Re: A new pthread patch

[Zack Weinberg <zack@rabi.columbia.edu>]

On Sun, 7 Feb 1999 21:00:28 -0800 (PST), H.J. Lu wrote:
>Hi,
>
>I don't like to call __libc_allocate_rtsig () for thread signals. It
>is very unreliable for gdb. Here is my patch for linuxthreads. It moves
>thread signals to glibc, which will be initialized at the link time. I
>will send you a patch for glibc in another email.

Isn't it a bit late to be breaking the debugger interface?  I don't
like the idea that libc 2.1 will only work with gdb .0.11 and vice
versa, the distribution people will hate it.

If you really want a more reliable debugger interface, make like the
dynamic linker: put a stub function on the interesting path and teach
gdb to set an internal breakpoint at it.  That's back compatible since
gdb can fall back to the old way if the stub isn't there.

Ulrich: I think __libc_allocate_rtsig or equivalent _should_ be user
visible.  Otherwise, someone will blindly write code that uses
SIGRTMAX in a threaded app, not realizing that libc wants it for
internal use.

zw