写道: > >> >> >> davidl wrote: >>> >>> I don't know why there should be different format object files. Couldn't they be much the same. And the linker finally decide how to link them to platform specific binary? >>> >>> MSVC perhaps has already get some idea of this kind, the link time code generation. >>> >>> I'm just curious why on the X86 arch , we get different object file formats just for platform specific. >> >> I would imagine it's because OPTLINK requires OMF, linux wants ELF and Mac OSX wants mach-o (or something). >> >> -- Daniel > > One nice thing of .net is your .net assemblies will not requrie further recompilation, and they can run directly by mono on linux. Ummm... how is that relevant? -- Daniel

Daniel Keep Wrote: > > > davidl wrote: > > > > I don't know why there should be different format object files. Couldn't they be much the same. And the linker finally decide how to link them to platform specific binary? > > > > MSVC perhaps has already get some idea of this kind, the link time code generation. > > > > I'm just curious why on the X86 arch , we get different object file formats just for platform specific. > > I would imagine it's because OPTLINK requires OMF, linux wants ELF and Mac OSX wants mach-o (or something). > > -- Daniel it's not a platform issue but more of a linker issue. after all, mingw uses elf on windows, right? obj files can be converted between different formats so I guess one can take elf obj files, convert them to omf and feed the result to optlink. it's all a matter of linker implementation and it would be nice to have all of this automated by dmd/optlink

Yigal Chripun wrote: > it's not a platform issue but more of a linker issue. after all, mingw uses elf on windows, right? Actually, I'm pretty sure it uses a COFF variant. (Unless this has changed recently?)