pg-transact

benchmarks

• The lack a with resource api for criterion is annoying

• abort is twice as fast as rollback

  Running 1 benchmarks...
  Benchmark benchmark: RUNNING...
  benchmarking rollback
  time                 114.1 μs   (112.2 μs .. 117.1 μs)
                       0.998 R²   (0.997 R² .. 1.000 R²)
  mean                 113.0 μs   (112.4 μs .. 113.9 μs)
  std dev              2.546 μs   (1.524 μs .. 4.065 μs)
  variance introduced by outliers: 18% (moderately inflated)
  
  benchmarking abort
  time                 53.08 μs   (52.89 μs .. 53.35 μs)
                       1.000 R²   (1.000 R² .. 1.000 R²)
  mean                 53.15 μs   (52.97 μs .. 53.56 μs)
  std dev              850.7 ns   (486.4 ns .. 1.343 μs)
  variance introduced by outliers: 11% (moderately inflated)

  but they are both fast I can’t see the difference mattering.

0.3.1.1

• No reason to have a abort take in an action. You can’t nest them. This is clearer if the type is abort :: DBT m a.
• Okay well I guess there is a reason. Throwing an abort is different then saying finally abort. So I think the current form is useful.
• However I think what is missing is a ROLLBACK in the runner.
• Scratch that. The runner already rollbacks on exceptions. I don’t need the extra rollback in abort call. Test pass. Ship it.