On 24 September 2010 10:24, François Bissey <f.r.bis...@massey.ac.nz> wrote: > Hi Dave,
Hi François >> If anyone is interested, here's a summary of the what I found. >> > > Seen the bug reports starting to show up on trac. I am interested :) Yes, some are quite interesting, as they show bugs that really should be fixed in Sage, that have nothing to do with AIX. >> 3) Four package that did built fully, produced a shared library with the >> .so extension which is common for most systems, but is in fact incorrect >> on AIX. The extension for both shared libraries and archive libraries on >> AIX should be .a and not .so. These four packages are the following. >> >> * Flint http://trac.sagemath.org/sage_trac/ticket/9996 >> * NTL http://trac.sagemath.org/sage_trac/ticket/9998 >> * Readline http://trac.sagemath.org/sage_trac/ticket/9987 >> * Zlib http://trac.sagemath.org/sage_trac/ticket/9988 >> > > A quick question: if both the shared and static libraries have the same > extension, is it possible at all to install both kind of library at the time? > Or do you have to choose one over the other? I don't know that one. I suspect not. To be honest, one has to question why one would need both in general. Perhaps IBM take the attitude you only need one or the other. I will ask on comp.unix.aix about this matter. I noticed a few .o files in /usr/lib. I'm not exactly sure what they are. I'm pretty sure sorting out these issues will be pretty simple, as the libraries build, but just have the wrong extension. I'd rather get fixes upstream if possible, rather than simply rename the libraries. I'm pretty much an AIX newbie, though I have owned an RS/6000 for about a decade, it has not seen much use. In fact, it is probably when the PSU blew up a couple of days after i switched it on for the first time in many years. I got a PSU for $50 + carriage from the USA which was a lot better than the price I was quoted in the UK, which was the US equivalent of about $300. Even with * Cost of PSU * carriage * 2 GB extra RAM * 9 GB disks and the all important holders. The disks are not much use, but the olders are. * A spare service processor card the seller chucked in for free. getting from the US worked out cheaper than buying just the PSU in the UK!! That's the trouble with a machine like an RS/6000 - its aimed at the professional market and has a price to match it. >> Rather annoyingly there are some small python scrips in things like the >> Fortran package which means Fortran never builds, so neither does a whole >> host of things that depend on Fortran. So I never got to see if ATLAS, R >> or anything like that would build. >> > > May be that should cleaned to shell if possible? Yes. It would allow much faster parallel builds too, as more things can be started earlier in the build process. In particular ATLAS could be started earlier, which can take ages to build. To start ATLAS earlier, we only need to * Remove a small amount of Python from ATLAS * Remove a very small amount of Python from the fortran package. To be honest, I could hack the Fortran package very easily, as it only creates a two line shell script which points to the location of gfortran >> 7) The packages that did fail to build, despite their perquisites being >> present were: >> >> * tachyon http://trac.sagemath.org/sage_trac/ticket/9997 >> * Singular http://trac.sagemath.org/sage_trac/ticket/9993 >> * GnuTLS http://trac.sagemath.org/sage_trac/ticket/9975 >> * Cliquer http://trac.sagemath.org/sage_trac/ticket/9979 >> * http://trac.sagemath.org/sage_trac/ticket/9990 > > The one you forgot to name is pari. Yes, it was Pari. I know the Pari developers were keen to get AIX sorted out, but had no AIX hardware. Hopefully if I can provide the hardware, they can sort out the AIX issues. There has been some discussions about IBM providing very fast AIX hardware, but they seem to have stalled a bit. But the discussions got me sufficiently interested to try an AIX build myself. >> If anyone wants an account on this AIX machine, I'm willing to give them >> one. I know a couple of people tried the metamodule open-AIX system, but >> it seems to be quite broken. I'm not saying mine is perfectly set up, but >> it seems to be working reasonably well. However, I'm not going to run this >> machine 24/7, due to the power costs. I don't think the machine is very >> poewr hungry, but with 6 hard drives in it, some of which are 15,000 rpm, >> it's not going to be a very "green" computer. >> > > I'd take you on the offer but I don't really have that much time to spare > until > mid-October when the university year comes to an end around here. > > Francois That's no problem. The machine is not going anywhere. The biggest problem with getting patches into Sage for HP-UX or AIX is getting people to review them. I think what I may do is make a few accounts and publish usernams and passwords openly, but only make the logins possible from the * The University of Washington * Uiversity where someone who is wiill to reivew patches will work * Any subset of IPs where someone might want. I'm reluctant to publish usernames and password that any idiot can use, but making the machine easily available will perhaps making getting things reviewed much easiler. But for now, I'll just report upstream problems, and hope they get sorted out by the time the Solaris ports are complete. I'm not going to spend a lot of time on it now, but the inital results were encouraging. AIX looks a lot easier to support than HP-UX, as for me at least the machine is fully POSIX compliant and has full C99 support. In contrast, the HP box I have is neither fully POSIX nor does it have C99 support. Also with at least 5 GHz CPUs available on AIX machines, Sage could be potentially very fast on such hardware. Dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
