NAME

forkprove - forking prove

SYNOPSIS

  forkprove -MMoose -MCatalyst -MDBIx::Class -j8 -lr t

DESCRIPTION

forkprove is a forking version of prove that allows you to preload modules with -M and then run each test after forking, potentially makes most of the tests to run faster without an overhead of loading the same modules again and again, while having an isolated, clean testing environment with a fork.

It works well with parallel testing as well, combined with -j option.

CONFIGURATION

You can specify an environment variable PERL_FORKPROVE_IGNORE to contain a regular expression to ignore files to test with forkprove.

  PERL_FORKPROVE_IGNORE=no-preload forkprove -lr t

This will run the tests whose filenames and directory match no-preload with the standard perl execution via prove.

CAVEATS

Following kind of tests might fail when tested under forkprove. There are workarounds in forkprove to detect/fix some of them, so don't be surprised if your tests still pass :)

  • Tests that rely on the value of $FindBin::Bin.

  • Tests that relies on CHECK and INIT block to run correctly.

See also http://search.cpan.org/~flora/Test-Aggregate-0.364/lib/Test/Aggregate.pm#CAVEATS, many of which could apply to forkprove as well.

WHY

Test::Aggregate allows you to create a nested TAP output by running a whole bunch of .t files from a directory.

forkprove shares the basic idea, but it creates a whole new test environment by making a copy of the process with fork for each test, which means it could run your tests that rely on the singleton state etc. without modifying the code to run with Test::Aggregate.

TIPS

  • If you have a bunch of modules you want to load, put them in a file t/Preload.pm and load it:

      forkprove -Mt::Preload

    If you're lazy and just load all of .pm files in lib and t: use lib::require::all

      forkprove -Mlib::require::all=lib,t

PERFORMANCE

Your mileage might vary, if forkprove runs your tests faster or not. For lightweight modules, we've found that using forkprove doesn't make any siginificant difference, due to the fact that fork overhead nullifies the benefit of preloading modules.

On the other hand, test suite with Moose, Catalyst and Dist::Zilla seems to run 80-100% faster with forkprove and -jN option.

AUTHOR

Tatsuhiko Miyagawa

SEE ALSO

prove, TAP::Parser