Slashdot Mirror
NetBSD Packages Collection No Longer Frozen
jschauma writes "As many users will probably have noticed by the increase in recent pkgsrc
commits, the NetBSD
Packages Collection freeze is now officially over.
Starting October
6th, 2003 and lasting almost two months, the NetBSD Packages
team concentrated upon stabilizing the over 4,000
software packages and the pkgsrc
infrastructure to prepare for a stable pkgsrc branch. During that time, the
number of broken packages during a i386 bulk build was brought down to a mere
15, and a large number of PRs was closed.
A new branch with the tag ``pkgsrc-2003Q4'' was created, allowing our users to
maintain a highly stabilized third-party software package managment
environment, as only pullups of significant importance (such as security
issues) are applied to this branch."
The best thing about this is propably that new stabilized branch. In the past I've used almost everytime the newest sources available to keep up with all the patches, but if this new branch has only the important patches applied to it, it's definetely going to be the one I'm using. If this is going to be updated in the future too, the name of the new branch (pkgsrc-2003Q4) wasn't the best one though.
I demand the Cone of Silence!
My Athlon XP 1800+, with 1.1 GB RAM and a fair bit of disk, was used. A "from-scratch" bulk-build takes between 5 and 6 days. The time for update bulk builds depends upon what was updated (duh). The machine was running NetBSD 1.6ZF (aka NetBSD-current), and the builds were done in a 1.6.2RC2 sandbox. The 15 broken packages which Jan mentioned can mostly be attributed to this setup: we found that pkgsrc/pkgtools/libkver works really well (it's a wrapper around sysctl(2)), but that Linux emulation has problems with libkver (because NetBSD and Linux use different ways of delivering errno for threaded programs). Using a wrapper for uname(1) allowed the packages which used Linux emulation during the build process to complete successfully, but imake uses sysctl(2) directly to get OS version information, and in the end we had to use a hybrid of the two to make packages. The bulk build results do not reflect this, which is probably the reason for the 15 packages.
Stop spreading FUD here. NetBSD shows 17 cpu types. Yes, Linux still supports more with ia64, ppc64, s390, etc. so at least your counting is somewhat off on both sides.
:)) but both are in heavy development and will quickly improve on all OSes. Obscure embedded platforms like v850, cris, h8300, etc. would make me nervous to use in a large production. They are not widely developed for and what info you can dig up suggests that they could be quirky with all the compiler weirdness, etc. that hasn't been shaken out yet. I'd much rather go with a more matured platform like ARM, MIPS, M68k, etc. regardless of the OS being chosen for the application.
:)
Your counts totally ignore edian issues. A playstation 2 and an SGI machine can't run the same binaries even though they are both MIPS because one is big endian and the other is little endian. ARMv2 is completely different architecturally than later versions. There are many other examples.
x86-64 is still a work in progress on both platforms and so is ppc64 (it's only a couple weeks old
There are times when a platform is more mature/complete/etc. (PA-RISC on linux is better supported) but NetBSD is generally very consistent and complete across all of the 40 platforms it currently supports.
The bottom line is use the right tool for the job. If I have a PA-RISC or s390 or wanted to build a PVR then I'd probably choose Linux and I would choose a BSD for most of the rest of my needs. You might choose a little different but both are good tools and very capable.
To get back ontopic. I use pkgsrc on several platforms ( BSD, Linux, OS X, Irix) and these fixes helped out on all platforms. I love the work that everyone has put into pkgsrc and can't wait to see it grow and develop more. Someone else needs to test it on Solaris Sparc/x86 since I don't have a box currently running it.