github.com/thejerf/abtime

abtime

go get github.com/thejerf/abtime

A library for abstracting away from the literal Go time library and the context's cancellation and timeout libraries, for testing and time control.

In any code that seriously uses time, such as billing or scheduling code, best software engineering practices are that you should not directly access the operating system time. This module provides you with code to implement that principle in Go.

See some discussions:

• blog.plover.com's Moonpig discussion
• Jon Skeet's discussion on Stack Overflow
• Jim McBeath's post

This module is fully covered with godoc, including examples, usage, and everything else you might expect from a README.md on GitHub. (DRY.)

Why abtime and not the more popular clock abstractions?

Most if not all other time testing abstractions for Go attempt to simulate the passage of time itself. That is, you can set a Timer for a second from now, then, you tell the time replacement module that one second has passed, and it will trigger the timer at that point.

That is indeed simpler for simple use cases than what I have here, and permits a drop-in interface replacement for the whole module. However, it does not permit you to test all scenarios, because it is built on a fundamentally false premise, which is that time is a monotonic, agreed-upon value for all goroutines. That is not how goroutines "perceive" time.

In reality, if you give one goroutine a timer for 1 second in the future, and another goroutine a timer for 1.1 seconds in the future, it is entirely possible for the second goroutine to entirely finish its execution before the first one even gets woken up. (The first goroutine may well have had its timer triggered, but then immediately descheduled for whatever reason, while the second runs to completion.)

Proper testing of complex time-dependent multi-goroutine coordination requires deeper levels of control than a compatible API can offer. This package takes the hit of having to add unique IDs to timers and tickers in order to permit a deeper level of testing, as proper testing of time-sensitive code must be able to consider the case where events in different goroutines happen "out of order", because they will.

If you only have one goroutine using time-based code then this package may be overkill. However, if you have multiple goroutines interacting with each other while also referring to the clock, you may find this package is worthwhile as it will permit you to set up important test scenarios that drop-in replacements for the time package simply can not express.

Changelog

• 1.0.7:
  • Fixups in the internal registration of triggerable events.
  • Added wrappers around context.WithTimeout and context.WithCancel that allow controlled cancellation of contexts like the rest of time-based code.
• 1.0.6:
  • Manual timer needs to reflect whether it was stopped, not that it was stopped.
  • This also cleans up some of the concurrency. This was one of my earlier libraries. The place where a goroutine is spawned to perform the various triggered actions should now be more correct.
• 1.0.5:
  • (INCORRECT) The manual timer is ALWAYS successfully stopped by a Stop call, so .Stop must always return true.
• 1.0.4:
  • Add ticker.Reset for Go 1.15. This version requires Go 1.15.
  • Add proper go module support.
• 1.0.3:
  • Fix locking for Unregister and UnregisterAll.
• 1.0.2

  • Adds support for unregistering triggers, so the ids can be reused with the same abtime object.

    As the godoc says, this is a sign of some sort of flaw, but it is not yet clear how to handle it. I still haven't found a good option for an API for this stuff. My original goal with abtime was to be as close to the original time API as possible, I'm considering abandoning that. Though I still don't know what exactly that would look like.

    (Plus, this need some sort of context support now.)

• 1.0.1
  • Issue 3 reports a reversal in the sense of the timer.Reset return value, which is fixed. While fixing this, a race condition in setting the underlying value was also fixed.
• 1.0.0
  • Initial Release.

On the Abtime API

Perusing the abtime API will quickly reveal that unlike other clock-mocking libraries in Go, the abtime API introduces additional ID parameters to the calls you make, which are ignored when using real time.

This is because as nice as the API may be, it is a mistake for a clock-mocking library to bake in the assumption that your code can assume the existence of a monotonic time, because in multi-threaded code, you can not count on that. Just because you have alarm A firing in 1000ms in one thread and alarm B firing in a different thread in 1001ms, that does not mean that you can assume A will fire before B. That's an error in multithreading code, and in my opinion, any mocking library needs to be able to help you test that even if B fires first, everything will still work as desired. Writing into the API that the only way to trigger alarms and such is by monotonically advancing the clock precludes this possibility from the beginning, however convenient it may be.

Abtime uses additional parameters to get around that.

You can still get into situations where testing is difficult or inconventient. However I have not been able to work out an API that is any better or safer, or I'd make a 2.0. Generally what's been provided by this library has been enough.

In my personal experince, the only thing I use the clock advancing for is test stability for code that needs to add time stamps for things. When testing triggers and alarms, it's often perfectly sufficient to just trigger them and not advance the clock at all. It's generally the logic you're testing, and the various paths code can execute through, not actual time handling itself. So in practice I don't generally miss having the nice "just advance the clock" API as much as you would think.

Stability

As I have been using this code for a while now and it has stopped changing, this is now at version 1.0.0.

Commit Signing

Starting with the commit after 3003eee879c, I will be signing this repository with the "jerf" keybase account. If you are viewing this repository through GitHub, you should see the commits as showing as "verified" in the commit view.

(Bear in mind that due to the nature of how git commit signing works, there may be runs of unverified commits; what matters is that the top one is signed.)