Would one of the powers that be invoke Nanosoldier following CI / when Nanosoldier rides again? This PR touches some performance-sensitive code. Thanks!

There is a JULIA_TESTFULL environment variable that we put a few of the time-consuming or comprehensive tests behind. I don't think we need to run the full-factorial set of combinations every time, we could select some sparser subset.

There is a JULIA_TESTFULL environment variable that we put a few of the time-consuming or comprehensive tests behind. I don't think we need to run the full-factorial set of combinations every time, we could select some sparser subset.

Perfect. I'll work that in. Thanks!

@Sacha0 last time I checked, it was (relatively) easy to get the benchmark suite to run on my Linux desktop if you wanted to try that. You could also add in some more sparse benchmarks since the state of that is pretty woeful. (Not to pile on more work, but you submit so many great PRs anyway...)

Without JULIA_TESTFULL, linalg/special runs (locally) in under thirty seconds again. (I pared down the regularly tested annotation types to (LowerTriangular, Symmetric) and the tested sparse/special matrix types to (SparseMatrixCSC, Diagonal).)

last time I checked, it was (relatively) easy to get the benchmark suite to run on my Linux desktop if you wanted to try that.

Cheers, I will have a go at running the benchmark suite locally in the next few days (unless Nanosoldier resurfaces first).

You could also add in some more sparse benchmarks since the state of that is pretty woeful. (Not to pile on more work, but you submit so many great PRs anyway...)

Thanks for the kind words! Perhaps once I've coaxed full back into its grave I'll have a look there. Best!

AFAIK good old nanosoldier is back from vacation.

@nanosoldier runbenchmarks(ALL, vs = ":master")

Travis OS X failure seems #16802 -esque? Edit: Scratch that. More probably related to discussion in #17823?

I'm not sure why Nanosoldier didn't respond to this one...

@nanosoldier runbenchmarks(ALL, vs = ":master")

EDIT: lol, @vchuravy you have to commit access to this repo to use Nanosoldier.

Because not contributor status.

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

Is this in shape to merge? Thanks!

Is this 0.5.x or 0.6? I'm guessing 0.6 since it's a behavioral change.

Or is it purely a performance enhancement?

changes result types of various concatenations, I don't think it's backport material

Should be 0.6. Could easily break things.

Additional thoughts? Thanks!

once more for good measure @nanosoldier runbenchmarks(ALL, vs = ":master")

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

I cannot locally reproduce any of the possible regressions the second nanosoldier run indicated. Thoughts? Thanks!

Hope it's noise and see whether the nightly results look consistent.

The nightly looks happy, so cheers!