Add mapAccumL'

[Daniel-Diaz]

A strict version of mapAccumL.

Motivation

I was getting

*** Exception: stack overflow

when querying for values very deep into the result of mapAccumL.

Solution

Once I switched to use the function added by this PR, it worked fine.

Thank you.

[Bodigrim]

Could it be expressed via foldr or other high-level combinator? Could there be rewrite rules for fusion similar to mapAccumLFB?

[Daniel-Diaz]

I'll take a look.

[Daniel-Diaz]

@Bodigrim

I wrote mapAccumL' following the definition for mapAccumL, and simply adding bang patterns on the accumulator. This works (not getting stack overflows, at least for now), but this benchmark shows the foldr version taking ~15% longer when I run it. I don't know whether I'm doing something wrong.

[Bodigrim]

The benchmark gives the same numbers on my machine, at least with GHC 9.10.2:

benchmarking mapAccumL/mapAccumL1
time                 30.84 μs   (30.80 μs .. 30.91 μs)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 30.89 μs   (30.84 μs .. 30.96 μs)
std dev              181.4 ns   (141.0 ns .. 232.0 ns)

benchmarking mapAccumL/mapAccumL2
time                 30.95 μs   (30.88 μs .. 31.04 μs)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 31.04 μs   (30.90 μs .. 31.38 μs)
std dev              646.1 ns   (180.1 ns .. 1.155 μs)
variance introduced by outliers: 18% (moderately inflated)

So let's proceed with the foldr version.

Could you please add tests, both to Properties.hs and Fusion.hs?

[Daniel-Diaz]

Thanks. That's good to know. I was using ghc-9.8.4. I'll see to add those tests later today.

[Daniel-Diaz]

@Bodigrim

Tests added in 90e40ab.

[Bodigrim]

Thanks!