Add checks for if instruments exist before running instrument functions.

[Fuyukai]

This provides a small speed-up.

Microbenchmark script:

import time

import trio


async def main():
    q = trio.Queue(0)

    async def i1():
        await q.put("123")

    async def i2():
        i = await q.get()

    start = time.time()
    print("Start time", start)
    async with trio.open_nursery() as n:
        for x in range(0, 5_000):
            n.start_soon(i1)
            n.start_soon(i2)
            if x % 500 == 0:
                await trio.sleep(0)

    end = time.time()
    print("End time", end)
    print("Elapsed", end - start)


trio.run(main)

Before:

Start time 1535036274.015487
End time 1535036274.5630355
Elapsed 0.5475485324859619

After:

Start time 1535036229.3229249
End time 1535036229.8288205
Elapsed 0.5058956146240234

I generally got a ~40-50ms speedup with this change.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@ac4a028). Click here to learn what that means.
The diff coverage is 90%.

@@            Coverage Diff            @@
##             master     #628   +/-   ##
=========================================
  Coverage          ?   99.28%           
=========================================
  Files             ?       91           
  Lines             ?    10763           
  Branches          ?      768           
=========================================
  Hits              ?    10686           
  Misses            ?       58           
  Partials          ?       19

Impacted Files	Coverage Δ
trio/_core/_run.py	99.68% <90%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ac4a028...ff7bbc8. Read the comment docs.

[njsmith]

eh, in the long run I think we should try to approach these things systematically based on benchmarks and bottlenecks people experience in real life (#604), but this particular change is so trivial and harmless that sure, why not.

[Fuyukai]

I actually discovered this by a profile - a suspicious amount of time was being spent inside instrument() so I added these guards and they seemed to reduce it by a bit.

[njsmith]

Oh interesting, what were you profiling?

[Fuyukai]

Just my Discord bot - I just wanted to see if there was any low hanging fruit in the main trio loop for a real program, saw the instrument calls, and wrote the change and a more pure benchmark to show it off.

[njsmith]

Neat. BTW, if it's possible to convert your bot code into a benchmark (like literally: take the existing code and wrap a harness around it), that could be a good way to start at #604 ... the more quirky real-world stuff is in a benchmark, the better it is for guiding these kinds of decisions :-).