Re: [Bug-readline] [PATCH] Enable visibility annotations

[Yury Gribov]

On 04/14/2016 04:10 PM, Mike Frysinger wrote:
> On 14 Apr 2016 15:28, Yury Gribov wrote:
> > On 04/14/2016 01:47 PM, Pedro Alves wrote:
> > > On 04/14/2016 07:45 AM, Yury Gribov wrote:
> > > > These two patches add visibility annotations to readline's public
> > > > functions and also enable -fvisibility=hidden on platforms where it's
> > > > available.
> > > >
> > > > This results in 9% size decrease (279K -> 254K) and 45% export symbols
> > > > reduction (655 -> 398) for libreadline on x86_64.
> > > >
> > > > The autoconf hunk is similar to one that I've recently submitted to
> > > > X.org's libICE, hope there's no licensing problem about this.
> > > >
> > > > Is this ok for trunk?
> > >
> > > FYI, this is probably doing to break (at least) gdb against system
> > > readline.  gdb relies on accessing a few private readline symbols...  :-/
> > >
> > > E.g.:
> > >
> > > tui/tui-io.c:437:  extern int _rl_echoing_p;
> > > completer.c:1677:  extern int _rl_complete_mark_directories;
> > > completer.c:1770:extern int _rl_completion_prefix_display_length;
> > > completer.c:1771:extern int _rl_print_completions_horizontally;
> > >
> > > Unless/until someone fixes these and adds the missing public APIs
> > > to readline, I think distros will need to either back off this
> > > patch, or build gdb against a static readline, which they'd rather
> > > not (at least Fedora doesn't want that).
> > >
> > > We may need a configure option as workaround, or perhaps export
> > > the _rl_foo symbols as well...
> >
> > Thanks, this sounds as a general problem with visibility enabling.  Is
> > there some (relatively straightforward) way to find all used symbols?
> > I'm afraid there'll be more than just _rl_*...
>
> just build gdb w/system readline enabled and see if it works :)

GDB - yes, sure but I'm more worried about other packages that silently 
use non-public readline symbols...

-Y