I have such a test in my Perl distribution template.

The test is a (not so) simple glue of Perl::PrereqScanner::Lite to find all used modules and some other stuff to remove known modules, and it (re)runs the test suite with each module blocked in turn.

This test basically works, except for the cases where I dynamically use different modules based (for example) on the operating system or other things that the test suite will not pick up on, because it (for exmample) only runs on Linux while a module is dynamically required on Windows.

Module::Used, Module::ScanDeps, Module::Extract::Use et al., Devel::TraceUse can all be useful in listing what modules you instructed perl to use (as opposed to what modules are "usefully used" which is what you want).

I assume the difficult bit will be to find a way to remove each of these modules from source in order to see the effect on compilation (as you suggested in a comment below). Removing from source code (e.g. with regex) can be trickier than finding them. Because finding them usually involves reading the %INC and I am not sure whether removing from %INC will have any effect at all.

One thing that comes to my mind, totally untested, is: instead of removing a module from source code and run, just prepend a dir in @INC which will contain "fake" module files (with empty content, bar the boilerplate). For example:

mkdir -p fakemodules/Data
echo "package Data::Dumper;1;" > fakemodules/Data/Dumper.pm
# this will fail:
perl -Ifakemodules -MData::Dumper -e'print Data::Dumper::Dumper([1])'
a=$?
# this succeeds:
perl -MData::Dumper -e'print Data::Dumper::Dumper([1])'
b=$?
if [ $a -ne 0 ] && [ $b -eq 0 ]; then echo "module is required"; fi
[download]

1 minute edit: I think choroba mentioned a similar technique for mocking modules and probably this is where I got it from

bw, bliako

Yes, that's one of the big problems with Perl modules. I have a fairly crude but working script which does this for C:

Remove unnecessary #include headers from C files

I saw this post yesterday:

Include-what-you-use: A tool to analyze includes in C and C++ source files (include-what-you-use.org)

But nothing at all for Perl. I guess what I'll probably go with, if nobody has a better idea, is something like that script for C files, like comment out the "use" line and see if it passes its tests.

That sounds very reasonable.

Self update: I've just seen App::perlimports mentioned on reddit, this might help. It seems to not actually detect unused modules by itself, but it adds a list of imports. So for example I found a script which has

use Math::Trig;

in it, but it doesn't actually use that, and App::perlimports changed that to

use Math::Trig ();

so I could tell it wasn't being used.