|
From: Heiko Z. <he...@zu...> - 2010-11-19 21:22:36
|
Hey,
I'm contemplating on how the lfssystem for 1.5 should look like.
DL 1.5 will be available as 32 and 64 bit version, so we need to either have
2 different lfssystems or one multilib.
I'm currently more leaning towards the multilib.
All the tools which get compiled for DL itself would of course then be only
for the target platform.
One other thought I have is to potentially compile everything statically, so
we don't have to worry about any libraries.
Thoughts?
--
Regards
Heiko Zuerker
<http://www.devil-linux.org> http://www.devil-linux.org
|
|
From: Bruce S. <bw...@re...> - 2010-11-20 01:29:57
|
> I’m contemplating on how the lfssystem for 1.5 should look like. > > DL 1.5 will be available as 32 and 64 bit version, so we need to either have > 2 different lfssystems or one multilib. > > I’m currently more leaning towards the multilib. Yes, I'd go with multilib. Do you know if there is any issues compiling DL (1.4 or 32bit) on a 64bit host system? I just upgraded from Ubuntu 10.04 32bit to Ubuntu 10.10 64bit, and I haven't tried a DL compile yet. > All the tools which get compiled for DL itself would of course then be only > for the target platform. > One other thought I have is to potentially compile everything statically, so > we don’t have to worry about any libraries. It's been while since I compiled anything static, but it seems like the resulting binaries are a HUGE compared; we might have to switch from a CD to a DVD distro. Plus it seems like the memory requirements of a static system may increase a lot too. I'm not a static expert, so please correct me if if I'm wrong. - BS |
|
From: Serge L. <ser...@gm...> - 2010-11-20 03:23:21
|
Hi, let me add my 2 cents. On 11/19/2010 04:59 PM, Bruce Smith wrote: >> I’m contemplating on how the lfssystem for 1.5 should look like. >> >> DL 1.5 will be available as 32 and 64 bit version, so we need to either have >> 2 different lfssystems or one multilib. >> >> I’m currently more leaning towards the multilib. > > Yes, I'd go with multilib. You know, I've been trying to build 64bit DL during a couple of years. And I even managed to get it several times, but the result didn't satisfy me. The main problem we'll face is compilation failures due to multi-arch libs availability. In my case a lot of 64bit applications try to link against 32lib :( Most cases are autoconf mistakes, which are possible to minimize by libs separation, but the efforts are huge. The easiest way was a pure 64bit lfssystems. However, I'm not the best release engineer and my problems might be related to lack of knowledge/experience/etc. So, lets try with the most attractive approach for majority of developers. As a plan B, I'd suggest thinking about 2 pure 32bit and 64bit systems, but single source and script base. Plus maybe some additional logic for handling this environment. More difficulties here, but less difficulties in the future. oftopp: x86 arch is not the only architecture for routers/firewalls/special servers. I'll be happy to see DL on arm some day. The "stackable" build environment could simplify that. > > Do you know if there is any issues compiling DL (1.4 or 32bit) on a > 64bit host system? I just upgraded from Ubuntu 10.04 32bit to Ubuntu > 10.10 64bit, and I haven't tried a DL compile yet. I use 64bit system and there is no problem. Host system are RHEL5/6 and OpenSuse. > >> All the tools which get compiled for DL itself would of course then be only >> for the target platform. > >> One other thought I have is to potentially compile everything statically, so >> we don’t have to worry about any libraries. > > It's been while since I compiled anything static, but it seems like > the resulting binaries are a HUGE compared; we might have to switch > from a CD to a DVD distro. Plus it seems like the memory requirements > of a static system may increase a lot too. > > I'm not a static expert, so please correct me if if I'm wrong. I have to agree, static linking demands additional memory and disk resources. Overhead might be too much. However, we can calculate it - just make test "static" system and compare sizes.. Serge |
|
From: Bruce S. <bw...@re...> - 2010-11-20 03:49:20
|
>>> I’m contemplating on how the lfssystem for 1.5 should look like. >>> >>> DL 1.5 will be available as 32 and 64 bit version, so we need to either have >>> 2 different lfssystems or one multilib. >>> >>> I’m currently more leaning towards the multilib. >> >> Yes, I'd go with multilib. > You know, I've been trying to build 64bit DL during a couple of years. And I > even managed to get it several times, but the result didn't satisfy me. The main > problem we'll face is compilation failures due to multi-arch libs availability. > In my case a lot of 64bit applications try to link against 32lib :( Most cases > are autoconf mistakes, which are possible to minimize by libs separation, but > the efforts are huge. The easiest way was a pure 64bit lfssystems. However, I'm > not the best release engineer and my problems might be related to lack of > knowledge/experience/etc. So, lets try with the most attractive approach for > majority of developers. As a plan B, I'd suggest thinking about 2 pure 32bit and > 64bit systems, but single source and script base. Plus maybe some additional > logic for handling this environment. More difficulties here, but less > difficulties in the future. Good point, I hadn't thought of sloppy autoconf's in the packages. It's more work to make 2 LFSsystems up front, but much less work afterwords, not having to fix a bunch of packages. Plus with 2 LFS's, we're sure a 64 bit compile is really 64 bit. I change my vote to 2 LFS's. :-) - BS |
|
From: Heiko Z. <he...@zu...> - 2010-11-20 13:41:39
|
Hey, Sorry I may not have been clear on the static thing. I was only contemplating to have the initial lfssystem statically linked, to prevent the multilb issues Serge mentioned. Speaking of the multilib issues, I'm glad you made me aware of that. I'll play around with it a bit, maybe I find a good workaround. Worst case scenario, we'll do 2 lfssystems as you guys already said. On the topic of support for other architectures (which I really would like to see for DL): I was once contemplating if we should get rid of our build system completely and use one maintained by somebody else. I played around with "buildroot" beginning of this year and I liked it. Maybe we need to change our approach completely and move to something like that. I'm not sure how this would be handled best, we'd probably need to create a fork of their sources. -- Regards Heiko Zuerker http://www.devil-linux.org |
|
From: Heiko Z. <he...@zu...> - 2010-11-20 21:00:20
|
Hey, After looking around I can confirm that we should stay away from a multilib system, since it just gets too complicated. I'm going to play around with "sysroot", this way we may be able support different architectures better. -- Regards Heiko Zuerker http://www.devil-linux.org |