Jed,
My experience so far is that rebuilding, while not a walk-away and wait process, is still pretty straight forward. The only problems I've encountered with rebuilding 3.0AS is just getting the initial environment right and bootstrapping my way up. There are some minor tweaks that could be made to spec files to insure clean builds. Also, a build script that sets and ensures the correct environment and assists with dependencies would be handy. These things wouldn't break compatibility. In fact, they may help assure it.
The "added value" that I was suggesting is to make it exactly a walk-away and wait process - for a build on the system itself.
What system are your doing your builds on? I think it important (added value in the open source sense) that a "walk-away and wait" build can be done on the system itself. I don't see so much value perhaps in the ability to build the whole system (all packages in one go) on the system so much as the ability to build every package individually - though I believe the two notions tend to the same result.
My current challenge is to try to get a self consistent, convergent build. One that the previous build closely matches a subsequent build. I'm still seeing some problems here, but I think they are related to the environment. Has anyone else checked this?
The environment is certainly an important consideration. That's an aspect of the "Builds from source!" that of course must be carefully specified. Perhaps it's important to tie down a user (root?) under which the build on the system can be done directly with the environment as it exists on a base install of the system? I think Simon's suggestion of a "package" (literally a package?) to do the whole system build is an excellent one.
I'm not sure I understand what you mean by "self consistent, convergent" and "the previous build closely matches a subsequent build." When you say "closely matches", what sort of comparison are you doing? Binary compares of the rpms? If so, what measure of closeness are you using?
Regarding:
Jed: One concern I have is that such an effort might in some ways compromise the Redhat compatibility...
VP Rosen replies: That is, indeed, a concern. I believe cAos is a fork in that respect.
I believe Shane is saying above that he believes these two notions ("builds from source!" and compatible with Redhat) are not opposed. Certainly they are two important values that we can try to achieve simultaneously - at least until they appear in obvious conflict.
Regarding:
Jed: ...distribute a system that would ACTUALLY BUILD with no problems, gotchas, etc. ON ITSELF?
Simon J Mudd wrote:
This implies a "package" which can do the build process which can be installed on a particular box and which will finish "unattended".
Specifically on a base build of the system itself in an existing environment on that system. Yes. That is what I believe will add value to the distribution on an open source sense that can be sold.
Regarding:
Simon J Mudd wrote:
In terms of making progress with the open source movement I´m not so sure. we are toeing a close line to where RH will not want us to be, and if they have to they can mess up the whole rebuild process, simply by doing something as simple as using a non-free compiler (or similar) or add lots of non-free RH tools which the rest of their environment relies on.
This is a key area where I believe Redhat needs to be faced with a clear choice. Either they continue to embrace the open source approach to software development - at least to benefit from it as they clearly do and not stand in the way of others benefiting from it - or they start erecting barriers to clear open source values. I believe "builds from source!" is a clear such open source value that can help clarify this choice. If Redhat was to start taking actions such as you note above (non-free compiler, etc.) then they will have made their intentions clear. At that point I believe a fork becomes a necessity - though as Shane suggests it may still be possible to benefit for some time by functional Redhat compatibility (mostly errata need essentially no mods) while still achieving this additional open source value.
I see this "Redhat rebuild" effort as a pivotal juncture in the open source movement. It seems to me to be testing the limits of the GPL applicability. I think it important the everybody see as clearly as possible what the consequences of various policy choices (e.g. charge per box, per year licenses as Redhat is doing) on their viability in the open source community.
--Jed http://www.nersc.gov/~jed/
rhel-rebuild mailing list [EMAIL PROTECTED] Hosted at the University of Innsbruck, Austria
