I fully agree. I have even tested this some time ago, but I have not submitted a pull request because I was not sure how to check whether dolfin is compiled with petsc-dev in python (for conditional running of examples and tests). And also I was thinking you prefer to wait for next petsc release (there were some minor adjustments in progress in the TAO interface inside petsc).
Garth may you do this or you want me to do it? Eventually I may try to find some time next week, but not sure. About TAO, I think it could be also useful to add the full nonlinear TAO solver, like for SNES. Differently from SNES, TAO requires also the definition of the functional to be minimized in NonlinearProblem (not only gradient and hessian). Otherwise the interface is very similar to SNES. The reason for which the current interface is only for quadratic functional is mainly due to the fact then when I added it SNES solver interface was not existing (and also that I was mainly interested in the quadratic problem ...). What is your opinion about that? Corrado > > Le 6 mars 2014 à 12:49, Anders Logg <[email protected]> a écrit : > >> On Thu, Mar 06, 2014 at 11:44:51AM +0000, Garth N. Wells wrote: >>> TAO has recently been merged into PETSc-dev, so to keep things >>> simple I propose that we drop support for TAO as an optional >>> dependency package and just configure it via PETSc. This means for >>> now that users of DOLFIN-dev will need to use PETSc-dev to access >>> TAO, and from the next PETSc release TAO will be available to all >>> users who configure DOLFIN with PETSc. >>> >>> Let me know if you object to the proposed change. >> >> Sounds very reasonable. >> >> -- >> Anders >> _______________________________________________ >> fenics mailing list >> [email protected] >> http://fenicsproject.org/mailman/listinfo/fenics > _______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
