On Tue, Mar 9, 2021 at 6:13 PM Nathan Dunfield <nat...@dunfield.info> wrote: > > An update: I just tried three different binary versions on Linux: The Ubuntu > 20.04 binary posted at SageMath.org, the sagemath/sagemath:develop Docker > image, and conda on RHEL 7. None of them "just worked" with "sage -i foo". > The Docker image and conda failed completely with > > make: *** No rule to make target 'all-toolchain'. Stop. > > I got farther with the Ubuntu binary. Choosing "sage -i pyflakes", it > successfully pip-installed pyflakes and then started rebuilding all of Sage > from scratch, starting with libpng, pkgconf, etc. So in some sense this > worked --- I was able to abort the build and import pyflakes --- but in the > end was equivalent to building Sage from source if I hadn't stopped it.
Yes, this has been a persistent problem. It's a problem on Sage-Windows as well. In the past I've gotten it to work, but every time people make changes to the build system (which is often) it tends to break again, and as Matthias pointed out there is a not a well-established process for testing this (I try to test it manually but don't always remember to; or sometimes I do test it, but throw up my hands when it doesn't work because I don't want to hold up the release any longer). A big part of the problem is that the system for installing packages is badly interwoven with the build system for Sage itself. There is a good reason for this: If one of sagelib's build dependencies is changed, it is necessary to rebuild sagelib. For optional/experimental packages I think we should try to disentangle them a bit better from the "normal" build system. They really should be "drop-in" and not require a rebuild of sagelib. Part of my original goals for developing the "DESTDIR" build system was so that it would eventually be easier to build binary packages for optional packages. I wanted to be able to use this on Windows, for example, by allowing users to select optional packages by just unpacking pre-built tarballs, but I never got around to materializing that goal. Of course, if such a system did exist it should be available for other platforms as well. > On Tuesday, March 9, 2021 at 9:00:22 AM UTC-6 Nathan Dunfield wrote: >> >> To what extent does installing optional packages "just work" with the >> current binary distributions of Sage? I'm thinking of both those posted on >> sagemath.org as well as things not directly under our control such as the >> sage packages for conda, debian, gentoo, etc. My past experience has been >> that "sage -i foo" works only if I had built Sage from source, though I >> haven't tried any of the binaries recently. >> >> I bring this up because the user impact of moving R or any other package to >> optional depends tremendously on whether "sage -i R" just works. If >> switching R to optional is tantamount to requiring users of R to build all >> of Sage from source, that would be very disruptive for those users of R in >> Sage. Building Sage from source is a huge hurdle for 95% users and a >> nontrivial hassle for the rest. >> >> Best, >> >> Nathan > > -- > You received this message because you are subscribed to the Google Groups > "sage-devel" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to sage-devel+unsubscr...@googlegroups.com. > To view this discussion on the web visit > https://groups.google.com/d/msgid/sage-devel/4c6b267c-b29d-4aae-8bbd-f52f7f9ae820n%40googlegroups.com. -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/CAOTD34bQKFJAGNqpXOajyAwioKCMJGiytHMw3tN%3Dc1L9P_eMcw%40mail.gmail.com.
