Test::Warnings - Test for warnings and the lack of them
version 0.033
use Test::More;
use Test::Warnings;
pass('yay!');
done_testing;
emits TAP:
ok 1 - yay!
ok 2 - no (unexpected) warnings (via done_testing)
1..2
and:
use Test::More tests => 3;
use Test::Warnings 0.005 ':all';
pass('yay!');
like(warning { warn "oh noes!" }, qr/^oh noes/, 'we warned');
emits TAP:
ok 1 - yay!
ok 2 - we warned
ok 3 - no (unexpected) warnings (via END block)
1..3
If you've ever tried to use Test::NoWarnings to confirm there are no warnings generated by your tests, combined with the convenience of done_testing
to not have to declare a test count, you'll have discovered that these two features do not play well together, as the test count will be calculated before the warnings test is run, resulting in a TAP error. (See examples/test_nowarnings.pl
in this distribution for a demonstration.)
This module is intended to be used as a drop-in replacement for Test::NoWarnings: it also adds an extra test, but runs this test before done_testing
calculates the test count, rather than after. It does this by hooking into done_testing
as well as via an END
block. You can declare a plan, or not, and things will still Just Work.
It is actually equivalent to:
use Test::NoWarnings 1.04 ':early';
as warnings are still printed normally as they occur. You are safe, and enthusiastically encouraged, to perform a global search-replace of the above with use Test::Warnings;
whether or not your tests have a plan.
It can also be used as a replacement for Test::Warn, if you wish to test the content of expected warnings; read on to find out how.
The following functions are available for import (not included by default; you can also get all of them by importing the tag :all
):
allow_warnings([bool])
- EXPERIMENTAL - MAY BE REMOVEDWhen passed a true value, or no value at all, subsequent warnings will not result in a test failure; when passed a false value, subsequent warnings will result in a test failure. Initial value is false
.
When warnings are allowed, any warnings will instead be emitted via Test::Builder::note.
allowing_warnings
- EXPERIMENTAL - MAY BE REMOVEDReturns whether we are currently allowing warnings (set by allow_warnings
as described above).
had_no_warnings(<optional test name>)
Tests whether there have been any warnings so far, not preceded by an allowing_warnings
call. It is run automatically at the end of all tests, but can also be called manually at any time, as often as desired.
warnings( { code } )
Given a code block, runs the block and returns a list of all the (not previously allowed via allow_warnings
) warnings issued within. This lets you test for the presence of warnings that you not only would allow, but must be issued. Testing functions are not provided; given the strings returned, you can test these yourself using your favourite testing functions, such as Test::More::is or Test::Deep::cmp_deeply.
You can use this construct as a replacement for Test::Warn::warnings_are:
is_deeply(
[ warnings { ... } ],
[
'warning message 1',
'warning message 2',
],
'got expected warnings',
);
or, to replace Test::Warn::warnings_like:
cmp_deeply(
[ warnings { ... } ],
bag( # ordering of messages doesn't matter
re(qr/warning message 1/),
re(qr/warning message 2/),
),
'got expected warnings (in any order)',
);
Warnings generated by this code block are NOT propagated further. However, since they are returned from this function with their filename and line numbers intact, you can re-issue them yourself immediately after calling warnings(...)
, if desired.
Note that use Test::Warnings 'warnings'
will give you a warnings
subroutine in your namespace (most likely main
, if you're writing a test), so you (or things you load) can't subsequently do warnings->import
-- it will result in the error: "Not enough arguments for Test::Warnings::warnings at ..., near "warnings->import"". To work around this, either use the fully-qualified form (Test::warnings
) or make your calls to the warnings
package first.
warning( { code } )
Same as warnings( { code } )
, except a scalar is always returned - the single warning produced, if there was one, or an arrayref otherwise -- which can be more convenient to use than warnings()
if you are expecting exactly one warning.
However, you are advised to capture the result from warning()
into a temp variable so you can dump its value if it doesn't contain what you expect. e.g. with this test:
like(
warning { foo() },
qr/^this is a warning/,
'got a warning from foo()',
);
if you get two warnings (or none) back instead of one, you'll get an arrayref, which will result in an unhelpful test failure message like:
# Failed test 'got a warning from foo()'
# at t/mytest.t line 10.
# 'ARRAY(0xdeadbeef)'
# doesn't match '(?^:^this is a warning)'
So instead, change your test to:
my $warning = warning { foo() };
like(
$warning,
qr/^this is a warning/,
'got a warning from foo()',
) or diag 'got warning(s): ', explain($warning);
allow_patterns(qr/always allow this warning/);
{
my $temp = allow_patterns(qr/only allow in this scope/, qr/another temporary warning/);
... stuff ...
}
Given one or more regular expressions, in qr/.../
form, add them to the allow-list (warnings will be emitted with note
rather than triggering the warning handler). If the return value is saved in a local variable, the warning exemption will only be in effect for that local scope (the addition is reversed at the end of the scope); otherwise, the effect is global.
Given one or more regular expressions, in qr/.../
form, remove it from the allow-list. The pattern must exactly match a pattern previously provided to "allow_patterns".
:all
Imports all functions listed above
:no_end_test
Disables the addition of a had_no_warnings
test via END
or done_testing
:fail_on_warning
When used, fail immediately when an unexempted warning is generated (as opposed to waiting until "had_no_warnings" or done_testing
is called).
I recommend you only turn this option on when debugging a test, to see where a surprise warning is coming from, and rely on the end-of-tests check otherwise.
:report_warnings
When used, had_no_warnings()
will print all the unexempted warning content, in case it had been suppressed earlier by other captures (such as "stderr_like" in Test::Output or "capture" in Capture::Tiny).
You can temporarily turn off the failure behaviour of this module, swapping it out for reporting (see :report_warnings
above) with:
$ENV{PERL_TEST_WARNINGS_ONLY_REPORT_WARNINGS} = 1;
This can be useful for working around problematic modules that have warnings in newer Perl versions.
Sometimes new warnings can appear in Perl that should not block installation -- for example, smartmatch was recently deprecated in perl 5.17.11, so now any distribution that uses smartmatch and also tests for warnings cannot be installed under 5.18.0. You might want to consider only making warnings fail tests in an author environment -- you can do this with the if pragma:
use if $ENV{AUTHOR_TESTING} || $ENV{RELEASE_TESTING}, 'Test::Warnings';
In future versions of this module, when interfaces are added to test the content of warnings, there will likely be additional sugar available to indicate that warnings should be checked only in author tests (or TODO when not in author testing), but will still provide exported subs. Comments are enthusiastically solicited - drop me an email, write up an RT ticket, or come by #perl-qa
on irc!
Achtung! This is not a great idea:
sub warning_like(&$;$) {
my ($code, $pattern, $name) = @_;
like( &warning($code), $pattern, $name );
}
warning_like( { ... }, qr/foo/, 'foo appears in the warning' );
If the code in the { ... }
is going to warn with a stack trace with the arguments to each subroutine in its call stack (for example via Carp::cluck
), the test name, "foo appears in the warning" will itself be matched by the regex (see examples/warning_like.t). Instead, write this:
like( warning { ... }, qr/foo/, 'foo appears in the warning' );
If you are using another module that sets its own warning handler (for example Devel::Confess or diagnostics) your results may be mixed, as those handlers will interfere with this module's ability to properly detect and capture warnings in their original form.
allow_warnings(qr/.../)
- allow some warnings and not others
more sophisticated handling in subtests - if we save some state on the Test::Builder object itself, we can allow warnings in a subtest and then the state will revert when the subtest ends, as well as check for warnings at the end of every subtest via done_testing
.
sugar for making failures TODO when testing outside an author environment
strictures - which makes all warnings fatal in tests, hence lessening the need for special warning testing
Bugs may be submitted through the RT bug tracker (or bug-Test-Warnings@rt.cpan.org).
There is also a mailing list available for users of this distribution, at http://lists.perl.org/list/perl-qa.html.
There is also an irc channel available for users of this distribution, at #perl
on irc.perl.org
.
I am also usually active on irc, as 'ether' at irc.perl.org
and irc.libera.chat
.
Karen Etheridge <ether@cpan.org>
Graham Knop <haarg@haarg.org>
A. Sinan Unur <nanis@cpan.org>
Leon Timmermans <fawaka@gmail.com>
Tina Mueller <cpan2@tinita.de>
This software is copyright (c) 2013 by Karen Etheridge.
This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.