dist_surveyor - determine exactly what dist versions are installed
dist_surveyor [options] /some/perl/lib/dir
This utility examines all the modules installed within the specified perl library directory and uses the metacpan API to work out what versions of what distributions could have provided those modules. It then works out which of those candidate distributions is the most likely one.
It is fairly robust and copes well with edge cases like installation of non-released versions from git repos and local modifications.
Distributions are written to stdout. Progress and issues are reported to stderr.
It can take a long time to run for the first time on a directory with a large number of modules and candidate distributions. The data fetched from metacpan is cached so future runs are much faster. (The system this code was tested on took about 60 minutes to process around 500 distributions with no cached data, and under 10 minutes for later runs that could reuse the cached data. The cache file ended up about 40MB in size.)
--verbose Show more detailed progress --debug Show much more information --match R Ignore modules that don't match regex R (unanchored) --perlver V Ignore modules that are shipped with perl version V --remnants Include old distribution versions that have left old modules behind --uncached Don't use or update the persistent cache --makecpan D Create a CPAN repository in directory D --output S List of field names to output, separate by spaces. --format S Printf format string with a %s for each field in --output
Creates a CPAN repository in the specified directory by fetching the selected distributions into authors/id/... and writing the index files into modules/...
If the directory already exists then selected distributions that already exist are not refetched, any distributions that already exist but aren't selected by this run are left in place.
New package distribution information is merged into the modules/02packages index file.
Some additional files are written into a dist_surveyor subdirectory:
This file lists one unique 'token package' per distribution. It's very useful to speed up re-running a full install after some distributions have failed.
Run a survey and create a mini-CPAN repository containing the distributions:
dist_surveyor --makecpan my_cpan /some/perl/lib/dir > installed_dists.txt
It's important to give the correct perl lib directory path.
It's important to check the results related to any modules that generated warnings during the run.
Install those distributions into a new library:
cpanm --mirror file:$PWD/my_cpan [--mirror-only] -l new_lib < installed_dists.txt
That will always reinstall all the listed distributions. If some distributions fail to install (typically due to test failures) then it's much faster to use the 'token package list' on later runs:
cpanm --mirror file:$PWD/my_cpan [--mirror-only] -l new_lib < my_cpan/dist_surveyor/token_packages.txt
Using package name allows cpanm to skip those that it knows are already installed.
When using dist_surveyor to migrate perl versions, use the old perl to run dist_surveyor and the new perl to run cpanm.
* Polish up, refactor, add tests etc. Including making Dist::Surveyor a proper module that exports functions (or uses methods) and changing dist_surveyor to use that interface. * Auto-detect when directory given isn't the root of a perl library dir tree. E.g. by matching file names to module names * Add support for matching Foo.pm.PL files (e.g. FCGI and common::sense) * For installed modules get the file modification time (last commit time) and use it to eliminate candidate dists that were released after that time. * Consider factoring in release status ('authorized') so rogue releases that ship copies of many other modules (like Net-Braintree-0.1.1) are given a lower priority. * Sort out ExtUtils::Perllocal::Parser situation Avoid hard-coded %distro_key_mod_names related to perllocal.pod where the dist name doesn't match the key module name. Or maybe just remove use of distro_key_mod_names and perllocal entirely? * Optimise use of metacpan. Check caching. Use ElasticSearch.pm. * Fully handle merging of pre-existing --makecpan directory data files. * Consider factoring install date in the output ordering. May help with edge cases where a package P is installed via distros A then B. If A is reinstalled after B then the reinstalled P will be from A but should be from B. (I don't know of any cases, but it's certainly a possibility. The LWP breakup and Class::MOP spring to mind as possible candidates.)