The ppc64 handbook seems to be a little vague/missing some important settings. Given that Ive tried builds and had them fail, Im marking this as 'major' (which for most people who need to follow the install doc, I think it will be). Feel free to mod it down to whatever you want. CHOSTS setting ============== The cflags setting doesnt list the ppc64 chosts value, which from the stage3 tarball seems to be CHOST="powerpc64-unknown-linux-gnu" It's not even mentioned. CFLAGS settings.... =================== Also (and I think this is a more generic weakness in the whole install documentation feature) the examples for CFLAGS are not ppc anything at all. I _REALLY_ think there needs to be a table for CFLAGS similar to the one for CHOSTS which lists _stable_ and needed CFLAGS. There arnt that many CHOSTS settings are there? It seems that a 20 line table would provide most users the correct settings.... take for exaple the page: http://www.freehackers.org/gentoo/gccflags/flag_gcc3.html Which seems to be _WAY_ more help than the install document (or any gentoo documentation that I could find). I know that generating the table will probably take talking to each of the different arch groups, but it would __REALLY__ improve the install docs to be able to offically state the [safe]/[fastest] CFLAGS settings for a given cpu. Missing vital CFLAGS? ===================== Im sure I read lots on the forums about the need for the CFLAGS setting -fno-strict-aliasing and if it is vital, then maybe it should be in the install docs? Otherwise you're going to get lots of failed builds/installs. Here we go... "One important thing to mention about gcc-3.3.3 is that it requires the -fno-strict-aliasing flag in your CFLAGS." in http://forums.gentoo.org/viewtopic.php?t=167465 Again, I think the install docs really need a table with cpu/chosts settings. Really. :) There cant be more than, what, 50 all up over all archs? Reproducible: Always Steps to Reproduce:
Having a CFLAGS table would spawn far more than 20 lines. For the architectures that I count and using 1 line for each subarch (since those have a different -march / -mcpu setting) there are 59 lines (and I didn't even search the information for all the supported architectures, only "HPPA, x86, MIPS and ARM"). I am assuming that the default setting used in the make.conf that accompanies the stage tarball is sufficient (and better since it's maintained by the arch-specific developers). The -fno-strict-aliasing isn't vital. I don't have it in my CFLAGS as it slowed down the benchmark app I used. It is however very well possible that the section on CHOSTS/CFLAGS/CXXFLAGS will be rewritten to be more clear on these aspects; perhaps we will write up a complete guide on those settings. As for this bug, I'm closing with the mention that the CFLAGS for PPC64 has been put in the guide :)
"...there are 59 lines (and I didn't even search the information for all the supported architectures, only "HPPA, x86, MIPS and ARM")." So, one 200 line document would remove all those forum posts with people offering a bizar range of flags to use? :) Isnt that an argument in itself? :) I mean it seems to me that a _LOT_ of problems are caused by incorrect CFLAGS etc. One 'from the developer' line would solve most of that :) "...default setting used in the make.conf that accompanies the stage tarball is sufficient (and better since it's maintained by the arch-specific developers)." Hehehehehe. So, it's easy to get the info then.... :) "...perhaps we will write up a complete guide on those settings." "As for this bug, I'm closing with the mention that the CFLAGS for PPC64 has been put in the guide :)" Thanks.
> So, one 200 line document would remove all those forum posts with people > offering a bizar range of flags to use? :) Isnt that an argument in itself? :) No; those postings would still exist as the "safe" values we would list in the document are *really* *low* (something like "-march=... -pipe -fomit-frame-pointer" or similar). Users always want better CFLAGS and we just can't deliver any information on those. We provide our users with the possibility to use their own CFLAGS, not with a limited list of allowed CFLAGS.
back from holidays... So the ppc64 handbook is ready for use? Markus
I'd still appreciate some testing but I assume this can be handled when it's online. I do need some Gentoo developer's (with PPC64 experience) name before putting it linked (just a formality) but I don't think that would be too much of a problem.
I will reinstall my G5 today using the guide step by step. When it will fail I'll know what to change ;-) Markus
ok... reinstallation done! some small changes: 1: Page "Choosing the right installation medium" Section 2.d.: There is a sentense at the beginning which mentions the OldWorld Bootloaders... I think this part of the sentence can be erased :-) 2: Page "Preparing the disks" Section 4.c.: The codeline "mac-fdisk /dev/hda" should be changed to "mac-fdisk /dev/sda", because there only exist G5 with serial ata drives. This are the mistakes I've found. Markus