Opened 5 years ago

Closed 4 years ago

Last modified 4 years ago

#51964 closed update (fixed)

sundials update

Reported by: jjstickel (Jonathan Stickel) Owned by: seanfarley (Sean Farley)
Priority: Normal Milestone:
Component: ports Version:
Keywords: Cc:
Port: sundials

Description

Sundials 2.6.2 is available (macports has 2.5.0):

http://computation.llnl.gov/projects/sundials-suite-nonlinear-differential-algebraic-equation-solvers/sundials-software

I started to make a patch to the portfile, but the build system is now cmake only. That plus mpi makes it a challenge for me. I might try after awhile if no one gets to it. Thanks.

Change History (10)

comment:1 Changed 5 years ago by mf2k (Frank Schima)

Cc: sean@… removed
Owner: changed from macports-tickets@… to sean@…
Version: 2.3.4

comment:2 Changed 5 years ago by seanfarley (Sean Farley)

Status: newassigned

SUNDIALS Is mostly just used as a time-stepper for PETSc and PETSc is still on 2.5. Could you file an issue with them to see if they can upgrade?

comment:3 in reply to:  2 Changed 5 years ago by jjstickel (Jonathan Stickel)

Replying to sean@…:

SUNDIALS Is mostly just used as a time-stepper for PETSc and PETSc is still on 2.5. Could you file an issue with them to see if they can upgrade?

That may be what you need sundials for, but there are other use cases. It is a dependency of cantera, for example. We use it independently, via the Fortran interface, to solve large systems of ODEs. There is a bug in the kinsol solver of sundials-2.5.0 that affected us, hence my interest in the update.

If PETSc is stuck with sundials-2.5.0, we may need to have multiple port versions, e.g., sundials25 and sundials26.

comment:4 Changed 5 years ago by seanfarley (Sean Farley)

Frankly, no one should be using SUNDIALS for their linear solvers. You should be directly using PETSc which has an option to interface with SUNDIALS but, honestly, has better time steppers now. I say this even though I'm good friends with Carol (she's a classmate of mine). LLNL just doesn't get the computational math funding like they used to.

I'll try to update SUNDIALS and patch PETSc if needed this time but cantera should really consider something better like PETSc.

comment:5 Changed 5 years ago by jjstickel (Jonathan Stickel)

I am an applied math user (not a developer) and didn't know about PETSc until now. We'll look into it for our next project.

I do see discussions of SUNDIALS in various forums, including scipy. So people are using it. Sometimes it's just that people like to use what they know and change is hard. Thank you for being willing to update it this time around.

comment:6 Changed 5 years ago by seanfarley (Sean Farley)

Good news: the SUNDIALS API is still the same (just tested with PETSc). I'll make the portfile now.

comment:7 Changed 4 years ago by seanfarley (Sean Farley)

Resolution: fixed
Status: assignedclosed

Should be fixed in r152186.

comment:8 Changed 4 years ago by jjstickel (Jonathan Stickel)

Thanks. In case you did not notice, I get this during install:

--->  Staging sundials into destroot
Warning: violation by /opt/local/examples
Warning: sundials violates the layout of the ports-filesystems!
Warning: Please fix or indicate this misbehavior (if it is intended), it will be an error in future releases!

comment:9 Changed 4 years ago by seanfarley (Sean Farley)

Huh. I didn't notice that at all. Thanks for the heads up.

comment:10 Changed 4 years ago by seanfarley (Sean Farley)

Fixed the layout in r153384, thanks!

Note: See TracTickets for help on using tickets.