Shimming for testing Perl 6 code released to CPAN
An entry of the Test Automation Tips discussing how we can use the Perl 5 testing system to run TAP producing test in any language.
This entry was first sent out as part of the Test Automation Tips. Visit here to subscribe.
Recently I have started to prepare a web site and I wanted to write the code behind it in Perl 6. For this I decided to reuse the Perl 6 port of HTML::Template that was created by the developers of the November wiki.
The problem is that it is part of the November project and it does not have a separate releases. I am not sure if November has any official release or they are just copying the files to the production machine? After all it is a very experimental project.
Running Perl 6 locally
Anyway as I was only running my code on my own development machine that was not a huge issue as I could use the checked out version of the November git repository by writing the following at the beginning of my code.
BEGIN {
@*INC.push('/path/to/november/p6w/');
}
This is of course not flexible enough and does not allow easy sharing of Perl 6 packages. So I decided I should try to put Perl 6 packages on CPAN and reuse the already existing packaging, *testing*, distribution, and installation system of Perl 5. Later as we learn more about what Perl 6 will need we can migrate over our code to a new kind of CPAN server.
Shimming for Testing
Packaging the code was not an issue but allowing the standard installation process to run the unit test was a challenge. The problem is that the
make test
phase of the standard Perl 5 installation process will run the test files with the current perl, which is perl 5.
It would be nice if I could tell Test::Harness to use some arbitrary executable with some arbitrary parameters that the package developer can define but it is not yet available. Luckily Adam Kennedy helped me out pointing to the idea of shim.
In our case that means that I renamed all my *.t test files to *.t6 files and during the first step of the installation process I create a small wrapper for every .t6 file using the same name but with the .t extension.
The .t file contains the following Perl 5 code:
use strict;
use warnings;
use Config;
use File::Spec;
my $sep = $Config{path_sep};
my $rakudo_dir = File::Spec->catdir($ENV{PARROT_DIR}, 'languages', 'rakudo');
my @libs = $ENV{PERL6LIB} ? split (/$sep/, $ENV{PERL6LIB}) : ();
$ENV{PERL6LIB} = join $sep, $rakudo_dir, @libs;
my $parrot = File::Spec->catfile($ENV{PARROT_DIR}, 'parrot')
. ($^O eq "MSWin32" ? '.exe' : '');
my $rakudo = File::Spec->catfile($ENV{PARROT_DIR},
'languages', 'rakudo', 'perl6.pbc');
(my $file = $0) =~ s/t$/t6/;
system "$parrot $rakudo $file";
That is, the only thing it will do is to run the .t6 files with the same name but using Rakudo over Parrot as the interpreter.
This means the TAP stream of my real tests will flow directly to the code executing the *.t files.
This way actually we could wrap TAP streams generated by any language such as Python, PHP or even JavaScript and integrate them into our standard reporting systems.
Improvements
Looking at the above code again I can see a couple of possibilities to improve it.
I should use exec instead of system just to keep the same process and to make sure the exit code does not get lost.
I think I could easily tell any of the packaging tools of Perl to look for some other files and not .t file such as shim_testname.pl files. That way I could keep my real test files as .t. I am not sure though how much added value that would bring to languages other than Perl 6. AFAIK Python or PHP tests files don't normally have .t extensions.
Shimming library
I could wrap both the above script and the code that generates the small shimming files into a module to make it easier to reuse.
e.g. Makefile.PL could contain something like this:
use Test::Shim
prepares => '*.t6',
executer => 'path/to/interpreter param param';
or
use Test::Shim
prepares => '*.t6',
recurse => 1,
executer => 'path/to/interpreter param param';
That would create a .t file for every .t6 file in the t/ subdirectory possibly recursing deeply.
The small .t file can then contain only
use Test::Shim run => 'filename.t6', with => 'path/to/interpreter param param';
that would run 'path/to/interpreter param param filename.t6'
Installing using CPAN.pm
After the testing was solved I tried to see if I can install the newly created package with CPAN.pm and indeed the command
cpan .
installed my new package into the standard Perl 5 installation tree.
I could then configure the PERL6LIB environment variable to list all the directories in the @INC of my perl 5.
Uploading to PAUSE and Indexing
Uploading to PAUSE is not an issue but then I got an e-mail from PAUSE telling me that it could not find any package in my distribution so it won't index anything.
It was not too surprising as PAUSE is very perl 5 oriented and it probably is trying to load my modules using perl 5 which it will fail. Even if it only parses my code looking for the "package Foo::Bar;" definition it won't find it as the Perl 6 code contains "class Foo::Bar;". I asked Andreas Koenig and he quickly replied that I should provide the indexing information in META.yml but I could not figure it out yet how exactly I am supposed to do it.
So for now you probably cannot use CPAN.pm directly to install the package. search.cpan.org does show it so you can check it out, it is called Perl6::Conf and probably you can also install it using pip which gets a URL of a .tar.gz file, fetches that file and then uses your already setup CPAN.pm to install all its dependencies and the module itself. It will work as this module does not have any dependency.
Dependencies
Which brings me to the probably last issue we will have to solve. The current standard build systems of Perl 5 will probably fail to recognize the existence of an already installed prerequisite Perl 6 module. We'll have to provide an alternative way to check if the modules our module requires are already installed and to tell CPAN.pm and CPANPLUS that they should install the dependencies before trying to install this module.
Conclusion
Well, this story might not be entirely about testing but I think the shimming technique suggested by Adam can be reused to integrate many testing libraries into one stream so it might be useful to you as well.
Comments and Discussion
If you would like to discuss this on a public mailing list, subscribe to the Test Automation discussion list
This entry was first sent out as part of the Test Automation Tips. Visit here to subscribe.