Deprecated default solvers

[rofinn]

This seems like the first step to dropping the DEFAULT_SOLVER behaviour and getting rid of the following warning:

┌ Warning: Package Convex does not have ECOS in its dependencies:
│ - If you have Convex checked out for development and have
│   added ECOS as a dependency but haven't updated your primary
│   environment's manifest file, try `Pkg.resolve()`.
│ - Otherwise you may need to report an issue with Convex

which causes precompilation to fail on 1.0. Related to #244

[ararslan]

There seems to be an issue with how often this gets called. See for example the number of times the warning is emitted in https://travis-ci.org/JuliaOpt/Convex.jl/jobs/464710210.

[rofinn]

Yeah, I just figured it was an issue with depwarn. I know we've had issues with that in the past.

[ararslan]

Ah, it seems to be because it's used in the default constructor for Problem. This might help:

--- a/src/problems.jl
+++ b/src/problems.jl
@@ -40,7 +40,7 @@ end
 
 # constructor if model is not specified
 function Problem(head::Symbol, objective::AbstractExpr, constraints::Array=Constraint[],
-                 solver::MathProgBase.AbstractMathProgSolver=get_default_solver())
+                 solver::MathProgBase.AbstractMathProgSolver=DEFAULT_SOLVER)
     Problem(head, objective, MathProgBase.ConicModel(solver), constraints)
 end

[rofinn]

I guess if we deprecate the set_default_solver function too then that should be fine.

[ararslan]

The warning still pops up a lot but it's at least less spammy than before it seems.

[rofinn]

Yeah, looks like the test code is calling get_default_solver() pretty often.

[ararslan]

That's probably not representative of external user/package code, so maybe fine? 😅

[iamed2]

Why did the depwarn cause a problem when that function was used as a default argument? Is it anything special to do with that, or is depwarn just generally not good at warning only once?

[rofinn]

I'm not sure why the depwarn was causing problems, but I have noticed that it isn't very good at only warning once. Maybe it's related to where/how the deprecated function is called? :/

[oxinabox]

It is because maxlog is broken, no?
JuliaLang/julia#28786

It is now fixed, but that fix is only in master, I believe.

[iamed2]

Looks like that made it in to 1.0.2

[oxinabox]

Ah, so it is.