Closed edd8e884-f507-429a-b577-5d554626c0fe closed 10 months ago
Note that this was built on a Core2 Duo CPU (the goal is to have SDL run on old machines with fewer CPU flags).
dup of #30635, I guess
Replying to @mkoeppe:
dup of #30635, I guess
Indeed, perhaps should both tickets be merged (the current ticket shows a different set of errors).
Should i understand that Coin is not the culprit, and that the other mip solvers may give the same amount of wrong results due to numerical noise ? It is important for me to know, because i might just remove cbc from the next SDL (to exchange speed for correctness).
There were multiple things wrong. Meta-ticket #32191 has a large number tickets that fixed how various Sage code was using the numerical results from solvers.
In addition, it appears that sage-numerical-backend-coin is not compatible with the CBC 2.10 series. I haven't had the time to look into this issue.
As the way forward I want to use #31962, which makes sage-numerical-backend-coin obsolete.
In the meantime, you can use --without-system-cbc
. Our spkg is still using CBC 2.9.x, as the update ticket #30644 has stalled
Dependencies: #34251
Since the issue is fixed, I set this issue to positive review.
With
cbc
installed on 9.6, and Coin as the default solver:there are a lot of broken doctests involving
MixedIntegerLinearProgram
, showing wrong (at least unexpected) results:Depends on #34251
CC: @dimpase
Component: packages: optional
Keywords: sdl
Issue created by migration from https://trac.sagemath.org/ticket/33929