warnings for some common performance traps

[mlubin]

This covers some of the common cases but not everything. I think we should wait for user feedback before expanding the scope of performance warnings like this.

[mlubin]

Added void returns to avoid unnecessary allocations, otherwise don't see any performance regressions

[joehuchette]

This doesn't play nicely with the vectorized stuff, which uses this fallback.

using JuMP

A = rand(250,250)
B = rand(250,250)
b = rand(250)

m = Model()
@defVar(m, x[1:250])

@addConstraint(m, A*x + B*x .<= b) # gives warning

[mlubin]

Hmm, any suggestions on how to improve the warning? Bump the threshold for operator_counter to 10000?

[joehuchette]

Actually I think the real problem should be in the _multiply! functions this calls, let me play around with it.

[joehuchette]

But yeah, bumping it up seems prudent as well.

[joehuchette]

This particular instance is fixed with dac3402

[mlubin]

At least the warnings help us find cases where we're doing too much operator overloading.

[joehuchette]

The example above still gives warnings on v0.3

[joehuchette]

(because @addConstraint generates different code on v0.3 and v0.4)

[mlubin]

Hmm. We could disable the performance warnings on 0.3.

[joehuchette]

I'm gonna see if I can do something clever here, but otherwise yeah, let's do that and bump up the counter too.