Don't change order of variables when NLPBlock is used #850
Conversation
Codecov Report
@@ Coverage Diff @@
## master #850 +/- ##
==========================================
+ Coverage 94.89% 94.95% +0.05%
==========================================
Files 72 72
Lines 7939 7942 +3
==========================================
+ Hits 7534 7541 +7
+ Misses 405 401 -4
Continue to review full report at Codecov.
|
|
I have verified that it does fix the failure on |
| end | ||
| # The `NLPBlock` assumes that the order of variables does not change (#849) | ||
| if MOI.NLPBlock() in MOI.get(src, MOI.ListOfModelAttributesSet()) | ||
| vector_of_variables_not_added = [ |
There was a problem hiding this comment.
Why do this only for NLPBlock and not always?
There was a problem hiding this comment.
You mean first determine how each variable is going to be added (either free or constrained) and then add them in the same order (hence splitting the add_variables call in separate one for each contiguous block) so that they are added in the same order even if some variables are contrained ?
Note that with #846, the need to not change the order of the variables might go away. The assumption that variables have indices 1 to n that is used by NLP in MOI does not match the rest of the MOI ecosystem and it will go away once NonlinearExpressionFunction implements map_variables.
Closes #849