Slow `read_from_file` for `.cbf` files containing SDP constraints

[mipals]

Hi there,

I am currently solving a few models with PSD constraints using MosekTools.jl. The models are all read from .cbf files using JuMP.read_from_file, but there seem to be a large overhead in the reading of the models.

As an example one the models look as follows

julia> model_mosek
A JuMP Model
Maximization problem with:
Variables: 100750
Objective function type: AffExpr
`Vector{VariableRef}`-in-`MathOptInterface.Nonnegatives`: 1 constraint
`Vector{VariableRef}`-in-`MathOptInterface.PositiveSemidefiniteConeTriangle`: 9632 constraints
`Vector{AffExpr}`-in-`MathOptInterface.Zeros`: 3 constraints
`Vector{AffExpr}`-in-`MathOptInterface.Nonpositives`: 1 constraint
Model mode: AUTOMATIC
CachingOptimizer state: ATTACHED_OPTIMIZER
Solver name: Mosek

Using @time optimize!(model_mosek) I get the following

 43.853957 seconds (2.12 G allocations: 142.302 GiB, 13.44% gc time)

The model itself is solved in around 6 seconds, so the main issue here seem to be in the conversion of the model.

When reading and solving the same models using Clarabel there is no overhead when reading in the .cbf files, which is making me think there is an issue hiding somewhere here.

[blegat]

Thanks for reporting the issue. Could you share the cbf file (or an alternative version that reproduces the issue if you have sensitive information) ?

[mipals]

Yes, I can share a few files over email if that is okay?