Prep for v1.46.0

[odow]

Basic

• version field of Project.toml has been updated
  - If a breaking change, increment the MAJOR field and reset others to 0
  - If adding new features, increment the MINOR field and reset PATCH to 0
  - If adding bug fixes or documentation changes, increment the PATCH field

Documentation

• Add a new entry to docs/src/changelog.md, following existing style

Tests

• https://github.com/jump-dev/MathOptInterface.jl/actions/runs/18326452732
• https://github.com/jump-dev/MathOptInterface.jl/actions/runs/18362251645
• https://github.com/jump-dev/MathOptInterface.jl/actions/runs/18511698884
• If new tests were added, ensure that MOI.Test.version_added is implemented.

[odow]

The Juniper and Pavito failures are super interesting and they are caused by #2857

• They have z --> {f(x) in S}, z in {0, 1}
• They have an outer solver that covers z in {0, 1}. They pass z --> {f(x) in S}, z == a to the inner solver. The inner solver says, sure, I know how to reformulate --> assuming z is binary. But z is fixed and isn't {0, 1}. The reformulation is still valid though, which is why the tests were passing before.

Maybe we should revert #2857 and add a check at the JuMP level?

Or maybe the bridge needs to say: I added a ZeroOne constraint even though I didn't just so we validate that an indicator can't be bridged by a continuous solver.

Or maybe we need to convert the error() AddConstraintNotAllowed.

[odow]

https://github.com/jump-dev/MathOptInterface.jl/actions/runs/18362251645

[odow]

• BARON: emailed Nick
• DSDP: Test failures DSDP.jl#45
• PolyJuMP: known
• SDPA: [SDPA] update to v7.3.19 JuliaPackaging/Yggdrasil#12251
• SumOfSquares: known
• BlockDecomposition: Modify compat for Graphs and MetaGraphs atoptima/BlockDecomposition.jl#111
• Coluna: Modify compat for Graphs and MetaGraphs atoptima/BlockDecomposition.jl#111
• EAGO: some generated function/Cassette thing
• FrankWolfe: Refactor test files into modules ZIB-IOL/FrankWolfe.jl#621
• SetProg: Does not compile on Julia 1.12 kalmarek/SymbolicWedderburn.jl#97

[odow]

Hearing no objections?

[blegat]

The Juniper and Pavito failures are super interesting and they are caused by #2857

* They have `z --> {f(x) in S}, z in {0, 1}`

* They have an outer solver that covers `z in {0, 1}`. They pass `z --> {f(x) in S}, z == a` to the inner solver. The inner solver says, sure, I know how to reformulate `-->` assuming `z` is binary. But `z` is fixed and isn't `{0, 1}`. The reformulation is still valid though, which is why the tests were passing before.

Maybe we should revert #2857 and add a check at the JuMP level?

Or maybe the bridge needs to say: I added a ZeroOne constraint even though I didn't just so we validate that an indicator can't be bridged by a continuous solver.

Or maybe we need to convert the error() AddConstraintNotAllowed.

Pavito and Juniper shouldn't support Indicator constraints so that it's bridged before they pass it down. It doesn't make sense for them to pass down an indicator constraints to a non-MIP solver. It's not like there is a chance for these non-MIP solvers to support indicator constraints.

[blegat]

Maybe this should be fixed before we release ?

[odow]

Already done in #2868

[odow]

Latest failures in Alpine and PowerModels are because of jump-dev/HiGHS.jl#302

[odow]

@blegat can you confirm you're okay with my changes with the ToMILPBridge?

[blegat]

Already done in #2868

Ok so the issue was that Juniper said that it supported indicator because it was redirecting to the inner solver which said it was supporting it. Now with https://github.com/jump-dev/MathOptInterface.jl/pull/2868/files, the inner solver says it does not support indicator so Juniper says the same so it is bridged in the bridge outer of Juniper, that's an elegant fix.

[odow]

that's an elegant fix.

Yip, although it isn't a perfect fix.

On the outside, Juniper fixes z to either 0 or 1. This flows to the inner solver, so the MIP reformulation would have been valid even without z in {0, 1}. But detecting that either z is ZeroOne, or that it has fixed bound of 0 or 1, or that it is integer and has bounds [0, 1], or that it is a constant, or that there is some other equality constraint that 1.0 * z == 1 (or that there's a constraint like z * (1 - z) = 0) seems like it is out-of-scope for the bridge.