rust/src/test/compiletest/compiletest.rs

747 lines
24 KiB
Rust
Raw Normal View History

The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
import std::option;
import std::getopts;
import std::test;
import std::fs;
import std::str;
import std::vec;
import std::ivec;
import std::io;
import std::generic_os::setenv;
import std::generic_os::getenv;
import std::os;
import std::run;
import std::task;
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
tag mode {
mode_compile_fail;
mode_run_fail;
mode_run_pass;
}
type config = rec(// The library paths required for running the compiler
str compile_lib_path,
// The library paths required for running compiled programs
str run_lib_path,
// The rustc executable
str rustc_path,
// The directory containing the tests to run
str src_base,
// The directory where programs should be built
str build_base,
// The name of the stage being built (stage1, etc)
str stage_id,
// The test mode, compile-fail, run-fail, run-pass
mode mode,
// Run ignored tests
bool run_ignored,
// Only run tests that match this filter
option::t[str] filter,
// A command line to prefix program execution with,
// for running under valgrind
option::t[str] runtool,
// Flags to pass to the compiler
option::t[str] rustcflags,
// Explain what's going on
bool verbose);
fn main(vec[str] args) {
auto ivec_args = {
auto ivec_args = ~[];
for (str arg in args) {
ivec_args += ~[arg];
}
ivec_args
};
auto config = parse_config(ivec_args);
log_config(config);
run_tests(config);
}
fn parse_config(&str[] args) -> config {
auto opts = ~[getopts::reqopt("compile-lib-path"),
getopts::reqopt("run-lib-path"),
getopts::reqopt("rustc-path"),
getopts::reqopt("src-base"),
getopts::reqopt("build-base"),
getopts::reqopt("stage-id"),
getopts::reqopt("mode"),
getopts::optflag("ignored"),
getopts::optopt("runtool"),
getopts::optopt("rustcflags"),
getopts::optflag("verbose")];
check ivec::is_not_empty(args);
auto args_ = ivec::tail(args);
auto match = alt (getopts::getopts_ivec(args_, opts)) {
getopts::success(?m) { m }
getopts::failure(?f) {
fail getopts::fail_str(f)
}
};
ret rec(compile_lib_path = getopts::opt_str(match, "compile-lib-path"),
run_lib_path = getopts::opt_str(match, "run-lib-path"),
rustc_path = getopts::opt_str(match, "rustc-path"),
src_base = getopts::opt_str(match, "src-base"),
build_base = getopts::opt_str(match, "build-base"),
stage_id = getopts::opt_str(match, "stage-id"),
mode = str_mode(getopts::opt_str(match, "mode")),
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
run_ignored = getopts::opt_present(match, "ignored"),
filter = if vec::len(match.free) > 0u {
option::some(match.free.(0))
} else {
option::none
},
runtool = getopts::opt_maybe_str(match, "runtool"),
rustcflags = getopts::opt_maybe_str(match, "rustcflags"),
verbose = getopts::opt_present(match, "verbose"));
}
fn log_config(&config config) {
auto c = config;
logv(c, #fmt("configuration:"));
logv(c, #fmt("compile_lib_path: %s", config.compile_lib_path));
logv(c, #fmt("run_lib_path: %s", config.run_lib_path));
logv(c, #fmt("rustc_path: %s", config.rustc_path));
logv(c, #fmt("src_base: %s", config.src_base));;
logv(c, #fmt("build_base: %s", config.build_base));
logv(c, #fmt("stage_id: %s", config.stage_id));
logv(c, #fmt("mode: %s", mode_str(config.mode)));
logv(c, #fmt("run_ignored: %b", config.run_ignored));
logv(c, #fmt("filter: %s", opt_str(config.filter)));
logv(c, #fmt("runtool: %s", opt_str(config.runtool)));
logv(c, #fmt("rustcflags: %s", opt_str(config.rustcflags)));
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
logv(c, #fmt("verbose: %b", config.verbose));
logv(c, #fmt("\n"));
}
fn opt_str(option::t[str] maybestr) -> str {
alt maybestr {
option::some(?s) { s }
option::none { "(none)" }
}
}
fn str_opt(str maybestr) -> option::t[str] {
if maybestr != "(none)" {
option::some(maybestr)
} else {
option::none
}
}
fn str_mode(str s) -> mode {
alt s {
"compile-fail" { mode_compile_fail }
"run-fail" { mode_run_fail }
"run-pass" { mode_run_pass }
_ { fail "invalid mode" }
}
}
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
fn mode_str(mode mode) -> str {
alt (mode) {
mode_compile_fail { "compile-fail" }
mode_run_fail { "run-fail" }
mode_run_pass { "run-pass" }
}
}
type cx = rec(config config,
procsrv::handle procsrv);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
fn run_tests(&config config) {
auto opts = test_opts(config);
auto cx = rec(config = config,
procsrv = procsrv::mk());
auto tests = make_tests(cx);
test::run_tests_console_(opts, tests.tests, tests.to_task);
procsrv::close(cx.procsrv);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn test_opts(&config config) -> test::test_opts {
rec(filter = config.filter,
run_ignored = config.run_ignored)
}
type tests_and_conv_fn = rec(test::test_desc[] tests,
fn(&fn()) -> task to_task);
fn make_tests(&cx cx) -> tests_and_conv_fn {
log #fmt("making tests from %s", cx.config.src_base);
auto configport = port[str]();
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
auto tests = ~[];
for (str file in fs::list_dir(cx.config.src_base)) {
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
log #fmt("inspecting file %s", file);
if (is_test(file)) {
tests += ~[make_test(cx, file, configport)];
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
}
ret rec(tests = tests,
to_task = bind closure_to_task(cx, configport, _));
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn is_test(&str testfile) -> bool {
auto name = fs::basename(testfile);
(str::ends_with(name, ".rs") || str::ends_with(name, ".rc"))
&& !(str::starts_with(name, ".")
|| str::starts_with(name, "#")
|| str::starts_with(name, "~"))
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn make_test(&cx cx, &str testfile,
&port[str] configport) -> test::test_desc {
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
rec(name = testfile,
fn = make_test_closure(testfile, chan(configport)),
ignore = is_test_ignored(cx.config, testfile))
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn is_test_ignored(&config config, &str testfile) -> bool {
auto found = false;
for each (str ln in iter_header(testfile)) {
// FIXME: Can't return or break from iterator
found = found || parse_name_directive(ln, "xfail-" + config.stage_id);
}
ret found;
}
iter iter_header(&str testfile) -> str {
auto rdr = io::file_reader(testfile);
while !rdr.eof() {
auto ln = rdr.read_line();
// Assume that any directives will be found before the
// first module or function. This doesn't seem to be an optimization
// with a warm page cache. Maybe with a cold one.
if str::starts_with(ln, "fn") || str::starts_with(ln, "mod") {
break;
} else {
put ln;
}
}
}
/*
So this is kind of crappy:
A test is just defined as a function, as you might expect, but tests have to
run their own tasks. Unfortunately, if your test needs dynamic data then it
needs to be a closure, and transferring closures across tasks without
committing a host of memory management transgressions is just impossible.
To get around this, the standard test runner allows you the opportunity do
your own conversion from a test function to a task. It gives you your function
and you give it back a task.
So that's what we're going to do. Here's where it gets stupid. To get the
the data out of the test function we are going to run the test function,
which will do nothing but send the data for that test to a port we've set
up. Then we'll spawn that data into another task and return the task.
Really convoluted. Need to think up of a better definition for tests.
*/
fn make_test_closure(&str testfile,
chan[str] configchan) -> test::test_fn {
bind send_config(testfile, configchan)
}
fn send_config(str testfile, chan[str] configchan) {
task::send(configchan, testfile);
}
/*
FIXME: Good god forgive me.
So actually shuttling structural data across tasks isn't possible at this
time, but we can send strings! Sadly, I need the whole config record, in the
test task so, instead of fixing the mechanism in the compiler I'm going to
break up the config record and pass everything individually to the spawned
function. */
fn closure_to_task(cx cx, port[str] configport, &fn() testfn) -> task{
testfn();
auto testfile = task::recv(configport);
ret spawn run_test_task(cx.config.compile_lib_path,
cx.config.run_lib_path,
cx.config.rustc_path,
cx.config.src_base,
cx.config.build_base,
cx.config.stage_id,
mode_str(cx.config.mode),
cx.config.run_ignored,
opt_str(cx.config.filter),
opt_str(cx.config.runtool),
opt_str(cx.config.rustcflags),
cx.config.verbose,
procsrv::clone(cx.procsrv).chan,
testfile);
}
fn run_test_task(str compile_lib_path,
str run_lib_path,
str rustc_path,
str src_base,
str build_base,
str stage_id,
str mode,
bool run_ignored,
str opt_filter,
str opt_runtool,
str opt_rustcflags,
bool verbose,
procsrv::reqchan procsrv_chan,
str testfile) {
auto config = rec(compile_lib_path = compile_lib_path,
run_lib_path = run_lib_path,
rustc_path = rustc_path,
src_base = src_base,
build_base = build_base,
stage_id = stage_id,
mode = str_mode(mode),
run_ignored = run_ignored,
filter = str_opt(opt_filter),
runtool = str_opt(opt_runtool),
rustcflags = str_opt(opt_rustcflags),
verbose = verbose);
auto procsrv = procsrv::from_chan(procsrv_chan);
auto cx = rec(config = config,
procsrv = procsrv);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
log #fmt("running %s", testfile);
task::unsupervise();
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
auto props = load_props(testfile);
alt (cx.config.mode) {
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
mode_compile_fail {
run_cfail_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
mode_run_fail {
run_rfail_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
mode_run_pass {
run_rpass_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
}
}
type test_props = rec(str[] error_patterns,
option::t[str] compile_flags);
// Load any test directives embedded in the file
fn load_props(&str testfile) -> test_props {
auto error_patterns = ~[];
auto compile_flags = option::none;
for each (str ln in iter_header(testfile)) {
alt parse_error_pattern(ln) {
option::some(?ep) { error_patterns += ~[ep]; }
option::none { }
}
if option::is_none(compile_flags) {
compile_flags = parse_compile_flags(ln);
}
}
ret rec(error_patterns = error_patterns,
compile_flags = compile_flags);
}
fn parse_error_pattern(&str line) -> option::t[str] {
parse_name_value_directive(line, "error-pattern")
}
fn parse_compile_flags(&str line) -> option::t[str] {
parse_name_value_directive(line, "compile-flags")
}
fn parse_name_directive(&str line, &str directive) -> bool {
str::find(line, directive) >= 0
}
fn parse_name_value_directive(&str line, &str directive) -> option::t[str] {
auto keycolon = directive + ":";
if str::find(line, keycolon) >= 0 {
auto colon = str::find(line, keycolon) as uint;
auto value = str::slice(line,
colon + str::byte_len(keycolon),
str::byte_len(line));
log #fmt("%s: %s", directive, value);
option::some(value)
} else {
option::none
}
}
fn run_cfail_test(&cx cx, &test_props props, &str testfile) {
auto procres = compile_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
if (procres.status == 0) {
fatal_procres("compile-fail test compiled successfully!", procres);
}
check_error_patterns(props, testfile, procres);
}
fn run_rfail_test(&cx cx, &test_props props, &str testfile) {
auto procres = compile_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
if (procres.status != 0) {
fatal_procres("compilation failed!", procres);
}
procres = exec_compiled_test(cx, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
if (procres.status == 0) {
fatal_procres("run-fail test didn't produce an error!",
procres);
}
check_error_patterns(props, testfile, procres);
}
fn run_rpass_test(&cx cx, &test_props props, &str testfile) {
auto procres = compile_test(cx, props, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
if (procres.status != 0) {
fatal_procres("compilation failed!", procres);
}
procres = exec_compiled_test(cx, testfile);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
if (procres.status != 0) {
fatal_procres("test run failed!", procres);
}
}
fn check_error_patterns(&test_props props, &str testfile,
&procres procres) {
if ivec::is_empty(props.error_patterns) {
fatal("no error pattern specified in " + testfile);
}
auto next_err_idx = 0u;
auto next_err_pat = props.error_patterns.(next_err_idx);
for (str line in str::split(procres.out, '\n' as u8)) {
if (str::find(line, next_err_pat) > 0) {
log #fmt("found error pattern %s", next_err_pat);
next_err_idx += 1u;
if next_err_idx == ivec::len(props.error_patterns) {
log "found all error patterns";
ret;
}
next_err_pat = props.error_patterns.(next_err_idx);
}
}
auto missing_patterns = ivec::slice(props.error_patterns,
next_err_idx,
ivec::len(props.error_patterns));
if (ivec::len(missing_patterns) == 1u) {
fatal_procres(#fmt("error pattern '%s' not found!",
missing_patterns.(0)),
procres);
} else {
for (str pattern in missing_patterns) {
error(#fmt("error pattern '%s' not found!", pattern));
}
fatal_procres("multiple error patterns not found", procres);
}
}
type procargs = rec(str prog, vec[str] args);
type procres = rec(int status, str out, str cmdline);
fn compile_test(&cx cx, &test_props props,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
&str testfile) -> procres {
compose_and_run(cx,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
testfile,
bind make_compile_args(_, props, _),
cx.config.compile_lib_path)
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn exec_compiled_test(&cx cx, &str testfile) -> procres {
compose_and_run(cx,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
testfile,
make_run_args,
cx.config.run_lib_path)
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn compose_and_run(&cx cx, &str testfile,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
fn(&config, &str) -> procargs make_args,
&str lib_path) -> procres {
auto procargs = make_args(cx.config, testfile);
ret program_output(cx, testfile, lib_path,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
procargs.prog, procargs.args);
}
fn make_compile_args(&config config, &test_props props,
&str testfile) -> procargs {
auto prog = config.rustc_path;
auto args = [testfile,
"-o", make_exe_name(config, testfile)];
args += split_maybe_args(config.rustcflags);
args += split_maybe_args(props.compile_flags);
ret rec(prog = prog,
args = args);
}
fn make_run_args(&config config, &str testfile) -> procargs {
// If we've got another tool to run under (valgrind),
// then split apart its command
auto args = split_maybe_args(config.runtool)
+ [make_exe_name(config, testfile)];
ret rec(prog = args.(0),
args = vec::slice(args, 1u, vec::len(args)));
}
fn split_maybe_args(&option::t[str] argstr) -> vec[str] {
alt (argstr) {
option::some(?s) { str::split(s, ' ' as u8) }
option::none { [] }
}
}
fn program_output(&cx cx, &str testfile,
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
&str lib_path, &str prog, &vec[str] args) -> procres {
auto cmdline = {
auto cmdline = make_cmdline(lib_path, prog, args);
logv(cx.config, #fmt("running %s", cmdline));
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
cmdline
};
auto res = procsrv::run(cx.procsrv, lib_path, prog, args);
dump_output(cx.config, testfile, res.out);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
ret rec(status = res.status,
out = res.out,
cmdline = cmdline);
}
fn make_cmdline(&str libpath, &str prog, &vec[str] args) -> str {
#fmt("%s %s %s",
lib_path_cmd_prefix(libpath),
prog,
str::connect(args, " "))
}
// Build the LD_LIBRARY_PATH variable as it would be seen on the command line
// for diagnostic purposes
fn lib_path_cmd_prefix(&str path) -> str {
#fmt("%s=\"%s\"", lib_path_env_var(), make_new_path(path))
}
fn make_new_path(&str path) -> str {
// Windows just uses PATH as the library search path, so we have to
// maintain the current value while adding our own
alt getenv(lib_path_env_var()) {
option::some(?curr) { #fmt("%s:%s", path, curr) }
option::none { path }
}
}
#[cfg(target_os = "linux")]
fn lib_path_env_var() -> str { "LD_LIBRARY_PATH" }
#[cfg(target_os = "macos")]
fn lib_path_env_var() -> str { "DYLD_LIBRARY_PATH" }
#[cfg(target_os = "win32")]
fn lib_path_env_var() -> str { "PATH" }
fn make_exe_name(&config config, &str testfile) -> str {
output_base_name(config, testfile) + os::exec_suffix()
}
fn output_base_name(&config config, &str testfile) -> str {
auto base = config.build_base;
auto filename = {
auto parts = str::split(fs::basename(testfile), '.' as u8);
parts = vec::slice(parts, 0u, vec::len(parts) - 1u);
str::connect(parts, ".")
};
#fmt("%s%s.%s", base, filename, config.stage_id)
}
#[cfg(target_os = "win32")]
#[cfg(target_os = "linux")]
fn dump_output(&config config, &str testfile, &str out) {
auto outfile = make_out_name(config, testfile);
auto writer = io::file_writer(outfile, [io::create, io::truncate]);
writer.write_str(out);
maybe_dump_to_stdout(config, out);
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
// FIXME (726): Can't use file_writer on mac
#[cfg(target_os = "macos")]
fn dump_output(&config config, &str testfile, &str out) {
maybe_dump_to_stdout(config, out);
}
fn maybe_dump_to_stdout(&config config, &str out) {
if (config.verbose) {
io::stdout().write_line("------------------------------------------");
io::stdout().write_line(out);
io::stdout().write_line("------------------------------------------");
}
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
}
fn make_out_name(&config config, &str testfile) -> str {
output_base_name(config, testfile) + ".out"
}
fn error(&str err) {
io::stdout().write_line(#fmt("\nerror: %s", err));
}
fn fatal(&str err) -> ! {
error(err);
fail;
}
fn fatal_procres(&str err, procres procres) -> ! {
auto msg = #fmt("\n\
error: %s\n\
command: %s\n\
output:\n\
------------------------------------------\n\
%s\n\
------------------------------------------\n\
\n",
err, procres.cmdline, procres.out);
io::stdout().write_str(msg);
fail;
}
fn logv(&config config, &str s) {
log s;
if (config.verbose) {
io::stdout().write_line(s);
}
}
// So when running tests in parallel there's a potential race on environment
// variables if we let each task spawn its own children - between the time the
// environment is set and the process is spawned another task could spawn its
// child process. Because of that we have to use a complicated scheme with a
// dedicated server for spawning processes.
mod procsrv {
export handle;
export mk;
export from_chan;
export clone;
export run;
export close;
export reqchan;
type reqchan = chan[request];
type handle = rec(option::t[task] task,
reqchan chan);
tag request {
exec(str, str, vec[str], chan[response]);
stop;
}
type response = rec(int pid, int outfd);
fn mk() -> handle {
auto res = task::worker(worker);
ret rec(task = option::some(res.task),
chan = res.chan);
}
fn from_chan(&reqchan ch) -> handle {
rec(task = option::none,
chan = ch)
}
fn clone(&handle handle) -> handle {
// Sharing tasks across tasks appears to be (yet another) recipe for
// disaster, so our handle clones will not get the task pointer.
rec(task = option::none,
chan = task::clone_chan(handle.chan))
}
fn close(&handle handle) {
task::send(handle.chan, stop);
task::join(option::get(handle.task));
}
fn run(&handle handle, &str lib_path,
&str prog, &vec[str] args) -> rec(int status, str out) {
auto p = port[response]();
auto ch = chan(p);
task::send(handle.chan,
exec(lib_path, prog, args, ch));
auto resp = task::recv(p);
// Copied from run::program_output
auto outfile = os::fd_FILE(resp.outfd);
auto reader = io::new_reader(io::FILE_buf_reader(outfile, false));
auto buf = "";
while (!reader.eof()) {
auto bytes = reader.read_bytes(4096u);
buf += str::unsafe_from_bytes(bytes);
}
os::libc::fclose(outfile);
ret rec(status = os::waitpid(resp.pid), out = buf);
}
fn worker(port[request] p) {
while (true) {
alt task::recv(p) {
exec(?lib_path, ?prog, ?args, ?respchan) {
// This is copied from run::start_program
auto pipe_in = os::pipe();
auto pipe_out = os::pipe();
auto spawnproc = bind run::spawn_process(
prog, args, pipe_in.in, pipe_out.out, 0);
auto pid = with_lib_path(lib_path, spawnproc);
if (pid == -1) { fail; }
os::libc::close(pipe_in.in);
os::libc::close(pipe_in.out);
os::libc::close(pipe_out.out);
task::send(respchan, rec(pid = pid,
outfd = pipe_out.in));
}
stop {
ret;
}
}
}
}
fn with_lib_path[T](&str path, fn() -> T f) -> T {
auto maybe_oldpath = getenv(lib_path_env_var());
append_lib_path(path);
auto res = f();
if option::is_some(maybe_oldpath) {
export_lib_path(option::get(maybe_oldpath));
} else {
// FIXME: This should really be unset but we don't have that yet
export_lib_path("");
}
ret res;
}
fn append_lib_path(&str path) {
export_lib_path(make_new_path(path));
}
fn export_lib_path(&str path) {
setenv(lib_path_env_var(), path);
}
}
The Big Test Suite Overhaul This replaces the make-based test runner with a set of Rust-based test runners. I believe that all existing functionality has been preserved. The primary objective is to dogfood the Rust test framework. A few main things happen here: 1) The run-pass/lib-* tests are all moved into src/test/stdtest. This is a standalone test crate intended for all standard library tests. It compiles to build/test/stdtest.stageN. 2) rustc now compiles into yet another build artifact, this one a test runner that runs any tests contained directly in the rustc crate. This allows much more fine-grained unit testing of the compiler. It compiles to build/test/rustctest.stageN. 3) There is a new custom test runner crate at src/test/compiletest that reproduces all the functionality for running the compile-fail, run-fail, run-pass and bench tests while integrating with Rust's test framework. It compiles to build/test/compiletest.stageN. 4) The build rules have been completely changed to use the new test runners, while also being less redundant, following the example of the recent stageN.mk rewrite. It adds two new features to the cfail/rfail/rpass/bench tests: 1) Tests can specify multiple 'error-pattern' directives which must be satisfied in order. 2) Tests can specify a 'compile-flags' directive which will make the test runner provide additional command line arguments to rustc. There are some downsides, the primary being that Rust has to be functioning pretty well just to run _any_ tests, which I imagine will be the source of some frustration when the entire test suite breaks. Will also cause some headaches during porting. Not having individual make rules, each rpass, etc test no longer remembers between runs whether it completed successfully. As a result, it's not possible to incrementally fix multiple tests by just running 'make check', fixing a test, and repeating without re-running all the tests contained in the test runner. Instead you can filter just the tests you want to run by using the TESTNAME environment variable. This also dispenses with the ability to run stage0 tests, but they tended to be broken more often than not anyway.
2011-07-12 19:01:09 -07:00
// Local Variables:
// fill-column: 78;
// indent-tabs-mode: nil
// c-basic-offset: 4
// buffer-file-coding-system: utf-8-unix
// compile-command: "make -k -C $RBUILD 2>&1 | sed -e 's/\\/x\\//x:\\//g'";
// End: