Bump minimum required LAPACK/BLAS version to 3.7.1
Dear all, As discussed in [1] and elsewhere extensively for the past roadblocks we have encountered, it finally seems like we can bump the minimum required LAPACK/BLAS version to 3.7.1. A preliminary search over the major Linux distros show that CentOS 7 and OpenSUSE Leap are the only ones that doesn't come with the liblapack with version number lower than 3.7.1. Windows and OSX can already use whichever version they need. However this doesn't mean that these systems are not going to be supported. For building SciPy you would need OpenBLAS/MKL anyways and those are independent from the system-wide liblapack machinery. Thus it doesn't seem like a roadblock. Many SciPy users who install pip or conda way don't even notice this but I would like to probe for a reason which this bump might cause an issue in some edge-case system. In turn this will finally allow us to get rid of past version issues and to get rid of certain legacy code. Please share your thoughts, Best, ilhan [1] https://github.com/scipy/scipy/issues/5266
On Sun, Mar 27, 2022 at 3:12 PM Ilhan Polat <ilhanpolat@gmail.com> wrote:
Dear all,
As discussed in [1] and elsewhere extensively for the past roadblocks we have encountered, it finally seems like we can bump the minimum required LAPACK/BLAS version to 3.7.1. A preliminary search over the major Linux distros show that CentOS 7 and OpenSUSE Leap are the only ones that doesn't come with the liblapack with version number lower than 3.7.1. Windows and OSX can already use whichever version they need.
However this doesn't mean that these systems are not going to be supported. For building SciPy you would need OpenBLAS/MKL anyways and those are independent from the system-wide liblapack machinery. Thus it doesn't seem like a roadblock. Many SciPy users who install pip or conda way don't even notice this but I would like to probe for a reason which this bump might cause an issue in some edge-case system.
In turn this will finally allow us to get rid of past version issues and to get rid of certain legacy code.
Please share your thoughts,
Thanks for pushing this forward Ilhan! No objections after >10 days, so let's go ahead I'd say. Cheers, Ralf
Best, ilhan
[1] https://github.com/scipy/scipy/issues/5266 _______________________________________________ SciPy-Dev mailing list -- scipy-dev@python.org To unsubscribe send an email to scipy-dev-leave@python.org https://mail.python.org/mailman3/lists/scipy-dev.python.org/ Member address: ralf.gommers@gmail.com
participants (2)
-
Ilhan Polat -
Ralf Gommers