Possible solution to template bloat problem?

[Daniel Murphy]

"H. S. Teoh" <hsteoh at quickfur.ath.cx> wrote in message 
news:mailman.213.1376962388.1719.digitalmars-d at puremagic.com...
>
> [...]
>> Reminds me: how hard is writing own linker is again? :)
>
> Honestly, I think it's about time linker technology is rethought and
> developed further. Possible developements are automatic elision of
> unused code sections (already implemented in some linkers), automatic
> merging of identical sections (not sure if implemented yet -- may
> require language support), link-time inlining, reordering of symbols to
> increase code locality during execution (optimize for CPU caches), etc..
>
> Or more ambitiously, better integration with compiler so that the linker
> has access to compile-time structures to help it make decisions about
> optimization. Present-day object file formats are too far along the
> process to their executable form to permit many optimizations that could
> in theory be performed by the linker, requiring instead hacks like weak
> symbols, unused section GC, etc..
>
> On the more mundane side, we need better algorithms for improving linker
> performance. Current symbol resolution algorithms don't scale very well
> when your object files are large or have large numbers of symbols.
> Surely there are ways of improving the asymptotic complexity of these
> things!
>
>

Check out llvm's lld.  Their choice of language sucks, but they do appear to 
be trying to rethink the whole mess.