Compatibility for AdaCore’s legacy testsuites¶
Although all the default behaviors in e3.testsuite
presented in this
documentation should be fine for most new projects, it is not realistic to
require existing big testsuites to migrate to them. A lot of testsuites at
AdaCore use similar formalisms (atomic testcases, dedicated test directories,
…), but different formats: no test.yaml
file, custom files for test
execution control, etc.
These testsuites contain a huge number of testcases, and thus it is a better investment of time to introduce compatible settings in testsuite scripts rather than reformat all testcases. This section presents compatibility helpers for legacy AdaCore testsuites.
Test finder¶
The e3.testsuite.testcase_finder.AdaCoreLegacyTestFinder
class can act as a
drop-in test finder for legacy AdaCore testsuites: all directories whose name
matches a TN (Ticket Number), i.e. matching the
[0-9A-Z]{2}[0-9]{2}-[A-Z0-9]{3}
regular expression, are considered as
containing a testcase. Legacy AdaCore testsuites have only one driver, so this
test finder always use the same driver. For instance:
@property
def test_finders(self):
# This will create a testcase for all directories whose name matches a
# TN, using the MyDriver test driver.
return [AdaCoreLegacyTestFinder(MyDriver)]
Test control¶
AdaCore legacy testsuites rely on a custom file format to lead testcase
execution control: test.opt
files.
Similarly to the YAML-based control descriptions,
this format provides a declarative formalism to describe settings depending on
the environment, and more precisely on a set of discriminants: simple case
insensitive names for environment specificities. For instance: linux
on a
Linux system, windows
on a Windows one, x86
on Intel 32 bits
architecture, vxworks
when targetting a VxWorks is involved, etc.
A parser for such files is included in e3.testsuite
(see the
optfileparser
module), and most importantly, a TestControlCreator
subclass binds it to the rest of the testsuite framework:
AdaCoreLegacyTestControlCreator
, from the e3.testsuite.control
module.
Its constructor requires the list of discriminants used to selectively evaluate
test.opt
directives.
This file format not only controls test execution with its DEAD
, XFAIL
and SKIP
commands: it also allows to control the name of the script file to
run (CMD
command), the name of the output baseline file (OUT
), the time
limit for the script (RLIMIT
), etc. For this reason,
AdaCoreLegacyTestControlCreator
works best with the AdaCore legacy test
driver: see the next section.
Test driver¶
All legacy AdaCore testsuites use actual/expected test output comparisons to
determine if a test passes, so the reference test driver for them derives from
DiffTestDriver
: e3.testsuite.driver.adacore.AdaCoreLegacyTestDriver
.
This driver is coupled with a custom test execution control mechanism:
test.opt
files (see the previous section), and thus overrides the
test_control_creator
property accordingly.
This driver has two requirements for Testsuite
subclasses using it:
- Put a process environment (string dictionary) for subprocesses in
self.env.test_environ
. By default they can just put a copy of the testsuite’s own environment:dict(os.environ)
. - Put the list of discriminants (list of strings) in
self.env.discs
. For the latter, starting from the result of thee3.env.AbstractEnv.discriminants
property can help, as it computes standard discriminants based on the current host/build/target platforms. Testsuites can then add more discriminants as needed.
For instance, imagine a testsuite that wants standard dircriminants plus the
valgrind
discriminant if the --valgrind
command-line option is passed
to the testsuite:
class MyTestsuite(Testsuite):
def add_options(self, parser):
parser.add_argument("--valgrind", action="store_true",
help="Run tests under Valgrind")
def set_up(self):
super(MyTestsuite, self).set_up()
self.env.test_environ = dict(os.environ)
self.env.discs = self.env.discriminants
if self.env.options.valgrind:
self.env.discs.append("valgrind")
There is little point describing precisely the convoluted behavior for this driver, so we will stick here to a summary, with a few pointers to go further:
- All testcases must provide a script to run. Depending on testsuite defaults
(
AdaCoreLegacyTestControlCreator.default_script
property) and the content of eachtest.opt
testcase file, this script can be a Windows batch script (*.cmd
), a Bourne-compatible shell script (*.sh
) or a Python script (*.py
). - It is the output of this script that is compared against the output baseline.
To hide environment-specific differences, output refiners turn backslashes
into forward slashes, remove
.exe
extensions and also remove occurences of the working directory. - On Unix systems, this driver has a very crude conversion of Windows batch
script to Bourne-compatible scripts: text substitution remove some
.exe
extensions, replaces%VAR%
environment variable references with$VAR
, etc. SeeAdaCoreLegacyTestDriver.get_script_command_line
. Note that subclasses can override this method to automatically generate a test script.
Curious readers are invited to read the sources to know the details: doing so is necessary anyway to override specific behaviors so that this driver fits the precise need of some testsuite. Hopefully, this documentation and inline comments have made this process easier.