New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BUG: SciPy 1.11.0rc1 not buildable on PPC due to boost submodule #18600
Comments
Do you know which long double format your version of PPC64LE is running (e.g. ibm128 or float128)? |
I don't actually... It's actually running through QEMU 7.2 on x64 agent. How would I be able to find that out? |
Is there a way to output |
Then it's almost certainly the IBM one. We're not on anything nearly as new as Fedora 36 (rather, something like CentOS 7). PS. Thanks a lot for taking a look at this so quickly, much appreciated! |
Here's a larger traceback so we see what code tries to pull this in:
I'm traveling right now so hard to check, but it'd be great to know if this is a regression in Boost.Math, or if it came in when we replaced a |
If you look at the backtrace you'll see that you are explicitly calling the pdf on BTW the reason we treat this as an error is that "double double"'s cause all kinds of issues in some of our code - it's simply not possible to reason about their behaviour in a sensible way because they have a "discontinuous" representation, this is reflected in their value for So... my contention has always been that it's just too much work to make these types function correctly inside Boost.Math, but I'm happy to be proved wrong! |
I completely agree, and I'd like to remove |
scipy/scipy/stats/_boost/include/code_gen.py Line 189 in ec22c55
Looks like in stats we generate a |
Thanks for everyone's input! I tested @mckib2's #18630 in conda-forge/scipy-feedstock#235 and compilation on PPC now passes. 🥳 If we need to keep |
Trying to build rc1 for conda-forge in conda-forge/scipy-feedstock#235, things are looking pretty OK with two exceptions. I'll open another issue for the OSX test failures, but on ppc, the build already fails during compilation:
The boostorg/math submodule (or some recent update thereof) seems to contain a static_assert that unconditionally fails compilation on PPC (or maybe we can toggle off the long double support on PPC?)
PPC is our "most niche" platform, but it's still a hard compilation error on a long-standing arch in conda-forge, so I'm tentatively milestoning this for 1.11.
CC @mckib2 @mdhaber @rgommers @tylerjereddy
The text was updated successfully, but these errors were encountered: