Improve promote_op

[pabloferz]

This is the promised fix for #17794 (at least it works locally for me).

CC @JeffBezanson, @tkelman

[tkelman]

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

[pabloferz]

@tkelman I made a mistake and force-pushed a commit. Can we try again?

[tkelman]

Okay, not sure exactly what this'll do. We'll also probably want to compare this against 0.5.0-rc0.

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

[nanosoldier]

Something went wrong when running your job: RemoteException(2,CapturedException(NanosoldierError: error when preparing/pushing to report repo: UndefVarError(:result),Any[( in printreport at BenchmarkJob.jl:481,1),( in open at iostream.jl:113,1),( in report at BenchmarkJob.jl:387,1),( in run at BenchmarkJob.jl:233,1),( in #503 at multi.jl:1410,1),( in run_work_thunk at multi.jl:996,1),( in macro expansion at multi.jl:1410 [inlined],1),( in #502 at event.jl:46,1)]))
cc @jrevels

[jrevels]

Okay, not sure exactly what this'll do.

It just adds the job to Nanosoldier's queue, so that it will run as soon as a node is free. It shouldn't mess anything up, other than possibly delivering job statuses in a weird order if the jobs happen to run concurrently with the old job (which will continue to run until completion or failure).

The error comment is Nanosoldier trying to report the build failure, but hitting a dumb typo in the error handling code that I've just fixed (some weekend I need to sit down and do a test coverage sprint for Nanosoldier...). I haven't yet deployed that fix, so error comments might be wonky for tonight, but I'll deploy it tomorrow.

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here. cc @jrevels

edit by tkelman: this was relative to master

[tkelman]

@nanosoldier runbenchmarks(ALL, vs = ":release-0.5")

[nanosoldier]

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

[tkelman]

So relative to master this is much better, relative to release-0.5 where #17389 was reverted there are just a couple of regressions and they're not too huge.

[pabloferz]

The remaining regressions seem to come from somewhere else. See https://github.com/JuliaCI/BaseBenchmarkReports/blob/07527ce784fc022b854f37a29230dad4816a6895/daily_2016_8_10/report.md

[tkelman]

Jeff, is this okay to merge?

[JeffBezanson]

Yes. Thanks @pabloferz .

[tkelman]

I'll plan to backport this for RC3, along with reverting dbdaf72 and e6f5d89