lld

Linkers can also do something called "Identical code folding", or ICF, whereby the linker notices that two pieces of code are exactly the same and can merge them. lld's sources include a little overview of how this is done, see https://github.com/llvm-mirror/lld/blob/master/ELF/ICF.cpp

> The leading '_' before symbols isn't just a macOS thing it definitely is - see e.g. https://github.com/llvm-mirror/lld/blob/master/lib/ReaderWri...

http://releases.llvm.org/3.9.0/tools/lld/docs/Readers.html Seems to indicate that many object file formats contain the information you need to split them up in such a fine-grained manner. But maybe, they require you to create special versions of object files where even references internal to each library are referenced there as if they live in a different object file? Is that even possible? Edit: corrections welcome, but https://github.com/llvm-mirror/lld/tree/master/lib/ReaderWri... seems to indicate that this only sopports Mach-o (and a yaml-based format used for debugging and testing)

As I wrote in the blog post, we didn't take a closer look at the performance difference, and I know that "measure, don't guess" is maxim. But if I risk it, I believe there are some differences that could make a difference in performance as described in the following chapter. (This is for COFF but applicable to ELF.) https://github.com/llvm-mirror/lld/blob/master/COFF/README.m... We can't reach a conclusion until we add a linker script support to lld, though.

I think this is the COFF document which they say "is directly applicable". https://github.com/llvm-mirror/lld/blob/master/COFF/README.m... I'm guessing the speed difference comes from this linker using a different approach to symbol resolution which requires visiting files fewer times, but I'm not sure.