Document an expected crash

[glessard]

Follow-up to #157.

Adds the test with an assertCrashes block, to monitor if and when the official Combine stops crashing in this situation (FB7722681).

Note that Just and Optional.Publisher exhibit the same behaviour.

[codecov]

Codecov Report

Merging #158 into master will decrease coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #158      +/-   ##
==========================================
- Coverage   98.40%   98.37%   -0.03%     
==========================================
  Files          84       84              
  Lines        4758     4744      -14     
==========================================
- Hits         4682     4667      -15     
- Misses         76       77       +1     

Impacted Files	Coverage Δ
.../OpenCombineDispatch/DispatchQueue+Scheduler.swift	99.31% <0.00%> (-0.69%)	⬇️

---

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 d680f09...a2376c8. Read the comment docs.

[OpenCombineBot]

LGTM

Generated by 🚫 Danger Swift against a2376c8

[glessard]

The proposed new test trips TSAN on Linux with:
ERROR: ThreadSanitizer: SEGV on unknown address 0x000000000000
This is not useful information, as we are expecting it to crash.

[broadwaylamb]

Honestly, I don't think this is the right direction. This is a data race, which is undefined behavior, meaning that anything could happen and, strictly speaking, a crash is not guaranteed to occur, for example, if GCD decides to map the global queue to a single CPU core.

[broadwaylamb]

@glessard I'm curious though, how did you encounter this crash?

[glessard]

There should not be a data race because cancel() is required to be thread-safe. This is in the contract for Subscription.
What got me down this road was trying to delay the request from a subscriber, but it may not have been this crash in particular. I found this one in particular by trying to reproduce/understand my original crash.