[Chicago-talk] perl tests with overhead

[Greg Fast]

On Tue, Jul 26, 2005 at 09:57:57AM -0500, JT Smith wrote:
> The problem comes in that I don't want each test to have to seperately
> establish a session before running if I can help it. What I'd like to do
> is create a session and then pass each test a reference to the session.

I recently hacked together a solution to a similar testing problem,
using the previously-unknown-to-me Test::Harness::Straps.  This is the
chunk of Harness that runs the test scripts and analyzes/summarizes
the output ("1..251", "ok 1", etc).

I overrode Straps and swapped out the execution of "perl t/$n.t" with
my own test driver.  This allowed me to factor a large chunk of common
code out of a test suite (basically: configure a server, fork a
server, configure a client, run a client, join and cleanup).

The relevant bits:

  package MyStraps;
  use base 'Test::Harness::Straps';

  sub analyze_file {
    my $self = shift;
    my $case_file = shift;
  
    # run "<driver> <case_file>"
    my $cmd = $self->_command_line( $self->driver_cmd ) . " $case_file";
    open( my $fh, "$cmd |" ) || die $!;
    return $self->analyze_fh( $case_file, $fh );
  }

and:

  use Test::Harness;
  $Test::Harness::Strap = MyStraps->new();
  runtests( @cases ); # t/1.t, t/2.t, ...

So this lets you do arbitrary work for each test case.  In your
example, you could easily replace the exec of the test command ("perl
t/1.t") with a in-process execution that provides a shared session
(Straps has an analyze() method that takes plain text input), or you
could write a driver script that loads and stores the session before
providing it to an exec'd script (to avoid local pollution).

Normally when you do a "make test", it basically runs runtests().  I
haven't bothered to hack up Makefile.PL to make it use my straps
(because this is an integration test suite that requires external
setup), but it should be possible.


-- 
Greg Fast
gdf at speakeasy.net
http://cken.chi.groogroo.com/