Benchmarks produce exceptions on 32bits OS

[fccm]

$ make bench
ocamlbuild -no-links -use-ocamlfind src/batteries.cma src/batteriesHelp.cmo src/batteriesThread.cma META src/batteries.cmxs src/batteriesThread.cmxs benchsuite/bench_int.native benchsuite/flip.native benchsuite/deque.native benchsuite/lines_of.native benchsuite/bitset.native benchsuite/bench_map.native benchsuite/bench_nreplace.native
Finished, 0 targets (0 cached) in 00:00:00.
Finished, 557 targets (526 cached) in 00:00:05.
rm -f bench.log
_build/benchsuite/bench_int.native | tee -a bench.log;   _build/benchsuite/flip.native | tee -a bench.log;   _build/benchsuite/deque.native | tee -a bench.log;   _build/benchsuite/lines_of.native | tee -a bench.log;   _build/benchsuite/bitset.native | tee -a bench.log;   _build/benchsuite/bench_map.native | tee -a bench.log;   _build/benchsuite/bench_nreplace.native | tee -a bench.log; 
Measuring: System Clock
Warming up
Fatal error: exception Invalid_argument("Array.make")
Estimating clock resolutionMeasuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Measuring: System Clock
Warming up
Fatal error: exception Invalid_argument("Array.make")
Estimating clock resolutionMeasuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Test with small maps (length = 500)
Fatal error: exception File "benchsuite/bench_map.ml", line 172, characters 4-10: Assertion failed
1000 iterations
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Benchmarking results are written to bench.log

[UnixJunkie]

Here is one of the problems with having tons of code "hanging around".
They also have a maintainance cost.
If you can send a PR to correct those, it should be merged.

[fccm]

I know what maintaining means :)
I assigned this issue to me ;)
I think this is probably something simple like the max-length of arrays that are smaller on a 32bit OS.

[gasche]

I cannot reproduce this on my machine.

(I vaguely remember @UnixJunkie suggesting that you are using a 32bit system, could this be related?)

There is one anomaly in the log that I can reproduce:

Fatal error: exception File "benchsuite/bench_map.ml", line 172, characters 4-10: Assertion failed

I will look into this.

[gasche]

I propose a fix for this bench_map failure in #1009.

[UnixJunkie]

Corrected by @gasche

[fccm]

testing with 59c7025:

make bench
[...]
For comparing 1000 pairs of random integers
mfp's compare (19.76 us) is probably (alpha=7.87%) same speed as
BatInt.compare (19.77 us) which is 3.2% faster than
naive compare (20.43 us) which is 18.4% faster than
external compare (25.03 us) which is 57.7% faster than
stdlib's compare (59.15 us)
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Measuring: System Clock
Warming up
Estimating clock resolutionFatal error: exception Invalid_argument("Array.make")
Test with small maps (length = 500)
1000 iterations
Measuring: System Clock
Warming up
Fatal error: exception Invalid_argument("Array.make")
Estimating clock resolutionFatal error: exception Sys_error("bench_nreplace.ml: No such file or directory")
Measuring: System Clock
Warming up
Fatal error: exception Invalid_argument("Array.make")
Estimating clock resolutionBenchmarking results are written to bench.log

[gasche]

I think that the issue that @fccm observes is 32bit-specific, not the one I fixed.

[UnixJunkie]

@fccm can you get the stack trace?

export OCAMLRUNPARAM=b
# run the tests on 32bits