[PATCH] D33726: [driver][netbsd] Build and pass `-L` arguments to the linker

[Kamil Rytarowski via cfe-commits]

I managed to link a hello world application, however the NetBSD-specific
nits stand for production programs. RPATH is the first difference that
matters.

Unfortunately I'm swamped by other LLVM projects to join LLD.. research
other differences and prepare patches myself.

On 09.06.2017 00:11, Rui Ueyama wrote:
> I don't remember the details, but I believe LLD-generated executables
> are fine with either Linux or on NetBSD
> 
> On Thu, Jun 8, 2017 at 3:08 PM, Kamil Rytarowski <n54 at gmx.com
> <mailto:n54 at gmx.com>> wrote:
> 
>     On 08.06.2017 22:39, Rui Ueyama wrote:
>     > On Wed, Jun 7, 2017 at 6:55 AM, Joerg Sonnenberger via Phabricator
>     > <reviews at reviews.llvm.org <mailto:reviews at reviews.llvm.org>
>     <mailto:reviews at reviews.llvm.org <mailto:reviews at reviews.llvm.org>>>
>     wrote:
>     >
>     >     joerg added a comment.
>     >
>     >     In https://reviews.llvm.org/D33726#774105
>     <https://reviews.llvm.org/D33726#774105>
>     >     <https://reviews.llvm.org/D33726#774105
>     <https://reviews.llvm.org/D33726#774105>>, @ruiu wrote:
>     >
>     >     > I'm totally against adding per-OS path knowledge to our linker. Compilers already know include paths and I don't want to maintain another list of paths in the linker. Also this can be more confusing than useful when you are doing cross-linking.
>     >
>     >
>     >     The only reason for compilers to maintain that list is for finding
>     >     crt*.o. They otherwise don't care about the library paths at all.
>     >     There is no confusion for cross-linking as long as proper sysroot
>     >     support is used. Which we have been doing on NetBSD for ages.
>     >
>     >
>     > That's not what clang is trying to do for all Unix-like systems (except
>     > NetBSD due to the bug), right? The compiler driver actually passes
>     > library paths to the linker. If you think that is wrong, you should make
>     > a change to stop doing that on all systems. I don't see a reason to not
>     > do this only on NetBSD.
>     >
> 
>     I'm convinced that Joerg is right, that there is need some knowledge on
>     the LLD side. I'm more relaxed about -L paths as long as they work for
>     regular use-cases, but DT_RPATH vs DT_RUNPATH behaves differently on
>     Linux and NetBSD.
> 
>     >     > For all OSes other than NetBSD, LLD works fine with the clang driver as the driver passes include paths to the linker. I don't see any reason not to do the same thing for NetBSD. That stands even if the linker has to have a list of include paths.
>     >
>     >     Sorry, but this is again ignorant and wrong. The very same problem
>     >     of build systems calling ld directly apply on most other systems.
>     >     Even then, the list of linker paths is not the only OS-specific
>     >     knowledge. Things like the DT_RPATH vs DT_RUNPATH mess, init vs
>     >     init_array all belong into this category. The list goes on.
>     >
>     >
>     >     Repository:
>     >       rL LLVM
>     >
>     >     https://reviews.llvm.org/D33726
>     <https://reviews.llvm.org/D33726> <https://reviews.llvm.org/D33726
>     <https://reviews.llvm.org/D33726>>
>     >
>     >
>     >
>     >
> 
> 
> 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20170609/192c00fd/attachment-0001.sig>