Re: [PATCH 2/2] linux: Make profil_counter a compat_symbol (BZ#17726)

[Carlos O'Donell <carlos at redhat dot com>]

On 8/6/19 3:18 PM, Florian Weimer wrote:
> * Carlos O'Donell:
>
> > On 8/5/19 3:41 PM, Florian Weimer wrote:
> > > * Joseph Myers:
> > >
> > > > On Mon, 5 Aug 2019, Adhemerval Zanella wrote:
> > > >
> > > > > Also that this symbol export was not meant in first place, so it is unlikely
> > > >
> > > > Like all other exports, it's listed in a Versions file.  So it must have
> > > > been deliberate (even if mistaken) at some point.
> > >
> > > libc.map used to have this:
> > >
> > >       # all functions and variables in the normal name space
> > >       a*; b*; c*; d*; e*; f*; g*; h*; i*; j*; k*; l*; m*;
> > >       n*; o*; p*; q*; r*; s*; t*; u*; v*; w*; x*; y*; z*;
> > >
> > > So the missing static likely triggered the unwanted export, and that
> > > mistake was carried over when libc.map was switched over to explicit
> > > function names.  I do not see evidence of a deliberate decision here.
> >    
> > This is why we say "strong presumption" :-)
> >
> > We want to err on the side of being cautious unless we have evidence
> > otherwise.
> >
> > If there is no evidence that a deliberate and conscious decision was
> > made, then perhaps simplifying the situation may be fine e.g. no compat
> > symbols.
>
> And if necessary, we can add back the symbol again.  As far as
> mistakes go, it wouldn't even be that costly, I think.

Just to be clear I have no sustained objection to this change, but also
no strong opinion one way or the other.

My point is only that I would like to see, what you have provided,
which is evidence that this is entirely a historical mistake.

I think that if Adhemerval proposes the patch, you review it, and nobody
objects, then we should just check it in. As you say we can add a compat
symbol later. We know we want to remove it.

--
Cheers,
Carlos.