Specs: add global timeout to individual specs on mt

[bcardiff]

Now and then the preview_mt specs fails.
Initially they used to run in verbose mode to get a notion at which point that happens.

I figure we can add some timeout detection to detect specs that will hang due to events that will not happen. Tight loops will not be aborted, but those should be present in single-thread also.

[straight-shoota]

If the spec still finishes at some point (after SPEC_TIMEOUT), it would report twice, wouldn't it? Not sure that's really any problem, just want to bring it up.

[bcardiff]

It's true that some examples could be reported twice. They will as long as the main fiber didn't exit. The following example has one of each.

require "spec"
require "./spec/support/mt_abort_timeout"

it "hangs (1)" do
  sleep 20
  raise "oh no"
end

it "hangs (2)" do
  sleep 20
end

$ bin/crystal spec -Dpreview_mt hang-spec.ign.cr 
FEF

Failures:

  1) hangs (1)
     Failure/Error: it "hangs (1)" do

       spec timeout

     # hang-spec.ign.cr:4

  2) hangs (2)
     Failure/Error: it "hangs (2)" do

       spec timeout

     # hang-spec.ign.cr:9

  3) hangs (1)

       oh no (Exception)
         from hang-spec.ign.cr:6:3 in '->'
         from src/primitives.cr:255:3 in 'internal_run'
         from src/spec/example.cr:32:73 in '->'
         from src/primitives.cr:255:3 in 'run'
         from spec/support/mt_abort_timeout.cr:10:7 in '->'
         from src/primitives.cr:255:3 in 'run'
         from src/fiber.cr:92:34 in '->'
       

Finished in 30.01 seconds
3 examples, 2 failures, 1 errors, 0 pending

Failed examples:

crystal spec hang-spec.ign.cr:4 # hangs (1)
crystal spec hang-spec.ign.cr:9 # hangs (2)
crystal spec hang-spec.ign.cr:4 # hangs (1)

But I'm fine with that.

The alternative would be to add some state Spec::Example to know if it has "ended" and hence skip later reports. But on the other hand, if the spec finishes (successfully or with error) it might be useful to know if the timeout is too low.

[straight-shoota]

I suppose the double reporting is fine. A downside however is that the tally is incorrect. It reports 3 examples, although there are only 2 (the one that errors was counted twice).

[bcardiff]

Yes, I see that as part of the reporting.

The other alternative is to go back to verbose mode on preview_mt spec.

But it will offer less information IMO. With this PR we allow the rest of the specs to run and avoid hanging them because 1 got stuck. Plus, we will know which one hanged in a nice format.

[bcardiff]

Note: The "HTTP::Server #bind fails after close" spec at spec/std/http/server/server_spec.cr:200 timeout recently https://app.circleci.com/pipelines/github/crystal-lang/crystal/4678/workflows/a9809af7-db21-4c67-96e9-02628f00b6ec/jobs/55559/steps