I'm posting to this forum, since it seems more active than handhelds.org/oe (I thought for a minute while looking at the archive that either the list or the archive was dead, but I guess there just aren't that many messages from 2004-2005). Please let me know if this is inappropriate. Please also let me know if this has already been discussed elsewhere.
Now that I've finally gotten an oe build to sort of succeed, I'd like to make it work a little better for my setup. The main thing I'd like to do is compile on 3 different machines, each a different architecture (x86, sparc, alpha), targeting two different machines (tosa, h3600), possibly with either 2.4 or 2.6. My servers (sparc, alpha) run 24/7, and are more suited to long-term builds. My PC (x86) is faster, but I turn it off at night. My top-level observations lead me to the following questions (I realize I could read the source to find many of the answers, but I thought I might get quicker answers here):
1) Is the cache dependent on the local.conf file at all? If not, I could at least build for different targets using a single cache.
2) Is the cache dependent on build architecture?
3) Why are symbolic links always expanded fully? This prevents me from moving directories around when I need space, or to use symbolic links to make different trees for different build architectures.
4) Is it possible to link at least the cross directory out of the main tmp tree? I keep most stuff on my NFS-shared directories, but I could probably point the cross directory to a local dir, just like I do with gentoo.
5) How much of the build tree could I safely share between target architectures (right now, I keep 4 build directories)? I suppose I could search the .bb files for references to the changed variables, but I thought I'd ask first if there's an easy answer.
6) Is there a faster, experimental version of bitbake I could try out on my slower machines? All the python programs I use (bitbake, gentoo) are pretty slow, even with the psyco jit compiler, on x86, but they are truly painfully slow on my non-x86 machines.
Finally, I should point out that when I said "sort of succeed", what I mean is that unlike the times I've tried it in the past, where I only wanted to spend limited time on it and expected any build problems to magically vanish in the future on the assumption that whoever put the bug in would fix it for himself, I fixed the few build problems I encountered (the bigest of which was getting the #$&%* gcc-2.95 built & installed on all my machines, but that's gentoo's fault, not oe/bitbake's). This leads me to wonder where to submit patches, and how to determine if my patch is appropriate (total 4 lines of change so far, so nothing big...)
What I also meant by "sort of succeed" was that I built bootstrap-image's for tosa, but I could not test. I tried a 2.6 bootstrap-image, but got nothing but a long stream filesystem errors followed by no init, and with a 2.4 bootstrap-image, I got no response (power button just turns charge light off - no apparent boot). I'll look into those problems myself later.