Skip to content

Latest commit

 

History

History
17 lines (13 loc) · 1.96 KB

README.md

File metadata and controls

17 lines (13 loc) · 1.96 KB
Raw

Day 3: Sometimes subproblems sustain state

Okay, not too bad here. My first pass has a lot of state management and object conversion that might be placed on methods, but there's not a nice boundary that I can immediately find. I'm not gonna write table tests for this, since it wasn't really helpful or needed, but I took the chance to write a benchmark. Simple. Side note - the -8 or whatever you get appended on your benchmark test name is the number of CPUs used to run the benchmark. By default, it's GOMAXPROCS. You can try different combos on the command line via:

go test ./... -bench=. -cpu=1,2,4,8
# BenchmarkSumWithStop     	   51266	     21286 ns/op
# BenchmarkSumWithStop-2   	   62448	     19139 ns/op
# BenchmarkSumWithStop-4   	   62978	     19035 ns/op
# BenchmarkSumWithStop-8   	   62290	     19211 ns/op

Takeaways:

  • The theme of this one was clearly regular expressions, parsing out huge chunks of text. Taking the chance to explore the regexp library a bit was useful - capture groups and NAMED capture groups are useful, although they left me craving some wrapper library for some of the bare-bones functionality of the core regexp library.
  • A lot of the state transition code smells to me, it's a lot of object conversion in ways that feels inelegant. The thought has to be - is this a one-to-one object conversion? For changePoint, it is; for scanRange, it actually takes many changePoint objects for one scanRange object, since a lot of changePoint objects don't actually change the scan range state. An interesting pattern to capture!
  • Sometimes subproblems sustain state, e.g. "don't decompose a problem if it requires global state". I tried to do part 2 line by line, but since the state of a line ("still multiplying") carries over to the next, it was easier just to concatenate all the line together rather than pass state between iterations. This is a fun inversion on day 2's problem, where each line was an independent problem ("report").