[PATCH] D19995: Optimize access to global variable references in PIE mode when linker supports copy relocations for PIE

[Joerg Sonnenberger via llvm-commits]

On Sun, May 08, 2016 at 02:10:55PM -0700, Xinliang David Li wrote:
> It is true that copyreloc is not perfect and has  its own limitations and
> problems, but do not deny its benefits because you do not like it. If you
> feel really strongly about this, I think you should propose alternatives
> and make it generally accepted as well standard.  (So far It is not even
> clear to me what your proposal is: asking user to change source to annotate
> with protected visibility? making all linkers to automatic rewrite (without
> additional overhead, size/icache etc)? or something else? )  When the
> alternative is ready, we can talk about getting rid of copyreloc (even for
> non-PIE usage).

It is clear to me that any further discussion is pointless as you don't
want to accept that copy relocations are a horrible idea. It is also
clear that you do not even want to look at the *existing* alternatives.
My objection stands: do not add copy relocations for a code path that
never used or needed them before. It is not a question of "get rid of
them later". The long history of ELF has proven one thing beyond doubt:
if a feature has introduced once, it is extremely difficult to get rid
of it later, even and especially if it is broken. "But it is optional"
hasn't worked for the mess of DT_RPATH vs DT_RUNPATH either.

Joerg