Re: Keep local symbols of rump libraries

[Ryota Ozaki <ozaki-r%netbsd.org@localhost>]

On Wed, Apr 12, 2017 at 9:27 AM, Chuck Silvers <chuq%chuq.com@localhost> wrote:
> On Mon, Apr 10, 2017 at 07:22:45PM +0900, Ryota Ozaki wrote:
>> Hi,
>>
>> I'm using ATF tests running rump kernels (rump_server)
>> for development and debugging. When a rump kernel gets
>> panic, it dumps a core file from which we can obtain
>> a backtrace by using gdb.
>>
>> Unfortunately local symbols (i.e., static functions)
>> in a backtrace are unresolved because they are stripped
>> by the linker (and objdump). That makes debugging a bit
>> difficult.
>>
>> The patch introduces a compile option for rump kernels
>> called RUMP_KEEPSYMBOLS to keep such symbols:
>>   http://www.netbsd.org/~ozaki-r/rumplibs-keep-symbols.diff
>>
>> The option is disabled by default to not increase
>> the size of all rump libraries.
>>
>> I'm not so familiar with the NetBSD build system and
>> Makefiles, so the patch may be wrong or clumsy.
>>
>> Any comments?
>
> I've been meaning to ask about something similar to this
> but for everything in the build (libraries and commands)
> for the benefit of dtrace.  dtrace "ustack()" stack traces
> are similarly less useful without the static symbols.
>
> my patch for that was similar to yours, but I changed the
> "-Wl,-x" to "-Wl,-X" rather than removing the option entirely.
> for OBJCOPYLIBFLAGS I did what you did, though I suppose using "-X"
> there too would have been more consistent.

Sure. Changing -x to -X works for me.

>
> if we had an option for preserving static symbols for everything,
> would you still want a rump-library-specific option?

If it's useful for other than rump libraries, having a new global
option would be adequate. I don't have an idea of the name though.

  ozaki-r