concretizer: remove vestigial code and comment
This commit is contained in:
parent
1571d6240b
commit
91e86f9d0d
@ -615,26 +615,6 @@ no_flags(Package, FlagType)
|
|||||||
%-----------------------------------------------------------------------------
|
%-----------------------------------------------------------------------------
|
||||||
% How to optimize the spec (high to low priority)
|
% How to optimize the spec (high to low priority)
|
||||||
%-----------------------------------------------------------------------------
|
%-----------------------------------------------------------------------------
|
||||||
% weight root preferences higher
|
|
||||||
%
|
|
||||||
% TODO: how best to deal with this issue? It's not clear how best to
|
|
||||||
% weight all the constraints. Without this root preference, `spack solve
|
|
||||||
% hdf5` will pick mpich instead of openmpi, even if openmpi is the
|
|
||||||
% preferred provider, because openmpi has a version constraint on hwloc.
|
|
||||||
% It ends up choosing between settling for an old version of hwloc, or
|
|
||||||
% picking the second-best provider. This workaround weights root
|
|
||||||
% preferences higher so that hdf5's prefs are more important, but it's
|
|
||||||
% not clear this is a general solution. It would be nice to weight by
|
|
||||||
% distance to root, but that seems to slow down the solve a lot.
|
|
||||||
%
|
|
||||||
% One option is to make preferences hard constraints. Or maybe we need
|
|
||||||
% to look more closely at where a constraint came from and factor that
|
|
||||||
% into our weights. e.g., a non-default variant resulting from a version
|
|
||||||
% constraint counts like a version constraint. Needs more thought later.
|
|
||||||
%
|
|
||||||
|
|
||||||
root(Package, 2) :- root(Package), node(Package).
|
|
||||||
root(Dependency, 1) :- not root(Dependency), node(Dependency).
|
|
||||||
|
|
||||||
% The highest priority is to minimize the:
|
% The highest priority is to minimize the:
|
||||||
% 1. Version weight
|
% 1. Version weight
|
||||||
|
Loading…
Reference in New Issue
Block a user