Re: Reason for __ aliases of variables

[Florian Weimer <fweimer at redhat dot com>]

On 03/07/2017 09:06 PM, Joseph Myers wrote:
> On Tue, 7 Mar 2017, Florian Weimer wrote:
>
> > We have some global variables in glibc which have internal __ aliases (e.g.,
> > daylight and __daylight).  Why do we do this?
>
> As discussed in the signgam context: if the __ alias is exported at a
> public symbol version, as well as the non-__ version as a weak alias, and
> glibc internally only references the __ version, then: if an application
> references the variable without defining its own copy then the application
> will end up with a dynamic relocation against the __ alias (thanks to
> static linker magic), while if it defines its own copy then that copy will
> be separate.  So this provides namespace protection.

This is quite interesting, thanks.  I guess this means that we want to 
keep exporting both __ and non-__ symbol sets with default (public) 
versions, just for consistency, but we do not want to define the __ 
symbols in header files.

(I'll remove the comment in my recent patch which alludes to a binutils 
bug in this area because this will not matter with two sets of public 
symbols.)

Thanks,
Florian