Subject: Re: sys_errlist
To: Chris G. Demetriou <email@example.com>
From: Tim Rightnour <firstname.lastname@example.org>
Date: 03/22/1998 04:55:10
On 21-Mar-98 Chris G. Demetriou spoke unto us all:
# > number update - much as it does now using __RENAME - and ld.so would
# > know how to retrieve the old versions for programs that were linked
# > with a previous major version number.
# Except, now, if you are willing to bump the major number (really, all
# major numbers in the system, if you're bumping libc's), you can kill
# all of the backward-compatibility bits.
I'm not convinced the "backward compatibility" really works. Every time I
upgrade I end up doing one of two things:
a: re-download base.tar for the old version and install the old libs. I have a
number of programs that suddenly break because they look for libc.so.3.0
b: making a number of gross symlinks and just praying nothing dies.
# I think the requirement that backward compatiblity bits work For All
# Time is a very bad one. It means that the libraries _must_ become
# bloated, over time, and it means added work if interfaces change
# (change the interface and change all backward compatibility bits all
# the way back to the beginning of time).
I agree, don't put too much backward compat in the libraries. I think
realisticly we could bump the version numbers, and then have a "compat" library
distribution, which contains older libraries for programs that expect them.
Thus if I have a machine which has a ton of user bins for like 1.1, or
0.9-current (which I do.. and I'm slowly hacking things over, and making them
work again) then I can install the libs and at least have things work in the
intermn. It would probably take some form of maintinence however to keep the
older libs compiling and working properly however..
# I'm not at all convinced that getting rid of the notion of shared
# library major numbers is at all a good thing...
Tim Rightnour - email@example.com