Release::Checklist - A QA checklist for CPAN releases
What follows is a list of categories or subjects that touch the quality - or what an end-user percieves as such - of a distribution end might need some attention.
This document is aiming at up-river authors that want/need improving the release status of their distribution.
Work is underway in merging this list with The Berlin Consensus into a toolset and/or service.
Some of those can be (semi-)automated (in tests), others need action from the maintainer. Not all subjects may apply to your project.
Each subject will mention or list modules that might help in improving or preserving high standards.
Various guidelines are also available in the perl core documentation
$ perldoc perlmodstyle
Test, test and test. The more you test, the lower the chance you will break your code with small changes.
use strict; use warnings; use Test::More; : done_testing ();
Separate your module tests and your author tests. This will lower the number of dependencies. Check your Pod syntax, documentation coverage, spelling, and minimum perl version requirements in xt/. There is no need to check that when the distribution runs its tests in user-space.
xt/
t/ xt/
If possible, do not use Test::* modules that you do not actually require, however fancy they may be. See the point about dependencies.
If you are still using any additional Test:: module, do not mix your own code with the functionality from a/the module. Be consistent: use all or use nothing. That is: if the module you (now) use comes with features you had scripted yourself before using that module, replace them too.
Test::
If adding tests after a bug-fix, add at least two tests: one that tests that the required (fixed) behavior now passes and that the invalid behavior fails.
Check to see if your tests support running in parallel
$ prove -vwb -j8
If you have Test2::Harness installed, also test with yath
$ yath $ yath -j8
Add tests cases from issues or tickets to your own tests. Adding references to the tickets or issues creates a self-documenting structure, reasoning and history.
Every module you use is a module your release will depend on. If a new release of such module fails, the likelyhood of your release being unable to install increases. So only use modules that you need. Each dependency should be well-considered.
Depending on modules that are included in the perl5 CORE distribution do not have this problem, as you are sure that this module is available already. However, be careful to ensure that that module is a core module in the lowest version of perl that you claim to support. Check it with Module::CoreList
$ corelist File::Temp Data for 2018-04-20 File::Temp was first released with perl v5.6.1
Then there are two types of modules you can depend on: functional modules, like DBI and XML::LibXML, and developer convenience modules, like Modern::Perl
You cannot get around needing the first type of modules, but the convenience modules should only be used in (local) perl scripts and not in CPAN modules, so please do not add additional dependencies on modules/pragma's like sanity, Modern::Perl, common::sense, or nonsense.
However useful they might be in your own working environment and force you into behaving well, adding them as a requirement to a CPAN module will increase the complexity of the requirements to probably no good use, as they are unlikely to be found on all your targeted systems and add a chance to break.
There is no problem with you using those in your own (non-CPAN) scripts and modules, but please do not add needless dependencies.
Finally, make sure all your dependencies are declared in your META structure, so all CPAN clients know what to do before your module can be tested.
"prereqs" : { "build" : { "requires" : { "ExtUtils::MakeMaker" : "0" } }, "configure" : { "requires" : { "ExtUtils::MakeMaker" : "0" } }, "runtime" : { "requires" : { "perl" : "5.006", "Test::More" : "0.88" } } }
Be very careful in choosing dependencies that have a different license than your own distribution. However useful a module might be, if it has a more restrictive license than your distribution, it may be off limits to some users. That would mean that your distribution cannot be used either.
The reverse is also true: if your release has a more restrictive license than most releases on CPAN, it may lead to others avoiding depending on your code.
Make sure that you have a clear, consice, and short SYNOPSIS section. This section should show the most important code as simple and clear as possible. If you have 3500 methods in your class, do not list all of the there. Just show how to create the object and show the 4 methods that a beginner would use. Note that the user that reads the manual will appreciate complete, correct, and up-to-date documentation per method more than a complete list of available methods in the SYNOPSIS.
Make sure your documentation covers all your methods and/or functions and all edge cases are documented. If you have private functions, mentions that in the documentation, so users can read that they might disappear without warning. A special section about errors, error messages, and/or exceptions is also a very nice thing to have.
Make sure your pod is correct and can also be parsed by the pod-modules in the lowest version of perl you support (or mention that you need at least version whatever to read the pod as intended).
Not every developer is of native English tongue. And even if, they also make (spelling) mistakes. There are enough tools available to prevent public display of misspellings and typoes. Use them.
It is a good plan to have someone else proofread your documentation. If you can, ask three readers: one who knows about what the module is about, one who can be seen as an end-user of this modules without any knowledge about the internals, and last someone who has no clue about programming. You might be surprised of what they will find in the documentation as weird, unclear, or even plain wrong.
Have examples of your code. Preferably both in the EXAMPLES section of the pod, as in a folder named examples.
examples
It is good practice to use your example code/scripts in your documentation too, as that gives you a two-way check (additional to your tests). Even better if the test scripts can be used as examples.
Do not just test what you think would be used. There will be users that try to bend the rules and invent ways for your module to be useful that you would never think of.
If every line of your code is tested, not only do you prevent unexpected breakage, but you also make sure that most corner cases are tested. Besides that, it will probably confront you with questions like "What can I possibly do to get into this part of my code?", Which may lead to optimizations and other fun.
Remove dead code. It is in your (git) repository anyway.
This is a hard one. If your release/dist requires specific versions of other modules, try to create an environment where you test your distribution against the required version and a version that does not meet the minimum version.
If your module requires Foo::Bar-0.123 because it supports correct UTF-8 encoding/decoding, and you wrote a test for that, your release is apt to fail in an environment where Foo::Bar-0.023 is installed.
This gets really hard to set up if your release has different code for versions of perl and for versions of required modules, but it pays off eventually. Note that monitoring CPANTESTERS can be a huge help.
If your code resides on GitHub, you can setup hooks to Travis CI. Just compose a .travis.yml and enable the hook. This supports a variety of perl versions and an environment where you can install modules for just the tests you need.
Your Makefile.PL (or whatever initial file the build system you use requires) will have to state a minimal supported perl version that ends up in META.json and META.yml
Do not guess. It is easy to check with
$ perlver --blame test.pl -------------------------------------------------- File : test.pl Line : 3 Char : 14 Rule : _perl_5010_operators Version : 5.010 -------------------------------------------------- // --------------------------------------------------
If you have multiple perls installed on your system, test your module or release with all of the versions that you claim to support before doing the release. Best would be to test with a threaded perl and a non-threaded perl. If you can test with a mixture of -Duselongdouble and 32bit/64bit perls, that would be even better.
-Duselongdouble
$ perl -wc lib/Foo/Bar.pm
Repeat this on as many architectures as you can (i586, x64, IA64, PA-RISC, Sparc, PowerPC, …)
Repeat this on as many Operating Systems as you can (Linux, NetBSD, OSX, HP-UX, Solaris, Windows, OpenVMS, AIX, …)
Testing against a -Duselongdouble compiled perl will surface bad tests, e.g. tests that match against NVs like 2.1:
use Test::More; my $i = 21000000000000001; $i /= 10e15; is ($i, 2.1); done_testing;
with -Uuselongdouble:
-Uuselongdouble
ok 1 1..1
with -Duselongdouble:
not ok 1 # Failed test at -e line 1. # got: '2.1000000000000001' # expected: '2.1' 1..1 # Looks like you failed 1 test of 1.
will show that 2.25 is probably a better choice that 2.1.
If you use XS, make sure you (try to) support the widest range of perl versions. As using XS is quite often using more delicate areas of perl and perl internals, it is especially important to test for both threaded and unthreaded perl and - if supported - on operating systems that have different linking procedures than Linux. AIX and Windows are known to show deficiencies early.
Your code allocates memory. Not only for the code itself, but also for all data-structures, a stack and other resources (modules you use or require).
Creating circular references or (in XS) variables that do not get freed will cause leaks. You might not notice in your tests, but if a long running process hits leaks and crashes with out-of-memory after 4 days, that is a problem. Tracing memory leaks might be hard, but some help is available
If you have a perl available with ASAN (Address Sanitizer) enabled, your may find corruptions during compilation.
Some see CPANTS as a game, but many of the tests it puts on your release have a reason. Before you upload, you can check most of that to prevent unhappy users.
$ perl Makefile.PL $ make $ make test $ make dist $ cpants_lint.pl Foo-Bar-0.01.tar.gz Checked dist: Foo-Bar-0.01.tar.gz Score: 144.44% (26/18) Congratulations for building a 'perfect' distribution! $
Some problems only surface when you do a make clean or make distclean. The develop cycle normally only adds and changes files, and if you forget to add those to the MANIFEST, your distribution will be incomplete and is likely to fail on other systems, whereas your tests locally still keep passing.
make clean
make distclean
Check MANIFEST and MANIFEST.skip are complete.
$ make dist $ make distclean
Add a CONTRIBUTING.md or similar file to guide others to consistency that will match your style (or, in case of joint effort, the style as agreed upon by the developers).
There are helper modules to enforce a style (given a configuration) or try to help contributors to come up with a path/change than matches the project's style and layout. Again: consistency helps. A lot.
Make sure your meta-data matches the expected requirements. That can be achieved by using a generator that produces conform the most recent specifications or by using tools to check handcrafted META-files against the META spec 1.4 (2008) or META spec 2.0 (2011):
It is highly appreciated if you declare resources, like your public repository URL and the preferred way to communicate in your META.json:
"resources" : { "x_IRC" : "irc://irc.perl.org/#toolchain", "repository" : { "type" : "git", "url" : "https://github.com/Tux/Release-Checklist", "web" : "https://github.com/Tux/Release-Checklist" }, "bugtracker" : { "web" : "https://github.com/Tux/Release-Checklist/issues" } }
Those are recognized and shown in the top-left section on meta::cpan.
Use a sane versioning system that the rest of the world might understand. Do not use the MD5 of the current date and time related to the phase of the moon or versions that include quotes or spaces. Keep it simple and clear.
Make sure it is a versioning system that increments
Make sure your ChangeLog or Changes file is up-to-date. Your release procedure might check the most recent mentioned date in that
Check if your release matches previous performance (if appropriate)
Make a clear statement about your license. (or choose a default, but at least state it).
Some target areas require a license in order to allow a CPAN module to be installed.
Add a file that states the purpose of your distribution.
The README should state the purpose, the minimal envirenment to test, build, and run and possible license issue. If there is a need to amend or create configuration files or set up databases, that should be mentioned too.
If your module is on [github][https://github.com] or similar, regularly check on submitted issues and deal with them: either explain why it is not an issue or comment with alternatives or even better: this has been fixed now. Additionally, you should check on possible forks of your code and check the commits in that fork. There are people that make very interesting changes that you could use to improve your code. If the changes they made are already taken into your own code, or your code offers a (better) alternative to their change you could comment on their commits with additional information.
Likewise if you use [RT][https://rt.perl.org] as bug tracker.
You have had reasons to make the changes leading up to a new distribution. If you really care about the users of your module, you should check if your new release would break any of the CPAN modules that (indirectly) depend on your module by testing with your previous release and your upcoming release and see if the new release would cause the other module(s) to break.
used_by.pl will check the depending modules with the upcoming version.
Of course it is imposible to cover every possible situation here. The DarkPAN (uses of your module beyond what is registered on CPAN) is huge.
Copyright (C) 2015-2018 H.Merijn Brand. All rights reserved.
This library is free software; you can redistribute and/or modify it under the same terms as Perl itself.
To install Release::Checklist, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Release::Checklist
CPAN shell
perl -MCPAN -e shell install Release::Checklist
For more information on module installation, please visit the detailed CPAN module installation guide.