BSD ld (was Re: MCLinker and llvm-config)

[David Chisnall]

On 29 Jul 2012, at 16:08, Erik Cederstrand wrote:

> MCLinker seems to me like the most promising project at the moment. But in the end, working code is what counts :-)

Agreed.  The top requirement on my list was 'actually exists' for that exact reason: vapourware is of no interest.  Unfortunately, while the lld design is very promising, the progress doesn't look like it will produce something that is ready in time for 10.0 (including the at-least six months of testing in -CURRENT that as important as a new linker will need).  

MC-Linker has made a huge amount of progress since I last looked at the project, and has gone from being a toy to a serious contender.  Hopefully we will get external toolchain support (it's on brooks@'s todo list, and something that I think a few of us will work on at the DevSummit) soon, which means that the requirement for our system toolchain can be limited to tier 1 platforms (and one of the requirements for promoting a platform to tier 1 can be full toolchain support).  Currently, that basically means x86 and x86-64, with ARM coming Real Soon Now™.  Maybe MIPS (an architecture that manages to combine the code density of Alpha with the simplicity and elegance of x86), all of which are supported by MC-Linker.

At this stage, it certainly looks as if MC-Linker is going to be the first to produce something that can link the base system, and that, as you say, is what matters in the end; working code.

David