[2.45 PATCH 0/3] Backport GLIBC_ABI version dependencies

[Sam James]

"H.J. Lu" <hjl.tools@gmail.com> writes:

> Richard would like to make GCC configure fatal by default unless users
> have safe glibc or explicitly ask for it:
>
> https://inbox.sourceware.org/gcc-patches/CAFiYyc2P5ioY=w2wuHZoa275yLsEKBnnQbGgpkDDys=7x718kA@mail.gmail.com/
>
> This patch set backports GLIBC_ABI version dependencies to 2.45 so that we
> can tell users to use it.

Nick, for some more context:

I'll note that we've backported the relevant symbols to appropriate
glibc branches (quite far back) based on discussion there, but we need
the GLIBC_ABI_GNU2_TLS dependency to be added by bfd for this to be useful.

Coordinating changes across the GNU toolchain isn't easy. But it is a
strength that we can do these things when necessary. My hope is we
can have these backports (even if extraordinary) and possibly have a
2.45.1 release for:
a) this, and
b) the strip fixes (already on the branch), and
c) the windres fix (already on the branch).

I've been using these for a while and not had any issues.

I will also note:
* the GLIBC_ABI_DT_X86_64_PLT change here fixes some confusion that hit
  distributors of glibc (https://sourceware.org/bugzilla/show_bug.cgi?id=33212#c0)
  becausee we were previously abusing GLIBC_2.36 for that;

* the GLIBC_ABI_GNU_TLS (not the GLIBC_ABI_GNU2_TLS one) change is
  needed to know if binaries are safe even without any default changes
  in GCC for i386 if TLS is used.

> [...]

sam