Skip to content

Latest commit

 

History

History
284 lines (181 loc) · 14.3 KB

File metadata and controls

284 lines (181 loc) · 14.3 KB

Building and Running CoreCLR Tests

This guide will walk you through building and running the CoreCLR tests. These are located within the src/tests subtree of the runtime repo.

Requirements

In order to build CoreCLR tests, you will need to have built the runtime and the libraries (that is, clr and libs subsets). You can find more detailed instructions per platform in their dedicated docs:

For CoreCLR testing purposes, it is more than enough to simply build the libs subset, as far as it concerns the libraries. If you want to know more in-depth about them, they have their own libraries dedicated docs section.

Overview

As mentioned in the introduction, all test-building work is done from the src/tests folder, so we will consider that our starting point for the rest of this guide.

Building the tests can be as simple as calling the build script without any arguments. This will by default look for a Debug built runtime and Release built libraries. However, by passing the appropriate flags, you can define which configurations you want to use. For example, let's suppose you have a Checked runtime with Debug libraries:

./src/tests/build.sh checked /p:LibrariesConfiguration=Debug

Note that for the libraries configuration, we are passing the argument directly to MSBuild instead of the build script, hence the /p:LibrariesConfiguration flag. Also, make sure you use the correct syntax depending on our platform. The cmd script takes the arguments by placing, while the sh script requires them to be with a hyphen.

NOTE: Building the whole test suite is a very lengthy process, so it is highly recommended you build individual tests, and/or test subtrees as you need them, to make your workflow more efficient. This is explained in detail later on in this doc.

Building the Core_Root

The Core_Root folder is some sort of "dev-easy-to-use full build" of the product. It contains the built runtime binaries, as well as the library packages required to run tests. It is required that you build the libraries subset (--subset libs) before this command can be run.

Note that, as mentioned in the section above, running the tests build script by default searches the libraries in Release mode, regardless of the runtime configuration you specify. If you built your libraries in another configuration, then you have to pass down the appropriate flag /p:LibrariesConfiguration=<your_config>.

The simplest command to generate the Core_Root from the repository's root path is the following:

.\src\tests\build.cmd generatelayoutonly

This example assumes you built CoreCLR on Debug mode and the Libraries on Release mode, hence no additional flags are needed. After the build is complete, you will be able to find the output in the artifacts/tests/coreclr/<OS>.<arch>.<configuration>/Tests/Core_Root folder.

Building the Tests

The following subsections will explain how to segment the test suite according to your needs. There are three main scopes of building tests:

  • Individual Test
  • Full Directory
  • Entire Subtree

When no set is specified, almost the whole test suite (src/tests) is built.

One of the most important attributes tests have is their priority. By default, only those tests with Priority 0 are built, regardless of whether it's an individual one, a directory, or a subtree, unless otherwise specified with the -priority flag. More info on this in its dedicated section. Regardless of which subset you build, all the outputs will be placed in artifacts/tests/coreclr/<OS>.<arch>.<configuration>.

NOTE: Some tests have native components to them. It is highly recommended you build all of those prior to attempting to build any managed test in the following sections, as it's not a very costly or lengthy process:

.\src\tests\build.cmd skipmanaged

Building an Individual Test

To build an individual test, you have to pass the -test flag along with the path to the test's csproj file to the build script. You can select more than one by repeating the -test flag. For example, let's try building a couple JIT tests:

On Windows:

.\src\tests\build.cmd test JIT\Intrinsics\MathRoundDouble_ro.csproj test JIT\Intrinsics\MathFloorDouble_ro.csproj

On macOS and Linux:

./src/tests/build.sh -test:JIT/Intrinsics/MathRoundDouble_ro.csproj -test:JIT/Intrinsics/MathFloorDouble_ro.csproj

Alternatively, you can call build directly using the dotnet.cmd/dotnet.sh script at the root of the repo and pass all arguments directly yourself:

./dotnet.sh build -c <Your Configuration> src/tests/path/to/test/csproj

Building a Test Directory

To build all the tests contained in an individual directory, you have to pass the -dir flag along with the directory's path to the build script. Just like with individual tests, you can select more than one by repeating the -dir flag. For example, let's try a couple of folders in the JIT subtree:

On Windows:

.\src\tests\build.cmd dir JIT\Methodical\Arrays\lcs dir JIT\Methodical\cctor\misc\Desktop

On macOS and Linux:

./src/tests/build.sh -dir:JIT/Methodical/Arrays/lcs -dir:JIT/Methodical/cctor/misc/Desktop

Building a Test Subtree

To build a whole subtree, you have to pass the path to the root of the subtree you want with the -tree flag. Just like with any other subset, you can select more than one by repeating the -tree flag. For example, let's try building all the base services exceptions, and methodical JIT tests:

On Windows:

.\src\tests\build.cmd tree baseservices\exceptions tree JIT\Methodical

On macOS and Linux:

./src/tests/build.sh -tree:baseservices/exceptions -tree:JIT/Methodical

Building C++/CLI Native Test Components Against the Live Ref Assemblies

By default, the C++/CLI native test components build against the ref pack from the SDK specified in the global.json file in the root of the repository. To build these components against the ref assemblies produced in the build, pass the -cmakeargs -DCPP_CLI_LIVE_REF_ASSEMBLIES=1 parameters to the test build. For example:

./src/tests/build.sh skipmanaged -cmakeargs -DCPP_CLI_LIVE_REF_ASSEMBLIES=1

Test Priorities

As mentioned earlier in this guide, each test has a priority number assigned to them, and only tests with Priority 0 are built by default.

Now, here is where things get a little complicated. Test priority filtering is orthogonal to specifying test subsets. This means that even if when specifying tests, directories, and/or subtrees, you have to explicitly provide the priority if the test(s) of interest are not priority 0. Otherwise, the build will skip them.

Another very important thing to keep in mind, is that priorities are accumulative. This means that if for example, you pass -priority=1 to the build script, all priority 0 AND priority 1 tests get built.

Let's take one of the examples used in the previous subsections. Assume you want to build all JIT Methodical Div tests, including both pri0 and pri1. This is how the command-line would look:

On Windows:

.\src\tests\build.cmd dir JIT\Methodical\divrem\div -priority=1

On macOS and Linux:

./src/tests/build.sh -dir:JIT/Methodical/divrem/div -priority1

NOTE: Yes, you're seeing it right. The priority flag is a bit different between the Windows and macOS/Linux scripts.

Running the Tests

The simplest way to run in-bundle the tests you've built is by using the run.cmd/run.sh script after you've worked with build.cmd/build.sh. The running script takes flags very similarly to the build one. Let's suppose you have a Checked runtime you want to test on an x64 machine. You'd run all your built tests with the following command-line:

.\src\tests\run.cmd x64 checked

The run.cmd/run.sh scripts also have a number of flags you can pass to set specific conditions and environment variables, such as JIT Stress, GC Stress, and so on. Run it with only any one of the help flags for more details.

Once your tests are done running, a report will be generated with all the results under the artifacts/log folder, and will be named TestRun_<Arch>_<Configuration>.html. The tests that failed will be listed in a file called TestRunResults_<OS>_<Arch>_<Configuration>.err.

For individual test results and outputs, those are written in a Reports folder within the test root's directory (artifacts/tests/coreclr/<OS>.<Arch>.<Configuration>). For example, the results for the JIT Intrinsics Math Round Double test shown in Building an Individual Test, would be placed in artifacts/tests/coreclr/<OS>.<Arch>.<Configuration>/Reports/JIT/Intrinsics/MathRoundDouble_ro.

Running Individual Tests

After you've built one (or more) tests, the way to run them is by calling the generated cmd/sh script alongside them. These scripts take three optional arguments:

  • -debug: Receives the path of a debugger to run the test under in.
  • -env: Path to a .env file to specify environment variables to be set for the test. More info about dotenv can be found in their repo.
  • -coreroot: The path to the Core_Root you wish to use. Note that this flag is mandatory unless you have the CORE_ROOT environment variable set. Then, you can omit it and the script will use that one.

If this list of parameters feels familiar, it's because it's virtually the same as the arguments that corerun receives. More info on corerun in its how-to-use doc.

These scripts have a couple more hidden functionalities, which can be activated by setting their environment variables prior to running the script:

  • Run with Crossgen2: RunCrossGen2=1
  • Build and run as composite: CompositeBuildMode=1. Note that this one depends on RunCrossGen2 being set.

Let's run one of the Intrinsics tests as an example:

On Windows Command Prompt:

set CORE_ROOT=<repo_root>\artifacts\tests\coreclr\windows.<Arch>.<Configuration>\Tests\Core_Root
cd path\to\JIT\Intrinsics\MathRoundDouble_ro
.\MathRoundDouble_ro.cmd

On macOS/Linux:

export CORE_ROOT=<repo_root>/artifacts/tests/coreclr/<OS>.<Arch>.<Configuration>/Tests/Core_Root
cd path/to/JIT/Intrinsics/MathRoundDouble_ro
./MathRoundDouble_ro.sh

On Powershell:

$Env:CORE_ROOT = '<repo_root>\artifacts\tests\coreclr\windows.<Arch>.<Configuration>\Tests\Core_Root'
cd path\to\JIT\Intrinsics\MathRoundDouble_ro
.\MathRoundDouble_ro.cmd

Alternatively, instead of setting the CORE_ROOT environment variable, you can specify it directly to the test's script via the -coreroot flag, as mentioned at the beginning of this section:

On Windows:

cd path\to\JIT\Intrinsics\MathRoundDouble_ro
.\MathRoundDouble_ro.cmd -coreroot <repo_root>\artifacts\tests\coreclr\windows.<Arch>.<Configuration>\Tests\Core_Root

On macOS/Linux:

cd path/to/JIT/Intrinsics/MathRoundDouble_ro
./MathRoundDouble_ro.sh -coreroot=<repo_root>/artifacts/tests/coreclr/<OS>.<Arch>.<Configuration>/Tests/Core_Root

Tests Without a Main Method

Guide on how to run tests without a Main() Method coming soon!

PAL Tests (macOS and Linux only)

The PAL layer tests are exclusive to Unix-based operating systems. This section will go on how to work with them.

Building PAL Tests

Firstly, build them by passing the paltests subset to the main build script of the repo:

./build.sh -s clr.paltests

Running PAL Tests

Once you're done building them, you can run them all either including or excluding the disabled tests:

Including Disabled Tests:

./src/coreclr/pal/tests/palsuite/runpaltests.sh artifacts/bin/coreclr/<OS>.<Arch>.<Configuration>/paltests

Excluding Disabled Tests:

cd artifacts/bin/coreclr/<OS>.<Arch>.<Configuration>
./paltests/runpaltests.sh paltests

In order to run only specific tests, edit paltestlist.txt in under src/coreclr/pal/tests/palsuite, and adjust it to your needs (don't check in those changes though). The test(s) results will be output to /tmp/PalTestOutput/default/pal_tests.xml.

To disable tests in the CI, edit src/coreclr/pal/tests/palsuite/issues.targets accordingly.

Modifying Tests

If you need to edit any given test's source code, simply make your changes and rebuild the test project. Then, you can re-run it as needed following the instructions detailed in the sections above.

Investigating Test Failures

Upon completing a test run with run.sh/run.cmd, you may find one or more tests failing. If this is the case, there will be additional files detailing the failures in each test's Reports folder (see Running the Tests for more info regarding reports). There are 2 main files of interest:

  • <Test>.output.txt: Contains all the information logged by the test.
  • <Test>.error.txt: Contains all the information reported by CoreRun when the test process crashed.

The test's report will also contain the test command exactly as it run it, so you can investigate it further, either by running the app in the way you see fit and/or as the Running Individual Tests details.