NAME Test2::Plugin::Cover - Fast and Minimal file coverage info. DESCRIPTION This plugin will collect minimal file coverage info, and will do so with minimal performance impact. Every time a subroutine is called this tool will do its best to find the filename the subroutine was defined in, and add it to a list. Also, anytime you attempt to open a file with open() or sysopen() the file will be added to the list. This list will be attached to a test2 event just before the test exits. In most formatters the event will only show up as a comment on STDOUT # This test covered N source files. . However tools such as Test2::Harness::UI can make full use of the coverage information contained in the event. INTENDED USE CASE This tool is not intended to record comprehensive coverage information, if you want that use Devel::Cover. This tool is intended to obtain and maintain lists of files that were opened, or which define subs which were executed by any given test. This information is useful if you want to determine what test files to run after any given code change. The collected coverage data is contained in test2 events, if you use Test2::Harness aka yath then this data can be logged and consumed by other tools such as Test2::Harness::UI. PERFORMANCE Unlike tools that need to record comprehensive coverage (Devel::Cover), This module is only concerned about what files you open, or defined subs executed directly or indirectly by a given test file. As a result this module can get away with a tiny bit of XS code that only fires when a subroutine is called. Most coverage tools fire off XS for every statement. LIMITATIONS This tool uses XS to inject a little bit of C code that runs every time a subroutine is called, or every time open() or sysopen() is called. This C code obtains the next op that will be run and tries to pull the filename from it. eval, XS, Moose, and other magic can sometimes mask the filename, this module only makes a minimal attempt to find the filename in these cases. Originally this module only collected the filenames touched by a test. Now in addition to that data it can give you separate lists of files where subs were called, and files that were touched via open(). Additionally the sub list includes the info about what subs were called. In all of these cases it is also possible to know what sections of your test called the subs or opened the files. THINGS THAT WILL NOT SHOW UP goto goto &sub does not enter the sub the normal way, the target sub (and its file, if different) will not be recorded. The sub doing the goto is recorded normally. constants Constants created with use constant, and any other subs inlined at compile time, never trigger a runtime sub call, so the file defining them will not be recorded unless something else in it is called. XS subs XS subs have no perl source file. Calls to them are recorded against the file of the next perl statement that executes, which is usually the caller's file. threads Coverage data is collected per-thread and is not merged. Data collected inside spawned ithreads is lost unless you merge it yourself. exotic open() forms Only 2 and 3 argument open() calls are recorded. The list form for piping to programs (open($fh, '-|', $prog, @args)) and 1-arg open() are ignored. REAL EXAMPLES The following data was gathered using prove to run the full Moose test suite: # Prove on its own Files=478, Tests=17326, 64 wallclock secs ( 1.62 usr 0.46 sys + 57.27 cusr 4.92 csys = 64.27 CPU) # Prove with Test2::Plugin::Cover (no coverage event) Files=478, Tests=17326, 67 wallclock secs ( 1.61 usr 0.46 sys + 60.98 cusr 5.31 csys = 68.36 CPU) # Prove with Devel::Cover Files=478, Tests=17324, 963 wallclock secs ( 2.39 usr 0.58 sys + 929.12 cusr 31.98 csys = 964.07 CPU) no coverage event - No report was generated. This was done to only measure the effect of the XS that adds the data collection overhead, and not the cost of the perl code that generates the report event at the end of every test. The Moose test suite was also run using Test2::Harness aka yath # Without Test2::Plugin::Cover Wall Time: 62.51 seconds CPU Time: 69.13 seconds (usr: 1.84s | sys: 0.08s | cusr: 60.77s | csys: 6.44s) # With Test2::Plugin::Cover (no coverage event) Wall Time: 75.46 seconds CPU Time: 82.00 seconds (usr: 1.96s | sys: 0.05s | cusr: 72.64s | csys: 7.35s) As you can see, there is a performance hit, but it is fairly small, specially compared to Devel::Cover. This is not to say anything bad about Devel::Cover which is amazing, but a bad choice for the use case Test2::Plugin::Cover was written to address. SYNOPSIS INLINE use Test2::Plugin::Cover; ... # Arrayref of files covered so far my $covered_files = Test2::Plugin::Cover->files; COMMAND LINE You can tell prove to use the module this way: HARNESS_PERL_SWITCHES=-MTest2::Plugin::Cover prove ... For yath: yath test --cover-files ... SUPPRESS REPORT You can suppress the final report (only collect data, do not send the Test2 event) CLI: HARNESS_PERL_SWITCHES=-MTest2::Plugin::Cover=no_event,1 prove ... INLINE: use Test2::Plugin::Cover no_event => 1; KNOWING WHAT CALLED WHAT If you use a system like Test::Class, Test::Class::Moose, or Test2::Tools::Spec then you divide your tests into subtests (or similar). In these cases it would be nice to track what subtest (or equivalent) touched what files. There are 3 methods related to this, set_from(), get_from(), and clear_from() which you can use to manage this meta-data: subtest foo => sub { # Note, this is a simple string, but the 'from' data can also be a data # structure. Test2::Plugin::Cover->set_from("foo"); # subroutine() from Some.pm will be recorded as having been called by 'foo'. Some::subroutine(); Test2::Plugin::Cover->clear_from(); }; Doing this manually for all blocks is not ideal, ideally you would hook your tool, such as Test::Class to call set_from() and clear_from() for you. Adding such a hook is left as an exercise to the reader, and if you make one for a popular tool please upload it to cpan and add a ticket or send an email for me to link to it here. Once you have these hooks in place the data will not only show files and subs that were called, but what called them. Please see the set_from() documentation for details on values. CLASS METHODS $class->touch_source_file($file) $class->touch_source_file($file, $sub) $class->touch_source_file($file, \@subs) $class->touch_source_file($file, $subs, $from) This can be used to manually add coverage data. The first argument is the source file to be "touched" by coverage. The second argument is optional, and may be either a subroutine name, or an arrayref of subroutine names. The third argument is optional and can be used to override the default "from" value, which is normally determined for you automatically. If no subroutines are specified it will default to using '*', which means the entire file is considered to be touched. $class->touch_data_file($file) $class->touch_data_file($file, $from) This can be used to manually add coverage data. The first argument is the file to be "touched" by coverage data. Optionally you can override the 'from' value which is normally determined automatically. This is the same as calling $class->touch_source_file($file, '<>'). Both touch methods are no-ops while coverage is disabled, see disable(). $class->enable() $class->disable() $bool = $class->enabled() Toggle or check enabled status. When disabled no coverage is recorded. $class->reload() Reset filter if $0 or __FILE__ have changed. This is advanced usage, you will probably never need this. $val = $class->get_from() Get the current 'from' value. The default is '*' when nothing has set a from value. $class->set_from($val) Set a 'from' value. This can be anything, a string, a hashref, etc. Be advised though that it will usually be serialized to JSON, so make sure anything you put in it will be serializable as json. If the value is, or contains, a CODE or GLOB reference it cannot be serialized into the final report, so a warning will be issued and the value will be ignored (the previous 'from' value stays in effect). This is not fatal because enabling coverage should never introduce new exceptions into the code being observed. $class->clear_from() Resets the clear value to '*' $bool = $class->was_from_modified() This will return true if anything has called set_from() or set_from_manager. This can be reset back to false using reset_from(), which also clears the 'from' and 'from_manager' values. $class->set_from_manager($module) This should be set to a module that implements the following method: sub test_parameters { my $class = shift; my ($test_file, \@from_values) = @_; ... return { # If true - run the test # If false - skip the test # If not present or undef - run the test run => $bool, # The following are optional argv => [ ... ], env => { ... }, stdin => "...", }; # OR # If true - run the test # If false - skip the test # If undef or empty list - run the test return $bool; } This will be used by Test2::Harness to determine what data needs to be passed to a test given a set of 'from' values to instruct the test to run the necessary parts/subtests/groups/methods/etc. The 'argv' data will be prepended before any other arguments provided to the test. The 'env' hashref will be merged with any other env vars needed, with these taking priority. The 'stdin' string will be used as STDIN for the test. $arrayref = $class->files() $arrayref = $class->files(root => $path) This will return an arrayref of all files touched so far. The list of files will be sorted alphabetically, and duplicates will be removed. If a root path is provided it may be a Path::Tiny instance or a plain string. This path will be used to filter out any files not under the root directory. The running test file ($0) and this plugin's own file are always excluded from the results. $hashref = $class->data() $hashref = $class->data(root => $path) This returns the processed coverage data that goes into the report event: { # Files where subs were called 'lib/Foo.pm' => { some_sub => [ list of 'from' values that called it ], # Called BEGIN/END/etc blocks, or subs whose name could not be # determined, fall under '*'. '*' => [ ... ], }, # Files opened with open()/sysopen(), or touched as data files 'data.json' => { '<>' => [ ... ] }, } Duplicate 'from' values are removed (compared by content, not reference), and each list is sorted deterministically. The root parameter behaves as it does in files(). $event = $class->report(%options) This will send a Test2 event containing coverage information. It will also return the event. Options: root => Path::Tiny->new("...") Normally this is set to the current directory at module load-time. This is used to filter out any source files that do not live under the current directory. This may be a Path::Tiny instance or a plain string. verbose => $BOOL If this is set to true then the comment stating how many source files were touched will be printed as a diagnostics message instead so that it shows up without a verbose harness. ctx => DO NOT USE This is used ONLY when the Test2::API is doing its final book-keeping. Most users will never want to use this. $class->reset_coverage() This will completely clear all coverage data so far. $class->reset_from() This will clear the 'from' value, as well as reset the 'was_from_modified' state to false. $class->full_reset() Calls both reset_coverage() and reset_from(). $file_or_undef = $class->filter($file) $file_or_undef = $class->filter($file, root => Path::Tiny->new('...')) This method is used as a callback when getting the final list of covered source files. The default implementation removes any files that are not under the current directory which lets you focus on files in the distribution you are testing. You may return a modified filename if you wish to normalize it here, the default implementation will turn it into a relative path. If you provide a custom root parameter, it may be a Path::Tiny instance or a plain string. A custom filter callback should look something like this: sub { my $class = shift; my ($file, %params) = @_; # clean_filename() does not exist, it is just an example $file = clean_filename($file, %params); # should_show() does not exist, it is just an example return $file if should_show(%params); # Return undef or an empty list if you do NOT want to show the file. return; } Please take a look at the source to see what and how filter() is implemented if you want all the details on how it works. $file_or_undef = $class->extract($file) $file_or_undef = $class->extract($file, %params) This method is responsible for extracting a sensible filename from whatever the XS found. Some magic such as eval or Moose can set the filename to strings like '(eval 123)' or 'foo bar (defined at FILE line LINE)' or even nonsensical strings, or text with no filenames. If a sensible file name can be extracted it will be returned, otherwise undef (or an empty list) is returned. The default implementation does not use any parameters, but they are passed in for custom implementations to use. A custom extract callback should look something like this: sub { my $class = shift; my ($file, %params) = @_; # It is a valid file return $file if -e $file; # Do not use this, just an example return $1 if $file =~ m/($VALID_FILE_REGEX)/; # Cannot find a file here return; } TRACING OPENS For debugging you can ask the plugin to record where every open() and sysopen() call happened: $Test2::Plugin::Cover::TRACE_OPENS = 1; Every recorded open will push an arrayref onto @Test2::Plugin::Cover::OPENS: [$filename, $file, $line, $package] Where $filename is what was being opened, and $file, $line and $package describe the code doing the open. This is a debugging aid only, the data is not included in coverage events. SEE ALSO Devel::Cover is by far the best and most complete coverage tool for perl. If you need comprehensive coverage use Devel::Cover. Test2::Plugin::Cover is only better for a limited use case. SOURCE The source code repository for Test2-Plugin-Cover can be found at https://github.com/Test-More/Test2-Plugin-Cover. MAINTAINERS Chad Granum AUTHORS Chad Granum COPYRIGHT Copyright 2020 Chad Granum . This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself. See http://dev.perl.org/licenses/