Re: libio patch

[Ulrich Drepper <drepper at ipd dot info dot uni-karlsruhe dot de>]

hjl@lucon.org (H.J. Lu) writes:

> 1. The libio is too old. Ulrich, I have sent you a patch for
> libio in glibc 2.1 a few days ago. I can resend them to you
> if you want. My patch supports libc 5, libg++, egcs, glibc 2.0
> and glibc 2.1.

I have your patches and I also integrated them mostly but I'm not at
all convinced this ugly mechanism to handle thunks is good.  I don't
want to carry the load of supporting systems without thunks around.
We shall make a clear cut.  Nobody should use glibc-2.0 to 2.0.4 and
we can always make a patch set available for libc5 (if wanted at all).

> 2. _G_HAVE_LONG_DOUBLE_IO is checked, but not defined. I
> changed it to _G_HAVE_GLIBC since I only implemented long
> double output using glibc.

Bad idea.  I introduced the general symbol since I intend get long
double I/O working with other systems as well.  Many more systems will
support this i future since ISO C 9X will define it.

Beside (I haven't tried to compile the egcs libio) you seem to assume
that the configure script shall generate a _G_config.h file for glibc
based systems.  This is not want I have in mind.  All those magic
symbols we need are defined in the _G_config.h file which comes with
glibc.  So _G_HAVE_LONG_DOUBLE_IO *is* defined is needed.

> 5. We have to support libio in libc which has an incompatible
> vtable thunks support with g++. That means we have to provide
> the incompatible part of libio.

As said above, I don't think this shall be part of the package.  See
how much ugly code this adds and just because of some old systems.

> 6. iostream has no MT-safe support. Applications with glibc
> many dump core because of it.

This is what I'm working on.


-- Uli
---------------.      drepper@cygnus.com  ,-.   Rubensstrasse 5
Ulrich Drepper  \    ,-------------------'   \  76149 Karlsruhe/Germany
Cygnus Solutions `--' drepper@gnu.ai.mit.edu  `------------------------