Manual eta expansion leads to orders of magnitude less allocations

While working on deterministic free variable computation I run into some performance issues. I narrowed it down to code that when eta expanded performed less allocations.

I ended up writing a benchmark and for large examples the eta expanded versions performs orders of magnitude less allocations.

The results of my test:

EtaExpanded test3
      50,450,112 bytes allocated in the heap
real 0.89

EtaReduced test3
   3,661,740,240 bytes allocated in the heap
real 3.66

The whole source is here: https://github.com/niteria/deterministic-fvs, I'm also attaching the two interesting implementations.

I experimented with it a little and GHC eta-expanded when I didn't have mutually recursive bindings.

Trac metadata

Trac field	Value
Version	7.10.2
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture