Age | Commit message (Collapse) | Author |
|
Can test selected directory using list file or argument
|
|
* Fix run-test.sh
This makes run-test.sh work in x64/linux.
Remove useless argument and add --runtime argument to pass runtime directory
Not works in cross compile yet (ex. build in x64 & run in arm)
* Fix run-test.sh
remove chmod and check RunTests.sh existing
|
|
|
|
|
|
* Fix casing for native binaries
* Convert native bin dir to 'native' from 'Native'
|
|
* Skip test runs where doesn`t have test cases for certain target, so
generate errors with exit code 127.
|
|
|
|
|
|
|
|
* add `--test-dir <path>` : specify a test directory path
* add `--test-dir-file <path>` : specify a file that contains test directory list
|
|
for running unit-tests one by one.
|
|
If a ctrl-c interrupt occurs, the test runner will stop and
return the total number of failed.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
|
|
After some discussion we have decided to back out the change made in
commit 0eb598b and disable the offending tests with multiple target
groups. A long-term fix will be submitted later.
|
|
Requires https://github.com/dotnet/buildtools/pull/751
This makes all the tests resolve using NETCoreApp1.0 instead of
DNXCore50. I've removed the netstandard1.6 import since that is no
longer needed. I've also removed the workarounds from tests that were
targeting netcoreapp1.0 on their own. For restore compat I've added an
DNXCore50 import to every test project. These can be removed once
test-runtime packages are updated to target NETCoreApp1.0.
I've updated all the test scripts which were assuming DNXCore50 as well
as docs which referred to this path.
|
|
Today when a test that specifies a target group is built it will place the
output for that target group in a subdirectory of the build with an empty
target group; this leads to a race condition in the build when
ArchiveTests is set to true. Instead, place each test build in its own
directory; and in cases where TargetGroup is empty use netstandard as the
subdirectory name.
Update the test runner scripts with the new default directory.
Pick up latest build tools for some fixes for the official build.
|
|
Fixes #8029
|
|
The build now generates a script that actually finishes layout out the
tests and invokes CoreRun, so we need to invoke that instead of calling
it ourselves.
|
|
This form: "A =bcd" has no effect.
There mustn't be whitespace between a variable and =.
|
|
pdates the netci.groovy and runtest.sh files to pass -notrait Category=IgnoreForCI to build.cmd and run-test.sh in Jenkins so that we can disable tests from running when they're being run in Jenkins.
|
|
The PR job definitions are shared between release/1.0.0-rc2 and master
However, in release/1.0.0-rc2, the way you invoke run-test.sh differs
from master because we don't have the .builds changes there and hence
the test layouts are different. Update the job definitions to use the
release/1.0.0-rc2 way of invoking run-test.sh and then update the
comment in run-test.sh that explains why we strip some stuff from
the test paths in some cases.
|
|
Since #6572 was merged, we only build test projects for the platforms
they test. This means the UnsupportedPlatforms metadata is no longer
needed (instead using the .builds file to control when the project is
built is the right thing to do).
Leverage this in run-test.sh, we now simply loop over all the tests in
the relevent folders under bin/tests, depending on the platform in
question.
|
|
|
|
|
|
|
|
|
|
It's leftover from #6491
|
|
|
|
Various tests depend on CurrentCulture, and when run in the service in CI, LANG is defaulting to empty which then causes the locale to default to "POSIX", which causes problems for some of the expectations in various tests. This ensures that LANG is set.
|
|
coreclr will print out the Environment.FailFast message to stderr, but only when the PAL_OUTPUTDEBUGSTRING environment variable is set. I'm setting it in run-test.sh so that fail fasts in CI and other test runs will show something in the log.
|
|
|
|
|
|
Without these, it can be difficult to spot the cause of an error in an output log.
|
|
https://github.com/dotnet/buildtools/pull/332 disabled running benchmarks as part of unit tests, but only on Windows. They're still running on Linux and OS X, and are causing builds to timeout, e.g. the Process tests are currently taking over 5 minutes:
```
11:03:02 System.Diagnostics.Process.Tests Total: 77, Errors: 0, Failed: 0, Skipped: 0, Time: 318.916s
```
|
|
|
|
Now that we can properly restore cross platform binaries when building
out the test library, we no longer need the the step where we copy over
the Linux product build to the overlay. There was some other overlay
related stuff that we could remove as well (e.g. the need for picking up
the test host from a seperate place).
We continue to copy the runtime itself and the native corefx bits. In
time we should remove the copy of the runtime and mscorlib.dll (but for
now we continue to use it to speed up the rate at which we can implement
features in CoreCLR and test them in CoreFX, since if we we used the
runtime that came down from NuGet it would always be at least a day
old).
I have also changed the defaults of run-test.sh to ease folks doing
local development (e.g. it defaults to $OS for the version of the test
binaries instead of always using Windows versions).
In addition, I updated build.sh to set TestNugetRuntimeId to
ubuntu.14.04-x64 or osx.10.10 if you are building on "Linux" or
"Darwin". Obviously this is not complete because we don't actually try
to figure out your distro and version but it better than just deploying
windows stuff when building the test runtimes and run-test.sh will
replace all the native bits which is the major thing that changes anyway.
|
|
Previously finding out which xunit projects failed was difficult because
unlike Windows we didn't write a message when xUnit returned with a non
zero exit code. In cases where tests failed, it was somewhat easy to
search the logs for xUnits "a test failed" messages but if the runtime
crashed, things were much harder.
Update run-test.sh to write an error message in the same style as what
we do on windows as part of the xUnit invocation from MSBuild.
While I was in the area, I reduced some noise in the log by pushing the
output of a grep for supported platforms to /dev/null. We print a
message anyway when we detect a project is unsupported, so the output of
grep is not interesting.
|
|
Before we moved to using packer to slim down our test drop, we had
Jenkins exclude the System.Threading.Tasks.Dataflow.WP8 assemblies when
uploading build artifacts. This meant that the OSX and Linux jobs would
never copy them down and hence they could not be picked up by
run-test.sh (and overwrite the correct version of
System.Threading.Tasks.Dataflow when building the test overlay).
With packer, we now archive everything, so depending on the order in
which find discovered files, we may have copied over the WP8 version,
which does not have ETW support.
For now, update run-test.sh to exclude this folder when copying stuff
over. The correct long term fix is to just do what we do on
Windows (have the build produce the correct test layouts and get out of
this run-test.sh test overlay game).
Fixes #3520
|
|
|
|
|
|
Add options to run-test.sh to generate code coverage reports for coreclr
|
|
Changes in run-test.sh to download the unix code-coverage report generator tools from myget package 'unix-code-coverage-tools.1.0.0.nupkg' and use them to generate html style graphic code-coverage reports for coreclr.
|
|
OSX doesn't have bash version 4, so globstar doesn't work. Use find -regex instead.
|
|
The test layout folder that is created when we run tests includes an
mscorlib native image. This native image is specific to windows (since
we use the windows version of the runtime when building a test layout,
even when building a test layout for Linux or OSX) and the runtime
will not boot if it is present.
When we run tests from CI, we instruct Jenkins to not copy native
images, but if you are running tests by hand you need to either make
sure you don't copy native images over or remove them before running
tests.
With more folks wanting to use run-test.sh by hand, we should just
make run-test.sh responsible for cleaning up these native images.
Fixes #2605
|
|
Remove run-test native binaries workaround
|
|
Becuase of a lack of a native binaries build leg, we were building the native corefx binaries during run-test. Now that we have those legs in CI, we can remove the workaround.
|
|
copying test assemblies to the 'Overlay' directory.
|
|
Globstar is not enabled by default, and a recent commit attempting to increase the number of directories being run for the Linux build actually skipped a bunch of stuff. Enable globstar in the script (this is local) to fix.
Mark DriveLetter_Unix as a failing test, which appears to have snuck in in the meantime
|
|
* Rename shims to match their underlying platform API names
* Update guidelines
* Add errno conversion TODO
* Account for OS X in run-test.sh
|
|
|
|
run-test.sh is currently looking for test projects of the form
src/**/tests/*.csproj, but some of our projects are nested a level down
from that, e.g. src/**/tests/something/*.csproj.
|