Improve matrix inequality support

[odow]

• Requires Fix querying dual of symmetric equality constraints #3797 to be merged first

Follow-up to #3766.

In our discussion of #3766, we never really talked about the fact that == is not ambiguous.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.40%. Comparing base (95da051) to head (4228d97).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3778   +/-   ##
=======================================
  Coverage   98.40%   98.40%           
=======================================
  Files          44       44           
  Lines        5956     5966   +10     
=======================================
+ Hits         5861     5871   +10     
  Misses         95       95           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[blegat]

Can we just replace Nonnegatives here:

JuMP.jl/src/macros/@constraint.jl

Line 583 in 301d46e

operator_to_set(::Function, ::Union{Val{:(>=)},Val{:(≥)}}) = Nonnegatives()

What are the other cases that you'd like to intercept ?

[odow]

I didn't do this, because it could be interpreted as a breaking change.

[odow]

Bump @blegat

[blegat]

I didn't do this, because it could be interpreted as a breaking change.

Breaking change for which function ? We are changing it to _OpGreaterThan before giving it to build_constraint in this PR as well anyway.

[odow]

Changes the return of operator_to_set, but perhaps this is minor: https://juliahub.com/ui/Search?type=code&q=operator_to_set

[blegat]

Indeed, the function could be used outside of a JuMP macro. I'd say it is minor. _intercept_operator is complicated or code too much just to accomplish to accomplish exactly the same thing as changing the returned value of operator_to_set

[odow]

I made a slight tweak to get rid of _intercept_operator but add a new method to operator_to_set.

[blegat]

I still think this complicating things too much just for such a small breaking change. Users are expected to implement operator_to_set, we never said that calling it would always return the same thing. I can't believe we haven't made something that could qualify as breaking for the returned value of such function since JuMP v1.0

[odow]

Okay how about now

[blegat]

I'd prefer to name things for what they are rather than what they are used for.
What about calling the set Nonnegative and saying that
@constraint(model, a - b in Nonnegative()) and @constraint(model, a >= b) are synonymous.

[odow]

Let's please not have Nonnegative and Nonnegatives for the same but slightly different thing.

[blegat]

How about GreaterThanZero ?

[odow]

This needs something similar to #3797

[blegat]

This needs something similar to #3797

What do you mean ?

[odow]

It needed the adjoint dual for the new constraints, which is why this PR is based against #3797

…

On Tue, 6 Aug 2024, 12:17 am Benoît Legat, ***@***.***> wrote: This needs something similar to #3797 <#3797> What do you mean ? — Reply to this email directly, view it on GitHub <#3778 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB6MQJJXGCZ54AZI6C6MI3DZP5UMHAVCNFSM6AAAAABJ4WXBR6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRYHEZTINBRHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[odow]

Okay, this is good to go on my end.