ExCALIBUR-NEPTUNE / NESO-Spack

Spack repository for installing NESO components and dependencies.
MIT License
2 stars 2 forks source link

Pin Scotch version to 6.0.4 when building Nektar #15

Closed oparry-ukaea closed 1 year ago

oparry-ukaea commented 1 year ago

Fixes Nektar's Scotch dependence to version 6.0.4. With Scotch 7.0.1 (Spack's default), running with certain global solve modes, often the default, causes a segfault. While this can be avoided by setting a suitable 'GlobalSysSoln' in the config file, I'd argue it's better to just rollback the version and not have to worry about it.

N.B.

cmacmackin commented 1 year ago

I'm currently on annual leave until the new year. Is it urgent that I review this before then?

On Mon, Dec 19, 2022 at 4:01 PM oparry-ukaea @.***> wrote:

@oparry-ukaea https://github.com/oparry-ukaea requested your review on:

15 https://github.com/ExCALIBUR-NEPTUNE/NESO-Spack/pull/15 Pin Scotch

version to 6.0.4 when building Nektar.

— Reply to this email directly, view it on GitHub https://github.com/ExCALIBUR-NEPTUNE/NESO-Spack/pull/15#event-8070833302, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB6ESPPZU7ZYH7MXB2OQU53WOCBGPANCNFSM6AAAAAATDQSAQA . You are receiving this because your review was requested.Message ID: @.***>

-- Chris MacMackin

oparry-ukaea commented 1 year ago

Ah right, sorry, didn't realise - I'll assign to somebody else. Have a good Christmas!

cmacmackin commented 1 year ago

I'm never keen on fixing package versions. I'd propose that instead we fix it in the specs we put in our spack.yaml file.

You mention this can be fixed with the config file. Is this something that can be done simply when writing our program, or would it require manual intervention? If the former, I think it would be worth considering.

jwscook commented 1 year ago

Btw, 6.0.4 doesn't fix the segfaults for me. https://github.com/ExCALIBUR-NEPTUNE/NESO/pull/116#issuecomment-1378869569

oparry-ukaea commented 1 year ago

You mention this can be fixed with the config file. Is this something that can be done simply when writing our program, or would it require manual intervention? If the former, I think it would be worth considering.

Certain choices of 'GlobalSysSoln', including the default in some cases, seem to cause intermittent seg faults when using Scotch 7 but running in serial. Setting a value that works is probably always possible, but it can take a while to diagnose the problem if you haven't seen it before / forget that you've seen it before...!

oparry-ukaea commented 1 year ago

I'm never keen on fixing package versions. I'd propose that instead we fix it in the specs we put in our spack.yaml file.

Yeah, fair point, that would do the same job in a more easily reversible way. I'll close this PR then.