The Official Perl 6 Test Suite (someday...)
This is the Official Perl 6 test suite. It evolved out of the pugs
test suite, and is now maintained in a separate repository.
Its purpose is to validate implementations that wish to be known
as a conforming Perl 6 implementation.
Please consider this test suite to be the result
of an ongoing negotiation: since many of these tests are inspired
by seeing how the various prototype implementations are screwed up
(or even more likely, how the design of Perl 6 is screwed up), this
test suite should be considered a work in progress until one or more
of the implementations gets close to passing the entire test suite,
at which point we will freeze version 6.0 of the test suite, and any
implementation that passes it can then call itself Perl 6.
As they develop, different implementations will certainly be in
different states of readiness with respect to the test suite, so
in order for the various implementations to track their progress
independently, we've established a mechanism for "fudging" the
tests in a kind of failsoft fashion. To pass a test officially,
an implementation must be able to run a test file unmodified, but an
implementation may (temporarily) skip tests or mark them as "todo" via
the fudging mechanism, which is implemented via the fudge preprocessor.
Individual implementations are not allowed to modify the actual test
code, but may insert line comments before each actual test (or block
of tests) that changes how those tests are to be treated for this
platform. The fudge preprocessor pays attention only to the comments
that belong to the current platform and ignores all the rest. If your
platform is named "humpty" then your special comment lines look like:
#?humpty: [NUM] VERB ARGS
(The colon is optional.)
The optional NUM says how many statements or blocks to apply the
verb to. (If not supplied, a value of 1 is assumed). A statement
is arbitrarily defined as one or more lines starting with a test call
and ending in semicolon (with an optional comment).
VERBs include:
skip "reason" # skip test entirely
eval "reason" # eval the test because it doesn't parse yet
try "reason" # try the test because it throws exception
todo "reason" # mark "todo" because "not ok" is expected
emit code # insert code (such as "skip_rest();") inline
All fudged tests return an exit code of 1 by default, so the test harness
will mark it as "dubious" even if all the tests supposedly pass.
There is also the following directive which modifies the test count of
the next construct:
#?DOES count
The count may be an expression as long as any variables referenced in
the expression are in scope at the location fudge eventually inserts a
"skip()" call.
When applied to a subsequent sub definition, registers the sub name as
doing that many tests when called. Note, however, that any skipping
is done at the point of the call, not within the subroutine, so the count
may not refer to any parameter of the sub.
When you run the fudge preprocessor, if it decides the test needs
fudging, it returns the new fudged filename; otherwise it returns
the original filename. (Generally you don't run "fudge" directly,
but your test harness runs the "fudgeall" program for you; see below.)
If there is already a fudged program in the directory that is newer
than the unfudged version, fudge just returns the fudged version
without regenerating it. If the fudged version is older, it removes
it and then decides anew whether to regenerate it based on the internal
fudge comments.
The "fudgeall" program may be called to process all the needed fudging
for a particular platform:
fudgeall humpty */*.t */*/*.t
will use the "fudge" program to translate any fudged files to a new
file where the extension is not *.t but instead is *.humpty to indicate
the platform dependency. It also returns the fudged list of filenames
to run, where unfudged tests are just passed through unchanged as *.t.
Each test comes through as either fudged or not, but never both.
The test harness then runs the selected test files as it normally
would (it shouldn't care whether they are named *.t or *.humpty).
In cases where the current working directory makes a difference, the tests
assume that the working directory is the root of the test suite, so that the
relative path to itself is t/spec/S\d\d-$section/$filename.