test: add a custom mock clock implementation

[abhinav]

benbjohnson/clock has been archived and no longer maintained.
Removing that dependency is desirable.

This change adds a mock clock adapted from Zap's mock clock
(uber-go/zap#1349), but with operations specific to Fx's needs.

There are two areas that require more scrutiny
because of divergence from Zap's mock clock:

WithTimeout requires us to implement a custom context.Context
so that it can report context.DeadlineExceeded when it's time.
We cannot just use context.WithCancelCause here because the cause
for a context failure is considered different from ctx.Err,
so ctx.Err would still report context.Canceled for a timeout.

When testing that a sleep behaves as expected, there's a data race
between the Sleep and the Add that progresses time.
To resolve this, this change adds an AwaitScheduled method
that blocks until there are operations scheduled for the future.
This is done by using a sync.Cond.
This gives us a way to wait until the sleep is scheduled
before we progress time.

This change also had to update Zap to pick up the release with the
custom clock to drop the benbjohnson/clock dependency from the go.mod
completely.

Refs uber-go/zap#1349
Resolves #1135

[abhinav]

Instead of just the system clock sanity test, the test is now used for both, system clock and mock clock to verify basic outer expectations of the clocks.

[rabbbit]

Looks like you don't intend to put the clock in an externally addressable location? :l I should probably copy paste this into uber/ratelimit too. Not only is it on bobjohnson but also on an ancient version of it - upgrading breaks the tests :l

…

On Fri, Feb 9, 2024, 03:18 Abhinav Gupta ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In internal/fxclock/clock_test.go <#1154 (comment)>: > "github.com/stretchr/testify/assert" ) -var _ Clock = clock.Clock(nil) - -// Just a basic sanity check that everything is in order. Instead of just the system clock sanity test, the test is now used for both, system clock and mock clock to verify basic outer expectations of the clocks. — Reply to this email directly, view it on GitHub <#1154 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACG3UUGE5SGHUG5YCNLUUDYSW5UDAVCNFSM6AAAAABDA6E5YWVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMYTQNZRG4ZDGNJVGY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[abhinav]

Looks like you don't intend to put the clock in an externally addressable location? :l

We do! @prashantv has a nicer version of the clock in progress for re-use between projects.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (fa4c995) 98.74% compared to head (7de231b) 98.78%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1154      +/-   ##
==========================================
+ Coverage   98.74%   98.78%   +0.04%     
==========================================
  Files          30       30              
  Lines        2950     3057     +107     
==========================================
+ Hits         2913     3020     +107     
  Misses         30       30              
  Partials        7        7              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[abhinav]

unclear what's going on here; i don't have visibility into it

[JacobOaks]

LGTM I think, just a couple questions/comments on your tests, and probably should wait for another maintainer to take a look too before merging.

[sywhang]

quick first pass, will come back to this later tonight