b0d29bc47d
Having kyua in the base system will simplify automated testing in CI and eliminates bootstrapping issues on new platforms. The build of kyua is controlled by WITH(OUT)_TESTS_SUPPORT. Reviewed by: emaste Obtained from: CheriBSD Sponsored by: DARPA Differential Revision: https://reviews.freebsd.org/D24103
1374 lines
47 KiB
C++
1374 lines
47 KiB
C++
// Copyright 2014 The Kyua Authors.
|
|
// All rights reserved.
|
|
//
|
|
// Redistribution and use in source and binary forms, with or without
|
|
// modification, are permitted provided that the following conditions are
|
|
// met:
|
|
//
|
|
// * Redistributions of source code must retain the above copyright
|
|
// notice, this list of conditions and the following disclaimer.
|
|
// * Redistributions in binary form must reproduce the above copyright
|
|
// notice, this list of conditions and the following disclaimer in the
|
|
// documentation and/or other materials provided with the distribution.
|
|
// * Neither the name of Google Inc. nor the names of its contributors
|
|
// may be used to endorse or promote products derived from this software
|
|
// without specific prior written permission.
|
|
//
|
|
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
#include "engine/scheduler.hpp"
|
|
|
|
extern "C" {
|
|
#include <unistd.h>
|
|
}
|
|
|
|
#include <cstdio>
|
|
#include <cstdlib>
|
|
#include <fstream>
|
|
#include <memory>
|
|
#include <stdexcept>
|
|
|
|
#include "engine/config.hpp"
|
|
#include "engine/exceptions.hpp"
|
|
#include "engine/requirements.hpp"
|
|
#include "model/context.hpp"
|
|
#include "model/metadata.hpp"
|
|
#include "model/test_case.hpp"
|
|
#include "model/test_program.hpp"
|
|
#include "model/test_result.hpp"
|
|
#include "utils/config/tree.ipp"
|
|
#include "utils/datetime.hpp"
|
|
#include "utils/defs.hpp"
|
|
#include "utils/env.hpp"
|
|
#include "utils/format/macros.hpp"
|
|
#include "utils/fs/directory.hpp"
|
|
#include "utils/fs/exceptions.hpp"
|
|
#include "utils/fs/operations.hpp"
|
|
#include "utils/fs/path.hpp"
|
|
#include "utils/logging/macros.hpp"
|
|
#include "utils/noncopyable.hpp"
|
|
#include "utils/optional.ipp"
|
|
#include "utils/passwd.hpp"
|
|
#include "utils/process/executor.ipp"
|
|
#include "utils/process/status.hpp"
|
|
#include "utils/sanity.hpp"
|
|
#include "utils/stacktrace.hpp"
|
|
#include "utils/stream.hpp"
|
|
#include "utils/text/operations.ipp"
|
|
|
|
namespace config = utils::config;
|
|
namespace datetime = utils::datetime;
|
|
namespace executor = utils::process::executor;
|
|
namespace fs = utils::fs;
|
|
namespace logging = utils::logging;
|
|
namespace passwd = utils::passwd;
|
|
namespace process = utils::process;
|
|
namespace scheduler = engine::scheduler;
|
|
namespace text = utils::text;
|
|
|
|
using utils::none;
|
|
using utils::optional;
|
|
|
|
|
|
/// Timeout for the test case cleanup operation.
|
|
///
|
|
/// TODO(jmmv): This is here only for testing purposes. Maybe we should expose
|
|
/// this setting as part of the user_config.
|
|
datetime::delta scheduler::cleanup_timeout(60, 0);
|
|
|
|
|
|
/// Timeout for the test case listing operation.
|
|
///
|
|
/// TODO(jmmv): This is here only for testing purposes. Maybe we should expose
|
|
/// this setting as part of the user_config.
|
|
datetime::delta scheduler::list_timeout(300, 0);
|
|
|
|
|
|
namespace {
|
|
|
|
|
|
/// Magic exit status to indicate that the test case was probably skipped.
|
|
///
|
|
/// The test case was only skipped if and only if we return this exit code and
|
|
/// we find the skipped_cookie file on disk.
|
|
static const int exit_skipped = 84;
|
|
|
|
|
|
/// Text file containing the skip reason for the test case.
|
|
///
|
|
/// This will only be present within unique_work_directory if the test case
|
|
/// exited with the exit_skipped code. However, there is no guarantee that the
|
|
/// file is there (say if the test really decided to exit with code exit_skipped
|
|
/// on its own).
|
|
static const char* skipped_cookie = "skipped.txt";
|
|
|
|
|
|
/// Mapping of interface names to interface definitions.
|
|
typedef std::map< std::string, std::shared_ptr< scheduler::interface > >
|
|
interfaces_map;
|
|
|
|
|
|
/// Mapping of interface names to interface definitions.
|
|
///
|
|
/// Use register_interface() to add an entry to this global table.
|
|
static interfaces_map interfaces;
|
|
|
|
|
|
/// Scans the contents of a directory and appends the file listing to a file.
|
|
///
|
|
/// \param dir_path The directory to scan.
|
|
/// \param output_file The file to which to append the listing.
|
|
///
|
|
/// \throw engine::error If there are problems listing the files.
|
|
static void
|
|
append_files_listing(const fs::path& dir_path, const fs::path& output_file)
|
|
{
|
|
std::ofstream output(output_file.c_str(), std::ios::app);
|
|
if (!output)
|
|
throw engine::error(F("Failed to open output file %s for append")
|
|
% output_file);
|
|
try {
|
|
std::set < std::string > names;
|
|
|
|
const fs::directory dir(dir_path);
|
|
for (fs::directory::const_iterator iter = dir.begin();
|
|
iter != dir.end(); ++iter) {
|
|
if (iter->name != "." && iter->name != "..")
|
|
names.insert(iter->name);
|
|
}
|
|
|
|
if (!names.empty()) {
|
|
output << "Files left in work directory after failure: "
|
|
<< text::join(names, ", ") << '\n';
|
|
}
|
|
} catch (const fs::error& e) {
|
|
throw engine::error(F("Cannot append files listing to %s: %s")
|
|
% output_file % e.what());
|
|
}
|
|
}
|
|
|
|
|
|
/// Maintenance data held while a test is being executed.
|
|
///
|
|
/// This data structure exists from the moment when a test is executed via
|
|
/// scheduler::spawn_test() or scheduler::impl::spawn_cleanup() to when it is
|
|
/// cleaned up with result_handle::cleanup().
|
|
///
|
|
/// This is a base data type intended to be extended for the test and cleanup
|
|
/// cases so that each contains only the relevant data.
|
|
struct exec_data : utils::noncopyable {
|
|
/// Test program data for this test case.
|
|
const model::test_program_ptr test_program;
|
|
|
|
/// Name of the test case.
|
|
const std::string test_case_name;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param test_program_ Test program data for this test case.
|
|
/// \param test_case_name_ Name of the test case.
|
|
exec_data(const model::test_program_ptr test_program_,
|
|
const std::string& test_case_name_) :
|
|
test_program(test_program_), test_case_name(test_case_name_)
|
|
{
|
|
}
|
|
|
|
/// Destructor.
|
|
virtual ~exec_data(void)
|
|
{
|
|
}
|
|
};
|
|
|
|
|
|
/// Maintenance data held while a test is being executed.
|
|
struct test_exec_data : public exec_data {
|
|
/// Test program-specific execution interface.
|
|
const std::shared_ptr< scheduler::interface > interface;
|
|
|
|
/// User configuration passed to the execution of the test. We need this
|
|
/// here to recover it later when chaining the execution of a cleanup
|
|
/// routine (if any).
|
|
const config::tree user_config;
|
|
|
|
/// Whether this test case still needs to have its cleanup routine executed.
|
|
///
|
|
/// This is set externally when the cleanup routine is actually invoked to
|
|
/// denote that no further attempts shall be made at cleaning this up.
|
|
bool needs_cleanup;
|
|
|
|
/// The exit_handle for this test once it has completed.
|
|
///
|
|
/// This is set externally when the test case has finished, as we need this
|
|
/// information to invoke the followup cleanup routine in the right context,
|
|
/// as indicated by needs_cleanup.
|
|
optional< executor::exit_handle > exit_handle;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param test_program_ Test program data for this test case.
|
|
/// \param test_case_name_ Name of the test case.
|
|
/// \param interface_ Test program-specific execution interface.
|
|
/// \param user_config_ User configuration passed to the test.
|
|
test_exec_data(const model::test_program_ptr test_program_,
|
|
const std::string& test_case_name_,
|
|
const std::shared_ptr< scheduler::interface > interface_,
|
|
const config::tree& user_config_) :
|
|
exec_data(test_program_, test_case_name_),
|
|
interface(interface_), user_config(user_config_)
|
|
{
|
|
const model::test_case& test_case = test_program->find(test_case_name);
|
|
needs_cleanup = test_case.get_metadata().has_cleanup();
|
|
}
|
|
};
|
|
|
|
|
|
/// Maintenance data held while a test cleanup routine is being executed.
|
|
///
|
|
/// Instances of this object are related to a previous test_exec_data, as
|
|
/// cleanup routines can only exist once the test has been run.
|
|
struct cleanup_exec_data : public exec_data {
|
|
/// The exit handle of the test. This is necessary so that we can return
|
|
/// the correct exit_handle to the user of the scheduler.
|
|
executor::exit_handle body_exit_handle;
|
|
|
|
/// The final result of the test's body. This is necessary to compute the
|
|
/// right return value for a test with a cleanup routine: the body result is
|
|
/// respected if it is a "bad" result; else the result of the cleanup
|
|
/// routine is used if it has failed.
|
|
model::test_result body_result;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param test_program_ Test program data for this test case.
|
|
/// \param test_case_name_ Name of the test case.
|
|
/// \param body_exit_handle_ If not none, exit handle of the body
|
|
/// corresponding to the cleanup routine represented by this exec_data.
|
|
/// \param body_result_ If not none, result of the body corresponding to the
|
|
/// cleanup routine represented by this exec_data.
|
|
cleanup_exec_data(const model::test_program_ptr test_program_,
|
|
const std::string& test_case_name_,
|
|
const executor::exit_handle& body_exit_handle_,
|
|
const model::test_result& body_result_) :
|
|
exec_data(test_program_, test_case_name_),
|
|
body_exit_handle(body_exit_handle_), body_result(body_result_)
|
|
{
|
|
}
|
|
};
|
|
|
|
|
|
/// Shared pointer to exec_data.
|
|
///
|
|
/// We require this because we want exec_data to not be copyable, and thus we
|
|
/// cannot just store it in the map without move constructors.
|
|
typedef std::shared_ptr< exec_data > exec_data_ptr;
|
|
|
|
|
|
/// Mapping of active PIDs to their maintenance data.
|
|
typedef std::map< int, exec_data_ptr > exec_data_map;
|
|
|
|
|
|
/// Enforces a test program to hold an absolute path.
|
|
///
|
|
/// TODO(jmmv): This function (which is a pretty ugly hack) exists because we
|
|
/// want the interface hooks to receive a test_program as their argument.
|
|
/// However, those hooks run after the test program has been isolated, which
|
|
/// means that the current directory has changed since when the test_program
|
|
/// objects were created. This causes the absolute_path() method of
|
|
/// test_program to return bogus values if the internal representation of their
|
|
/// path is relative. We should fix somehow: maybe making the fs module grab
|
|
/// its "current_path" view at program startup time; or maybe by grabbing the
|
|
/// current path at test_program creation time; or maybe something else.
|
|
///
|
|
/// \param program The test program to modify.
|
|
///
|
|
/// \return A new test program whose internal paths are absolute.
|
|
static model::test_program
|
|
force_absolute_paths(const model::test_program program)
|
|
{
|
|
const std::string& relative = program.relative_path().str();
|
|
const std::string absolute = program.absolute_path().str();
|
|
|
|
const std::string root = absolute.substr(
|
|
0, absolute.length() - relative.length());
|
|
|
|
return model::test_program(
|
|
program.interface_name(),
|
|
program.relative_path(), fs::path(root),
|
|
program.test_suite_name(),
|
|
program.get_metadata(), program.test_cases());
|
|
}
|
|
|
|
|
|
/// Functor to list the test cases of a test program.
|
|
class list_test_cases {
|
|
/// Interface of the test program to execute.
|
|
std::shared_ptr< scheduler::interface > _interface;
|
|
|
|
/// Test program to execute.
|
|
const model::test_program _test_program;
|
|
|
|
/// User-provided configuration variables.
|
|
const config::tree& _user_config;
|
|
|
|
public:
|
|
/// Constructor.
|
|
///
|
|
/// \param interface Interface of the test program to execute.
|
|
/// \param test_program Test program to execute.
|
|
/// \param user_config User-provided configuration variables.
|
|
list_test_cases(
|
|
const std::shared_ptr< scheduler::interface > interface,
|
|
const model::test_program* test_program,
|
|
const config::tree& user_config) :
|
|
_interface(interface),
|
|
_test_program(force_absolute_paths(*test_program)),
|
|
_user_config(user_config)
|
|
{
|
|
}
|
|
|
|
/// Body of the subprocess.
|
|
void
|
|
operator()(const fs::path& /* control_directory */)
|
|
{
|
|
const config::properties_map vars = scheduler::generate_config(
|
|
_user_config, _test_program.test_suite_name());
|
|
_interface->exec_list(_test_program, vars);
|
|
}
|
|
};
|
|
|
|
|
|
/// Functor to execute a test program in a child process.
|
|
class run_test_program {
|
|
/// Interface of the test program to execute.
|
|
std::shared_ptr< scheduler::interface > _interface;
|
|
|
|
/// Test program to execute.
|
|
const model::test_program _test_program;
|
|
|
|
/// Name of the test case to execute.
|
|
const std::string& _test_case_name;
|
|
|
|
/// User-provided configuration variables.
|
|
const config::tree& _user_config;
|
|
|
|
/// Verifies if the test case needs to be skipped or not.
|
|
///
|
|
/// We could very well run this on the scheduler parent process before
|
|
/// issuing the fork. However, doing this here in the child process is
|
|
/// better for two reasons: first, it allows us to continue using the simple
|
|
/// spawn/wait abstraction of the scheduler; and, second, we parallelize the
|
|
/// requirements checks among tests.
|
|
///
|
|
/// \post If the test's preconditions are not met, the caller process is
|
|
/// terminated with a special exit code and a "skipped cookie" is written to
|
|
/// the disk with the reason for the failure.
|
|
///
|
|
/// \param skipped_cookie_path File to create with the skip reason details
|
|
/// if this test is skipped.
|
|
void
|
|
do_requirements_check(const fs::path& skipped_cookie_path)
|
|
{
|
|
const model::test_case& test_case = _test_program.find(
|
|
_test_case_name);
|
|
|
|
const std::string skip_reason = engine::check_reqs(
|
|
test_case.get_metadata(), _user_config,
|
|
_test_program.test_suite_name(),
|
|
fs::current_path());
|
|
if (skip_reason.empty())
|
|
return;
|
|
|
|
std::ofstream output(skipped_cookie_path.c_str());
|
|
if (!output) {
|
|
std::perror((F("Failed to open %s for write") %
|
|
skipped_cookie_path).str().c_str());
|
|
std::abort();
|
|
}
|
|
output << skip_reason;
|
|
output.close();
|
|
|
|
// Abruptly terminate the process. We don't want to run any destructors
|
|
// inherited from the parent process by mistake, which could, for
|
|
// example, delete our own control files!
|
|
::_exit(exit_skipped);
|
|
}
|
|
|
|
public:
|
|
/// Constructor.
|
|
///
|
|
/// \param interface Interface of the test program to execute.
|
|
/// \param test_program Test program to execute.
|
|
/// \param test_case_name Name of the test case to execute.
|
|
/// \param user_config User-provided configuration variables.
|
|
run_test_program(
|
|
const std::shared_ptr< scheduler::interface > interface,
|
|
const model::test_program_ptr test_program,
|
|
const std::string& test_case_name,
|
|
const config::tree& user_config) :
|
|
_interface(interface),
|
|
_test_program(force_absolute_paths(*test_program)),
|
|
_test_case_name(test_case_name),
|
|
_user_config(user_config)
|
|
{
|
|
}
|
|
|
|
/// Body of the subprocess.
|
|
///
|
|
/// \param control_directory The testcase directory where files will be
|
|
/// read from.
|
|
void
|
|
operator()(const fs::path& control_directory)
|
|
{
|
|
const model::test_case& test_case = _test_program.find(
|
|
_test_case_name);
|
|
if (test_case.fake_result())
|
|
::_exit(EXIT_SUCCESS);
|
|
|
|
do_requirements_check(control_directory / skipped_cookie);
|
|
|
|
const config::properties_map vars = scheduler::generate_config(
|
|
_user_config, _test_program.test_suite_name());
|
|
_interface->exec_test(_test_program, _test_case_name, vars,
|
|
control_directory);
|
|
}
|
|
};
|
|
|
|
|
|
/// Functor to execute a test program in a child process.
|
|
class run_test_cleanup {
|
|
/// Interface of the test program to execute.
|
|
std::shared_ptr< scheduler::interface > _interface;
|
|
|
|
/// Test program to execute.
|
|
const model::test_program _test_program;
|
|
|
|
/// Name of the test case to execute.
|
|
const std::string& _test_case_name;
|
|
|
|
/// User-provided configuration variables.
|
|
const config::tree& _user_config;
|
|
|
|
public:
|
|
/// Constructor.
|
|
///
|
|
/// \param interface Interface of the test program to execute.
|
|
/// \param test_program Test program to execute.
|
|
/// \param test_case_name Name of the test case to execute.
|
|
/// \param user_config User-provided configuration variables.
|
|
run_test_cleanup(
|
|
const std::shared_ptr< scheduler::interface > interface,
|
|
const model::test_program_ptr test_program,
|
|
const std::string& test_case_name,
|
|
const config::tree& user_config) :
|
|
_interface(interface),
|
|
_test_program(force_absolute_paths(*test_program)),
|
|
_test_case_name(test_case_name),
|
|
_user_config(user_config)
|
|
{
|
|
}
|
|
|
|
/// Body of the subprocess.
|
|
///
|
|
/// \param control_directory The testcase directory where cleanup will be
|
|
/// run from.
|
|
void
|
|
operator()(const fs::path& control_directory)
|
|
{
|
|
const config::properties_map vars = scheduler::generate_config(
|
|
_user_config, _test_program.test_suite_name());
|
|
_interface->exec_cleanup(_test_program, _test_case_name, vars,
|
|
control_directory);
|
|
}
|
|
};
|
|
|
|
|
|
/// Obtains the right scheduler interface for a given test program.
|
|
///
|
|
/// \param name The name of the interface of the test program.
|
|
///
|
|
/// \return An scheduler interface.
|
|
std::shared_ptr< scheduler::interface >
|
|
find_interface(const std::string& name)
|
|
{
|
|
const interfaces_map::const_iterator iter = interfaces.find(name);
|
|
PRE(interfaces.find(name) != interfaces.end());
|
|
return (*iter).second;
|
|
}
|
|
|
|
|
|
} // anonymous namespace
|
|
|
|
|
|
void
|
|
scheduler::interface::exec_cleanup(
|
|
const model::test_program& /* test_program */,
|
|
const std::string& /* test_case_name */,
|
|
const config::properties_map& /* vars */,
|
|
const utils::fs::path& /* control_directory */) const
|
|
{
|
|
// Most test interfaces do not support standalone cleanup routines so
|
|
// provide a default implementation that does nothing.
|
|
UNREACHABLE_MSG("exec_cleanup not implemented for an interface that "
|
|
"supports standalone cleanup routines");
|
|
}
|
|
|
|
|
|
/// Internal implementation of a lazy_test_program.
|
|
struct engine::scheduler::lazy_test_program::impl : utils::noncopyable {
|
|
/// Whether the test cases list has been yet loaded or not.
|
|
bool _loaded;
|
|
|
|
/// User configuration to pass to the test program list operation.
|
|
config::tree _user_config;
|
|
|
|
/// Scheduler context to use to load test cases.
|
|
scheduler::scheduler_handle& _scheduler_handle;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param user_config_ User configuration to pass to the test program list
|
|
/// operation.
|
|
/// \param scheduler_handle_ Scheduler context to use when loading test
|
|
/// cases.
|
|
impl(const config::tree& user_config_,
|
|
scheduler::scheduler_handle& scheduler_handle_) :
|
|
_loaded(false), _user_config(user_config_),
|
|
_scheduler_handle(scheduler_handle_)
|
|
{
|
|
}
|
|
};
|
|
|
|
|
|
/// Constructs a new test program.
|
|
///
|
|
/// \param interface_name_ Name of the test program interface.
|
|
/// \param binary_ The name of the test program binary relative to root_.
|
|
/// \param root_ The root of the test suite containing the test program.
|
|
/// \param test_suite_name_ The name of the test suite this program belongs to.
|
|
/// \param md_ Metadata of the test program.
|
|
/// \param user_config_ User configuration to pass to the scheduler.
|
|
/// \param scheduler_handle_ Scheduler context to use to load test cases.
|
|
scheduler::lazy_test_program::lazy_test_program(
|
|
const std::string& interface_name_,
|
|
const fs::path& binary_,
|
|
const fs::path& root_,
|
|
const std::string& test_suite_name_,
|
|
const model::metadata& md_,
|
|
const config::tree& user_config_,
|
|
scheduler::scheduler_handle& scheduler_handle_) :
|
|
test_program(interface_name_, binary_, root_, test_suite_name_, md_,
|
|
model::test_cases_map()),
|
|
_pimpl(new impl(user_config_, scheduler_handle_))
|
|
{
|
|
}
|
|
|
|
|
|
/// Gets or loads the list of test cases from the test program.
|
|
///
|
|
/// \return The list of test cases provided by the test program.
|
|
const model::test_cases_map&
|
|
scheduler::lazy_test_program::test_cases(void) const
|
|
{
|
|
_pimpl->_scheduler_handle.check_interrupt();
|
|
|
|
if (!_pimpl->_loaded) {
|
|
const model::test_cases_map tcs = _pimpl->_scheduler_handle.list_tests(
|
|
this, _pimpl->_user_config);
|
|
|
|
// Due to the restrictions on when set_test_cases() may be called (as a
|
|
// way to lazily initialize the test cases list before it is ever
|
|
// returned), this cast is valid.
|
|
const_cast< scheduler::lazy_test_program* >(this)->set_test_cases(tcs);
|
|
|
|
_pimpl->_loaded = true;
|
|
|
|
_pimpl->_scheduler_handle.check_interrupt();
|
|
}
|
|
|
|
INV(_pimpl->_loaded);
|
|
return test_program::test_cases();
|
|
}
|
|
|
|
|
|
/// Internal implementation for the result_handle class.
|
|
struct engine::scheduler::result_handle::bimpl : utils::noncopyable {
|
|
/// Generic executor exit handle for this result handle.
|
|
executor::exit_handle generic;
|
|
|
|
/// Mutable pointer to the corresponding scheduler state.
|
|
///
|
|
/// This object references a member of the scheduler_handle that yielded
|
|
/// this result_handle instance. We need this direct access to clean up
|
|
/// after ourselves when the result is destroyed.
|
|
exec_data_map& all_exec_data;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param generic_ Generic executor exit handle for this result handle.
|
|
/// \param [in,out] all_exec_data_ Global object keeping track of all active
|
|
/// executions for an scheduler. This is a pointer to a member of the
|
|
/// scheduler_handle object.
|
|
bimpl(const executor::exit_handle generic_, exec_data_map& all_exec_data_) :
|
|
generic(generic_), all_exec_data(all_exec_data_)
|
|
{
|
|
}
|
|
|
|
/// Destructor.
|
|
~bimpl(void)
|
|
{
|
|
LD(F("Removing %s from all_exec_data") % generic.original_pid());
|
|
all_exec_data.erase(generic.original_pid());
|
|
}
|
|
};
|
|
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param pbimpl Constructed internal implementation.
|
|
scheduler::result_handle::result_handle(std::shared_ptr< bimpl > pbimpl) :
|
|
_pbimpl(pbimpl)
|
|
{
|
|
}
|
|
|
|
|
|
/// Destructor.
|
|
scheduler::result_handle::~result_handle(void)
|
|
{
|
|
}
|
|
|
|
|
|
/// Cleans up the test case results.
|
|
///
|
|
/// This function should be called explicitly as it provides the means to
|
|
/// control any exceptions raised during cleanup. Do not rely on the destructor
|
|
/// to clean things up.
|
|
///
|
|
/// \throw engine::error If the cleanup fails, especially due to the inability
|
|
/// to remove the work directory.
|
|
void
|
|
scheduler::result_handle::cleanup(void)
|
|
{
|
|
_pbimpl->generic.cleanup();
|
|
}
|
|
|
|
|
|
/// Returns the original PID corresponding to this result.
|
|
///
|
|
/// \return An exec_handle.
|
|
int
|
|
scheduler::result_handle::original_pid(void) const
|
|
{
|
|
return _pbimpl->generic.original_pid();
|
|
}
|
|
|
|
|
|
/// Returns the timestamp of when spawn_test was called.
|
|
///
|
|
/// \return A timestamp.
|
|
const datetime::timestamp&
|
|
scheduler::result_handle::start_time(void) const
|
|
{
|
|
return _pbimpl->generic.start_time();
|
|
}
|
|
|
|
|
|
/// Returns the timestamp of when wait_any_test returned this object.
|
|
///
|
|
/// \return A timestamp.
|
|
const datetime::timestamp&
|
|
scheduler::result_handle::end_time(void) const
|
|
{
|
|
return _pbimpl->generic.end_time();
|
|
}
|
|
|
|
|
|
/// Returns the path to the test-specific work directory.
|
|
///
|
|
/// This is guaranteed to be clear of files created by the scheduler.
|
|
///
|
|
/// \return The path to a directory that exists until cleanup() is called.
|
|
fs::path
|
|
scheduler::result_handle::work_directory(void) const
|
|
{
|
|
return _pbimpl->generic.work_directory();
|
|
}
|
|
|
|
|
|
/// Returns the path to the test's stdout file.
|
|
///
|
|
/// \return The path to a file that exists until cleanup() is called.
|
|
const fs::path&
|
|
scheduler::result_handle::stdout_file(void) const
|
|
{
|
|
return _pbimpl->generic.stdout_file();
|
|
}
|
|
|
|
|
|
/// Returns the path to the test's stderr file.
|
|
///
|
|
/// \return The path to a file that exists until cleanup() is called.
|
|
const fs::path&
|
|
scheduler::result_handle::stderr_file(void) const
|
|
{
|
|
return _pbimpl->generic.stderr_file();
|
|
}
|
|
|
|
|
|
/// Internal implementation for the test_result_handle class.
|
|
struct engine::scheduler::test_result_handle::impl : utils::noncopyable {
|
|
/// Test program data for this test case.
|
|
model::test_program_ptr test_program;
|
|
|
|
/// Name of the test case.
|
|
std::string test_case_name;
|
|
|
|
/// The actual result of the test execution.
|
|
const model::test_result test_result;
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param test_program_ Test program data for this test case.
|
|
/// \param test_case_name_ Name of the test case.
|
|
/// \param test_result_ The actual result of the test execution.
|
|
impl(const model::test_program_ptr test_program_,
|
|
const std::string& test_case_name_,
|
|
const model::test_result& test_result_) :
|
|
test_program(test_program_),
|
|
test_case_name(test_case_name_),
|
|
test_result(test_result_)
|
|
{
|
|
}
|
|
};
|
|
|
|
|
|
/// Constructor.
|
|
///
|
|
/// \param pbimpl Constructed internal implementation for the base object.
|
|
/// \param pimpl Constructed internal implementation.
|
|
scheduler::test_result_handle::test_result_handle(
|
|
std::shared_ptr< bimpl > pbimpl, std::shared_ptr< impl > pimpl) :
|
|
result_handle(pbimpl), _pimpl(pimpl)
|
|
{
|
|
}
|
|
|
|
|
|
/// Destructor.
|
|
scheduler::test_result_handle::~test_result_handle(void)
|
|
{
|
|
}
|
|
|
|
|
|
/// Returns the test program that yielded this result.
|
|
///
|
|
/// \return A test program.
|
|
const model::test_program_ptr
|
|
scheduler::test_result_handle::test_program(void) const
|
|
{
|
|
return _pimpl->test_program;
|
|
}
|
|
|
|
|
|
/// Returns the name of the test case that yielded this result.
|
|
///
|
|
/// \return A test case name
|
|
const std::string&
|
|
scheduler::test_result_handle::test_case_name(void) const
|
|
{
|
|
return _pimpl->test_case_name;
|
|
}
|
|
|
|
|
|
/// Returns the actual result of the test execution.
|
|
///
|
|
/// \return A test result.
|
|
const model::test_result&
|
|
scheduler::test_result_handle::test_result(void) const
|
|
{
|
|
return _pimpl->test_result;
|
|
}
|
|
|
|
|
|
/// Internal implementation for the scheduler_handle.
|
|
struct engine::scheduler::scheduler_handle::impl : utils::noncopyable {
|
|
/// Generic executor instance encapsulated by this one.
|
|
executor::executor_handle generic;
|
|
|
|
/// Mapping of exec handles to the data required at run time.
|
|
exec_data_map all_exec_data;
|
|
|
|
/// Collection of test_exec_data objects.
|
|
typedef std::vector< const test_exec_data* > test_exec_data_vector;
|
|
|
|
/// Constructor.
|
|
impl(void) : generic(executor::setup())
|
|
{
|
|
}
|
|
|
|
/// Destructor.
|
|
///
|
|
/// This runs any pending cleanup routines, which should only happen if the
|
|
/// scheduler is abruptly terminated (aka if a signal is received).
|
|
~impl(void)
|
|
{
|
|
const test_exec_data_vector tests_data = tests_needing_cleanup();
|
|
|
|
for (test_exec_data_vector::const_iterator iter = tests_data.begin();
|
|
iter != tests_data.end(); ++iter) {
|
|
const test_exec_data* test_data = *iter;
|
|
|
|
try {
|
|
sync_cleanup(test_data);
|
|
} catch (const std::runtime_error& e) {
|
|
LW(F("Failed to run cleanup routine for %s:%s on abrupt "
|
|
"termination")
|
|
% test_data->test_program->relative_path()
|
|
% test_data->test_case_name);
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Finds any pending exec_datas that correspond to tests needing cleanup.
|
|
///
|
|
/// \return The collection of test_exec_data objects that have their
|
|
/// needs_cleanup property set to true.
|
|
test_exec_data_vector
|
|
tests_needing_cleanup(void)
|
|
{
|
|
test_exec_data_vector tests_data;
|
|
|
|
for (exec_data_map::const_iterator iter = all_exec_data.begin();
|
|
iter != all_exec_data.end(); ++iter) {
|
|
const exec_data_ptr data = (*iter).second;
|
|
|
|
try {
|
|
test_exec_data* test_data = &dynamic_cast< test_exec_data& >(
|
|
*data.get());
|
|
if (test_data->needs_cleanup) {
|
|
tests_data.push_back(test_data);
|
|
test_data->needs_cleanup = false;
|
|
}
|
|
} catch (const std::bad_cast& e) {
|
|
// Do nothing for cleanup_exec_data objects.
|
|
}
|
|
}
|
|
|
|
return tests_data;
|
|
}
|
|
|
|
/// Cleans up a single test case synchronously.
|
|
///
|
|
/// \param test_data The data of the previously executed test case to be
|
|
/// cleaned up.
|
|
void
|
|
sync_cleanup(const test_exec_data* test_data)
|
|
{
|
|
// The message in this result should never be seen by the user, but use
|
|
// something reasonable just in case it leaks and we need to pinpoint
|
|
// the call site.
|
|
model::test_result result(model::test_result_broken,
|
|
"Test case died abruptly");
|
|
|
|
const executor::exec_handle cleanup_handle = spawn_cleanup(
|
|
test_data->test_program, test_data->test_case_name,
|
|
test_data->user_config, test_data->exit_handle.get(),
|
|
result);
|
|
generic.wait(cleanup_handle);
|
|
}
|
|
|
|
/// Forks and executes a test case cleanup routine asynchronously.
|
|
///
|
|
/// \param test_program The container test program.
|
|
/// \param test_case_name The name of the test case to run.
|
|
/// \param user_config User-provided configuration variables.
|
|
/// \param body_handle The exit handle of the test case's corresponding
|
|
/// body. The cleanup will be executed in the same context.
|
|
/// \param body_result The result of the test case's corresponding body.
|
|
///
|
|
/// \return A handle for the background operation. Used to match the result
|
|
/// of the execution returned by wait_any() with this invocation.
|
|
executor::exec_handle
|
|
spawn_cleanup(const model::test_program_ptr test_program,
|
|
const std::string& test_case_name,
|
|
const config::tree& user_config,
|
|
const executor::exit_handle& body_handle,
|
|
const model::test_result& body_result)
|
|
{
|
|
generic.check_interrupt();
|
|
|
|
const std::shared_ptr< scheduler::interface > interface =
|
|
find_interface(test_program->interface_name());
|
|
|
|
LI(F("Spawning %s:%s (cleanup)") % test_program->absolute_path() %
|
|
test_case_name);
|
|
|
|
const executor::exec_handle handle = generic.spawn_followup(
|
|
run_test_cleanup(interface, test_program, test_case_name,
|
|
user_config),
|
|
body_handle, cleanup_timeout);
|
|
|
|
const exec_data_ptr data(new cleanup_exec_data(
|
|
test_program, test_case_name, body_handle, body_result));
|
|
LD(F("Inserting %s into all_exec_data (cleanup)") % handle.pid());
|
|
INV_MSG(all_exec_data.find(handle.pid()) == all_exec_data.end(),
|
|
F("PID %s already in all_exec_data; not properly cleaned "
|
|
"up or reused too fast") % handle.pid());;
|
|
all_exec_data.insert(exec_data_map::value_type(handle.pid(), data));
|
|
|
|
return handle;
|
|
}
|
|
};
|
|
|
|
|
|
/// Constructor.
|
|
scheduler::scheduler_handle::scheduler_handle(void) : _pimpl(new impl())
|
|
{
|
|
}
|
|
|
|
|
|
/// Destructor.
|
|
scheduler::scheduler_handle::~scheduler_handle(void)
|
|
{
|
|
}
|
|
|
|
|
|
/// Queries the path to the root of the work directory for all tests.
|
|
///
|
|
/// \return A path.
|
|
const fs::path&
|
|
scheduler::scheduler_handle::root_work_directory(void) const
|
|
{
|
|
return _pimpl->generic.root_work_directory();
|
|
}
|
|
|
|
|
|
/// Cleans up the scheduler state.
|
|
///
|
|
/// This function should be called explicitly as it provides the means to
|
|
/// control any exceptions raised during cleanup. Do not rely on the destructor
|
|
/// to clean things up.
|
|
///
|
|
/// \throw engine::error If there are problems cleaning up the scheduler.
|
|
void
|
|
scheduler::scheduler_handle::cleanup(void)
|
|
{
|
|
_pimpl->generic.cleanup();
|
|
}
|
|
|
|
|
|
/// Checks if the given interface name is valid.
|
|
///
|
|
/// \param name The name of the interface to validate.
|
|
///
|
|
/// \throw engine::error If the given interface is not supported.
|
|
void
|
|
scheduler::ensure_valid_interface(const std::string& name)
|
|
{
|
|
if (interfaces.find(name) == interfaces.end())
|
|
throw engine::error(F("Unsupported test interface '%s'") % name);
|
|
}
|
|
|
|
|
|
/// Registers a new interface.
|
|
///
|
|
/// \param name The name of the interface. Must not have yet been registered.
|
|
/// \param spec Interface specification.
|
|
void
|
|
scheduler::register_interface(const std::string& name,
|
|
const std::shared_ptr< interface > spec)
|
|
{
|
|
PRE(interfaces.find(name) == interfaces.end());
|
|
interfaces.insert(interfaces_map::value_type(name, spec));
|
|
}
|
|
|
|
|
|
/// Returns the names of all registered interfaces.
|
|
///
|
|
/// \return A collection of interface names.
|
|
std::set< std::string >
|
|
scheduler::registered_interface_names(void)
|
|
{
|
|
std::set< std::string > names;
|
|
for (interfaces_map::const_iterator iter = interfaces.begin();
|
|
iter != interfaces.end(); ++iter) {
|
|
names.insert((*iter).first);
|
|
}
|
|
return names;
|
|
}
|
|
|
|
|
|
/// Initializes the scheduler.
|
|
///
|
|
/// \pre This function can only be called if there is no other scheduler_handle
|
|
/// object alive.
|
|
///
|
|
/// \return A handle to the operations of the scheduler.
|
|
scheduler::scheduler_handle
|
|
scheduler::setup(void)
|
|
{
|
|
return scheduler_handle();
|
|
}
|
|
|
|
|
|
/// Retrieves the list of test cases from a test program.
|
|
///
|
|
/// This operation is currently synchronous.
|
|
///
|
|
/// This operation should never throw. Any errors during the processing of the
|
|
/// test case list are subsumed into a single test case in the return value that
|
|
/// represents the failed retrieval.
|
|
///
|
|
/// \param test_program The test program from which to obtain the list of test
|
|
/// cases.
|
|
/// \param user_config User-provided configuration variables.
|
|
///
|
|
/// \return The list of test cases.
|
|
model::test_cases_map
|
|
scheduler::scheduler_handle::list_tests(
|
|
const model::test_program* test_program,
|
|
const config::tree& user_config)
|
|
{
|
|
_pimpl->generic.check_interrupt();
|
|
|
|
const std::shared_ptr< scheduler::interface > interface = find_interface(
|
|
test_program->interface_name());
|
|
|
|
try {
|
|
const executor::exec_handle exec_handle = _pimpl->generic.spawn(
|
|
list_test_cases(interface, test_program, user_config),
|
|
list_timeout, none);
|
|
executor::exit_handle exit_handle = _pimpl->generic.wait(exec_handle);
|
|
|
|
const model::test_cases_map test_cases = interface->parse_list(
|
|
exit_handle.status(),
|
|
exit_handle.stdout_file(),
|
|
exit_handle.stderr_file());
|
|
|
|
exit_handle.cleanup();
|
|
|
|
if (test_cases.empty())
|
|
throw std::runtime_error("Empty test cases list");
|
|
|
|
return test_cases;
|
|
} catch (const std::runtime_error& e) {
|
|
// TODO(jmmv): This is a very ugly workaround for the fact that we
|
|
// cannot report failures at the test-program level.
|
|
LW(F("Failed to load test cases list: %s") % e.what());
|
|
model::test_cases_map fake_test_cases;
|
|
fake_test_cases.insert(model::test_cases_map::value_type(
|
|
"__test_cases_list__",
|
|
model::test_case(
|
|
"__test_cases_list__",
|
|
"Represents the correct processing of the test cases list",
|
|
model::test_result(model::test_result_broken, e.what()))));
|
|
return fake_test_cases;
|
|
}
|
|
}
|
|
|
|
|
|
/// Forks and executes a test case asynchronously.
|
|
///
|
|
/// Note that the caller needn't know if the test has a cleanup routine or not.
|
|
/// If there indeed is a cleanup routine, we trigger it at wait_any() time.
|
|
///
|
|
/// \param test_program The container test program.
|
|
/// \param test_case_name The name of the test case to run.
|
|
/// \param user_config User-provided configuration variables.
|
|
///
|
|
/// \return A handle for the background operation. Used to match the result of
|
|
/// the execution returned by wait_any() with this invocation.
|
|
scheduler::exec_handle
|
|
scheduler::scheduler_handle::spawn_test(
|
|
const model::test_program_ptr test_program,
|
|
const std::string& test_case_name,
|
|
const config::tree& user_config)
|
|
{
|
|
_pimpl->generic.check_interrupt();
|
|
|
|
const std::shared_ptr< scheduler::interface > interface = find_interface(
|
|
test_program->interface_name());
|
|
|
|
LI(F("Spawning %s:%s") % test_program->absolute_path() % test_case_name);
|
|
|
|
const model::test_case& test_case = test_program->find(test_case_name);
|
|
|
|
optional< passwd::user > unprivileged_user;
|
|
if (user_config.is_set("unprivileged_user") &&
|
|
test_case.get_metadata().required_user() == "unprivileged") {
|
|
unprivileged_user = user_config.lookup< engine::user_node >(
|
|
"unprivileged_user");
|
|
}
|
|
|
|
const executor::exec_handle handle = _pimpl->generic.spawn(
|
|
run_test_program(interface, test_program, test_case_name,
|
|
user_config),
|
|
test_case.get_metadata().timeout(),
|
|
unprivileged_user);
|
|
|
|
const exec_data_ptr data(new test_exec_data(
|
|
test_program, test_case_name, interface, user_config));
|
|
LD(F("Inserting %s into all_exec_data") % handle.pid());
|
|
INV_MSG(
|
|
_pimpl->all_exec_data.find(handle.pid()) == _pimpl->all_exec_data.end(),
|
|
F("PID %s already in all_exec_data; not cleaned up or reused too fast")
|
|
% handle.pid());;
|
|
_pimpl->all_exec_data.insert(exec_data_map::value_type(handle.pid(), data));
|
|
|
|
return handle.pid();
|
|
}
|
|
|
|
|
|
/// Waits for completion of any forked test case.
|
|
///
|
|
/// Note that if the terminated test case has a cleanup routine, this function
|
|
/// is the one in charge of spawning the cleanup routine asynchronously.
|
|
///
|
|
/// \return The result of the execution of a subprocess. This is a dynamically
|
|
/// allocated object because the scheduler can spawn subprocesses of various
|
|
/// types and, at wait time, we don't know upfront what we are going to get.
|
|
scheduler::result_handle_ptr
|
|
scheduler::scheduler_handle::wait_any(void)
|
|
{
|
|
_pimpl->generic.check_interrupt();
|
|
|
|
executor::exit_handle handle = _pimpl->generic.wait_any();
|
|
|
|
const exec_data_map::iterator iter = _pimpl->all_exec_data.find(
|
|
handle.original_pid());
|
|
exec_data_ptr data = (*iter).second;
|
|
|
|
utils::dump_stacktrace_if_available(data->test_program->absolute_path(),
|
|
_pimpl->generic, handle);
|
|
|
|
optional< model::test_result > result;
|
|
try {
|
|
test_exec_data* test_data = &dynamic_cast< test_exec_data& >(
|
|
*data.get());
|
|
LD(F("Got %s from all_exec_data") % handle.original_pid());
|
|
|
|
test_data->exit_handle = handle;
|
|
|
|
const model::test_case& test_case = test_data->test_program->find(
|
|
test_data->test_case_name);
|
|
|
|
result = test_case.fake_result();
|
|
|
|
if (!result && handle.status() && handle.status().get().exited() &&
|
|
handle.status().get().exitstatus() == exit_skipped) {
|
|
// If the test's process terminated with our magic "exit_skipped"
|
|
// status, there are two cases to handle. The first is the case
|
|
// where the "skipped cookie" exists, in which case we never got to
|
|
// actually invoke the test program; if that's the case, handle it
|
|
// here. The second case is where the test case actually decided to
|
|
// exit with the "exit_skipped" status; in that case, just fall back
|
|
// to the regular status handling.
|
|
const fs::path skipped_cookie_path = handle.control_directory() /
|
|
skipped_cookie;
|
|
std::ifstream input(skipped_cookie_path.c_str());
|
|
if (input) {
|
|
result = model::test_result(model::test_result_skipped,
|
|
utils::read_stream(input));
|
|
input.close();
|
|
|
|
// If we determined that the test needs to be skipped, we do not
|
|
// want to run the cleanup routine because doing so could result
|
|
// in errors. However, we still want to run the cleanup routine
|
|
// if the test's body reports a skip (because actions could have
|
|
// already been taken).
|
|
test_data->needs_cleanup = false;
|
|
}
|
|
}
|
|
if (!result) {
|
|
result = test_data->interface->compute_result(
|
|
handle.status(),
|
|
handle.control_directory(),
|
|
handle.stdout_file(),
|
|
handle.stderr_file());
|
|
}
|
|
INV(result);
|
|
|
|
if (!result.get().good()) {
|
|
append_files_listing(handle.work_directory(),
|
|
handle.stderr_file());
|
|
}
|
|
|
|
if (test_data->needs_cleanup) {
|
|
INV(test_case.get_metadata().has_cleanup());
|
|
// The test body has completed and we have processed it. If there
|
|
// is a cleanup routine, trigger it now and wait for any other test
|
|
// completion. The caller never knows about cleanup routines.
|
|
_pimpl->spawn_cleanup(test_data->test_program,
|
|
test_data->test_case_name,
|
|
test_data->user_config, handle, result.get());
|
|
test_data->needs_cleanup = false;
|
|
|
|
// TODO(jmmv): Chaining this call is ugly. We'd be better off by
|
|
// looping over terminated processes until we got a result suitable
|
|
// for user consumption. For the time being this is good enough and
|
|
// not a problem because the call chain won't get big: the majority
|
|
// of test cases do not have cleanup routines.
|
|
return wait_any();
|
|
}
|
|
} catch (const std::bad_cast& e) {
|
|
const cleanup_exec_data* cleanup_data =
|
|
&dynamic_cast< const cleanup_exec_data& >(*data.get());
|
|
LD(F("Got %s from all_exec_data (cleanup)") % handle.original_pid());
|
|
|
|
// Handle the completion of cleanup subprocesses internally: the caller
|
|
// is not aware that these exist so, when we return, we must return the
|
|
// data for the original test that triggered this routine. For example,
|
|
// because the caller wants to see the exact same exec_handle that was
|
|
// returned by spawn_test.
|
|
|
|
const model::test_result& body_result = cleanup_data->body_result;
|
|
if (body_result.good()) {
|
|
if (!handle.status()) {
|
|
result = model::test_result(model::test_result_broken,
|
|
"Test case cleanup timed out");
|
|
} else {
|
|
if (!handle.status().get().exited() ||
|
|
handle.status().get().exitstatus() != EXIT_SUCCESS) {
|
|
result = model::test_result(
|
|
model::test_result_broken,
|
|
"Test case cleanup did not terminate successfully");
|
|
} else {
|
|
result = body_result;
|
|
}
|
|
}
|
|
} else {
|
|
result = body_result;
|
|
}
|
|
|
|
// Untrack the cleanup process. This must be done explicitly because we
|
|
// do not create a result_handle object for the cleanup, and that is the
|
|
// one in charge of doing so in the regular (non-cleanup) case.
|
|
LD(F("Removing %s from all_exec_data (cleanup) in favor of %s")
|
|
% handle.original_pid()
|
|
% cleanup_data->body_exit_handle.original_pid());
|
|
_pimpl->all_exec_data.erase(handle.original_pid());
|
|
|
|
handle = cleanup_data->body_exit_handle;
|
|
}
|
|
INV(result);
|
|
|
|
std::shared_ptr< result_handle::bimpl > result_handle_bimpl(
|
|
new result_handle::bimpl(handle, _pimpl->all_exec_data));
|
|
std::shared_ptr< test_result_handle::impl > test_result_handle_impl(
|
|
new test_result_handle::impl(
|
|
data->test_program, data->test_case_name, result.get()));
|
|
return result_handle_ptr(new test_result_handle(result_handle_bimpl,
|
|
test_result_handle_impl));
|
|
}
|
|
|
|
|
|
/// Forks and executes a test case synchronously for debugging.
|
|
///
|
|
/// \pre No other processes should be in execution by the scheduler.
|
|
///
|
|
/// \param test_program The container test program.
|
|
/// \param test_case_name The name of the test case to run.
|
|
/// \param user_config User-provided configuration variables.
|
|
/// \param stdout_target File to which to write the stdout of the test case.
|
|
/// \param stderr_target File to which to write the stderr of the test case.
|
|
///
|
|
/// \return The result of the execution of the test.
|
|
scheduler::result_handle_ptr
|
|
scheduler::scheduler_handle::debug_test(
|
|
const model::test_program_ptr test_program,
|
|
const std::string& test_case_name,
|
|
const config::tree& user_config,
|
|
const fs::path& stdout_target,
|
|
const fs::path& stderr_target)
|
|
{
|
|
const exec_handle exec_handle = spawn_test(
|
|
test_program, test_case_name, user_config);
|
|
result_handle_ptr result_handle = wait_any();
|
|
|
|
// TODO(jmmv): We need to do this while the subprocess is alive. This is
|
|
// important for debugging purposes, as we should see the contents of stdout
|
|
// or stderr as they come in.
|
|
//
|
|
// Unfortunately, we cannot do so. We cannot just read and block from a
|
|
// file, waiting for further output to appear... as this only works on pipes
|
|
// or sockets. We need a better interface for this whole thing.
|
|
{
|
|
std::auto_ptr< std::ostream > output = utils::open_ostream(
|
|
stdout_target);
|
|
*output << utils::read_file(result_handle->stdout_file());
|
|
}
|
|
{
|
|
std::auto_ptr< std::ostream > output = utils::open_ostream(
|
|
stderr_target);
|
|
*output << utils::read_file(result_handle->stderr_file());
|
|
}
|
|
|
|
INV(result_handle->original_pid() == exec_handle);
|
|
return result_handle;
|
|
}
|
|
|
|
|
|
/// Checks if an interrupt has fired.
|
|
///
|
|
/// Calls to this function should be sprinkled in strategic places through the
|
|
/// code protected by an interrupts_handler object.
|
|
///
|
|
/// This is just a wrapper over signals::check_interrupt() to avoid leaking this
|
|
/// dependency to the caller.
|
|
///
|
|
/// \throw signals::interrupted_error If there has been an interrupt.
|
|
void
|
|
scheduler::scheduler_handle::check_interrupt(void) const
|
|
{
|
|
_pimpl->generic.check_interrupt();
|
|
}
|
|
|
|
|
|
/// Queries the current execution context.
|
|
///
|
|
/// \return The queried context.
|
|
model::context
|
|
scheduler::current_context(void)
|
|
{
|
|
return model::context(fs::current_path(), utils::getallenv());
|
|
}
|
|
|
|
|
|
/// Generates the set of configuration variables for a test program.
|
|
///
|
|
/// \param user_config The configuration variables provided by the user.
|
|
/// \param test_suite The name of the test suite.
|
|
///
|
|
/// \return The mapping of configuration variables for the test program.
|
|
config::properties_map
|
|
scheduler::generate_config(const config::tree& user_config,
|
|
const std::string& test_suite)
|
|
{
|
|
config::properties_map props;
|
|
|
|
try {
|
|
props = user_config.all_properties(F("test_suites.%s") % test_suite,
|
|
true);
|
|
} catch (const config::unknown_key_error& unused_error) {
|
|
// Ignore: not all test suites have entries in the configuration.
|
|
}
|
|
|
|
// TODO(jmmv): This is a hack that exists for the ATF interface only, so it
|
|
// should be moved there.
|
|
if (user_config.is_set("unprivileged_user")) {
|
|
const passwd::user& user =
|
|
user_config.lookup< engine::user_node >("unprivileged_user");
|
|
props["unprivileged-user"] = user.name;
|
|
}
|
|
|
|
return props;
|
|
}
|