suite stats were not correct, a mixup between skipped and filtered
suites was leading to inconsistent numbers. This is now fixed.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Do not report status issues as errors, very confusing and developer end
up looking at the wrong thing, instead, treat those as warnings and
count them and report them at the end.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
It is confusing to report filtered testcases as testcases that were
selected but not exexuted. If they are filtered, then there should not
be considered as selected.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
custom simulator needs an executable defined, or else we will have tests
marked runnable with nothing to run them which would result in an
error/warning.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Fix Twister TestCase statuses left not assigned ('NONE') in these 'NOTRUN'
situations:
* a test suite has `integration platform` which is not available to run.
* `--cmake-only` execution.
Signed-off-by: Dmitrii Golovanov <dmitrii.golovanov@intel.com>
Twister shall now verify that the user does not
use an obsolete Python version. If user's Python
is deemed too old, it will raise a relevant error.
This check will also happen when running Twister
via west.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
When using the --build-only into --test-only
Twister setup, NOTRUN statuses were not properly rerun.
Now they are properly run again if runnable.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
ExecutionCounter has been expanded and now hold i.a.
more information on the statuses of TestCases.
This information is now incorporated in relevant summaries
- runner.py and reports.py.
Layout of those was changed to present that
and previous information in a clear and concise way.
TestInstance execution counter now is more intuitive.
Instances filtered out before running are no longer included there.
Retries now properly reset the counter.
TestCases with None and other incorrect final statuses
are logged as errors, but do not
exit Twister with a nonzero exit code.
This is because None statuses, although incorrect,
are currently common.
Inconsistent spacing in ERROR and FAILED fixed.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
scripts: Dmitri fix
Fix of a problem noticed by Dmitri
Removed unnecessary additional spaces when
printing FAILED and ERROR status.
Now TwisterStatus.get_color is used more.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
As per its creators, CMake is written with a capital "M".
The initial reason for this change is that I want Twister to print
"ERROR : CMake build failure" instead of "ERROR : Cmake build failure".
Signed-off-by: Reto Schneider <reto.schneider@husqvarnagroup.com>
Set soc_root using Path to avoid wrong generated path in the list of
soc_roots and other roots read from module.yml file
Fixes#80531
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Cleanup notrun tests with the -M option. This is used in the weekly
build and fails because devices will run out of space.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Fixes: #69548
Support extending an existing board with new board variants.
This commit introduces the following changes to allow a board to be
extended out-of-tree.
The board yaml schema is extended to support an extend field which
will be used to identify the board to be extended.
A board 'plank' can be extended like this:
> board:
> extend: plank
> variants:
> - name: ext
> qualifier: soc1
For the rest of the build system this means that there is no longer a
single board directory.
The existing CMake variable BOARD_DIR is kept and reference the
directory which defines the board.
A new CMake variable BOARD_DIRECTORIES provides a list of all
directories which defines board targets for the board.
This means the directory which defines the board as well as all
directories that extends the board.
Signed-off-by: Torsten Rasmussen <Torsten.Rasmussen@nordicsemi.no>
if alias or shorthand name is provided on the command line, we need
convert this to complete target name for the filters to work.
Fixes#80332
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
--device-testing implies runnable tests, this was the case before the
rework and was missed in one spot, so instead do that directly in the
argument parser.
Fixes#80428
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
We now use hwmv2 to list boards instead of relying on twister specific
config files.
One yaml files (twister.yaml for now) will have all the data needed for
all possible targets and variations of a board reusing most of the data
where possible and variations can override the top level data.
Twister keeps track of 'aliases' of boards and identifies that for
example native_sim is the same as native_sim/native, so either names
will be possible in both test yaml files or on the command line,
however, the reporting will always use the full name, so no there is no
confusion about what is being tested/built.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Adds a new NOTRUN status, which indicates
that a test was successfully built, but
not run on account of being not
runnable in given test instance.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
like extra_confs, add extr_args filterable by
soc:<arch>:<extra_args> or
platfrom:<platform name>: <extra_args>
simulation:<simulator name>: <extra_args>
which will only apply to given platform or arch
Signed-off-by: Hake Huang <hake.huang@oss.nxp.com>
Testcase execution time doesn't match between twister.xml and
twister.log. Testcase execution time is the sum of the previous
testcases' execution time plus its own execution time in
twister.xml.
This patch fixes the issue above.
Signed-off-by: Ioannis Damigos <ioannis.damigos.uj@renesas.com>
Fix trailing `\\r\\n` (escaped CR/LF) didn't cut off because of rstrip()
removed by #58338, so the CR/LF suffix was never found as the actual line
end was `\\r\\n\n`.
Add ANSI code sequence to `test_handlers` Twister unit test.
Signed-off-by: Dmitrii Golovanov <dmitrii.golovanov@intel.com>
Separate the pickled EDT generation from the C-Macro header
generation in gen_defines.py to have a more clear responsibility
of the scripts in the DTS parsing process.
Signed-off-by: Benedikt Schmidt <benedikt.schmidt@embedded-solutions.at>
Currently, debug logging in the console and verbosity
are tightly coupled - verbosity of level 2 and higher
enables logging at the debug level.
This change introduces a separate Twister flag
responsible for controlling the debug logging,
while leaving the rest of verbosity unchanged.
This allows for controlling the verbosity on
both logging levels, according to one's needs.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
In case of problem with parsing hex data from coverage dump,
do not create empty gcda file.
Such empty file will break gcovr parsing.
Signed-off-by: Piotr Kosycarz <piotr.kosycarz@nordicsemi.no>
This small change concerns the following filter functions:
1. `dt_compat_enabled(C)`:
There's a node with compatible `C` and status "okay".
2. `dt_enabled_alias_with_parent_compat(A, C)`:
There's a node with alias `A` and status "okay", and its parent
has compatible `C`.
3. `dt_label_with_parent_compat_enabled(L, C)`:
There's a node with label `L`, and its parent has compatible `C`
and status "okay".
All three functions involve checking whether some node or its parent has
a given compatible, but the way this has been checked is inconsistent.
Function (1) has done it with this Python conditional:
compat in node.compats
while (2) and (3) have used:
parent.matching_compat == compat
The first check works well with nodes that have multiple compatibles,
and it is more aligned with the notion of "has_compat" as seen in the
devicetree macros for C, CMake, and Kconfig.
Arguably, `matching_compat` shouldn't have been used here, because it is
actually a property of a node's binding, moreso than of the node itself.
In practice, it's usually equal to the first compatible for which edtlib
has found a binding, which at first glance is just more constrained than
the `node.compats` check. However, there also exist obscure cases where
the `node.compats` are empty, while the `node.matching_compat` is not.
For now, the three functions can use a combined check, to improve
consistency and utility while avoiding breakage:
node.matching_compat == compat or compat in node.compats
Signed-off-by: Grzegorz Swiderski <grzegorz.swiderski@nordicsemi.no>
It has been deprecated since Zephyr v2.6.0.
The filter that replaced it - `dt_enabled_alias_with_parent_compat` -
had shared code with it, which can now be inlined into `ast_expr()` to
match the parser's overall code structure.
Signed-off-by: Grzegorz Swiderski <grzegorz.swiderski@nordicsemi.no>
When at verbosity 1, we print out the status of TestInstances.
This makes it harder to notice changes at TestCase level,
which require perusing the logs.
This adds TestCase status and reason printing
if verbosity level is 2 or more.
Reason printing is suppressed if the reason is empty or None.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
Running sysbuilt tests fails because of missing "run" target.
This adds the default domain context to the command.
Signed-off-by: Wilfried Chauveau <wilfried.chauveau@arm.com>
Some dict.get() calls did not use a TwisterStatus
as a default value, thus using a NoneType where
TwisterStatus should appear.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
Fixes bad usage of single backticks in lieu of double backticks for
rendering inline literals, or simple '*' for italics.
When appropriate, a better construct than double backticks has been
selected (ex. :file:, :kconfig:option:, :c:func:, ...), or proper :ref:
have been used if the original intention was to have a link.
Signed-off-by: Benjamin Cabé <benjamin@zephyrproject.org>
If you interrupt process() operation, we want Twister
to exit as gracefully as it can. This avoids the
UnboundLocalError that could appear e.g.
when interrupting the operation via SIGINT.
Signed-off-by: Lukasz Mrugala <lukaszx.mrugala@intel.com>
This changes how some arguments are set in the `Handler`s.
`options`, `generator_cmd` and `suite_name_check` are now passed as
arguments to the constructor rather than injected from an other module.
Signed-off-by: Wilfried Chauveau <wilfried.chauveau@arm.com>
Anas Nashif
Dmitrii Golovanov
Lukasz Mrugala
Armin Kessler
Reto Schneider
Torsten Rasmussen
Mara Furland
Hake Huang
Ioannis Damigos
Benedikt Schmidt
Piotr Kosycarz
Grzegorz Swiderski
Wilfried Chauveau
Benjamin Cabé
Pieter De Gendt