| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
This moves all of the classes that derive from Test into a package
called test. This patch looks very large, but is mostly just moving
files around and changing imports to account for this.
So why all of this code churn? The big advantage is that through
__init__.py magic there is one module to be imported, framework.test.
This module contains all of the public classes, functions, and
constants, in one place while hiding the rest, but allowing them to be
accessed explicitly.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
|
|
The GLSLParserTest and ShaderTest subclasses of the PiglitTest class
were prepending the TEST_BIN_DIR variable onto the executable name
(such as "glslparsertest" and "shader_runner"). But the PiglitTest
class itself also does that so we were prepending TEST_BIN_DIR twice.
With unix paths, os.path.join("/abs/path1", "/abs/path2") returns
"/abs/path2".
But with cygwin, os.path.join("C:\\abspath1", "C:\\abspath2") returns
"C:\\abspath1/C:\\abspath2" which clearly causes trouble.
The solution is to only prepend the TEST_BIN_DIR path in the PiglitTest
class.
This issue was only encountered when the PIGLIT_BUILD_DIR env var was set,
which I just recently tried for the first time.
v2: don't change GleanTest, it's a subclass of Test, not PiglitTest.
Reviewed-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
Group is a subclass of dict, with no overrides. This is silly and
confusing, it's a dictionary, just use a dictionary. This has the
additional advantage of removing a lot of spaghetti dependency issues,
since nearly every module needed access to Group()
Most of this was done with the shell command:
find . -name '*py' | xargs sed -i -e 's!Group()!{}!g'
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
|
|
For whatever reason the concurrency flag was not passed to
GLSLParserTest and ShaderTest after the refactor. This patch fixes that
by always passing that keyword argument as was done before.
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
From python's PEP8:
"Constants are usually defined on a module level and written in all
capital letters with underscores separating words. Examples include
MAX_OVERFLOW and TOTAL."
This patch generated with the following shell command:
'find . -name '*py' | xargs sed -i -e 's!testBinDir!TEST_BIN_DIR!g'
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
|
|
Since PlainExecTest is either used directly or as a parent for running
"native" piglit tests, it seems fitting to give it a name that reflects
that purpose.
This patch was mostly generated with the following command:
find . -name '*.py' | xargs sed -i -e 's!PlainExecTest!PiglitTest!g
Then a handful of related changes were made manually (These were mostly
docstrings or framework unit tests related)
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
This is mostly consumed by Test's children, so putting it in the same
module makes sense. This should reduce the number of imports from core
as well, since most consumers of testBinDir also make use of ExecTest or
PlainExecTest.
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
This patch almost completely reworks ShaderTest. It was originally much
more complicated than was necessary, this code should be much easier to
work with, and removes some quirks. One of those was that ShaderTest
inherited from PlainExecTest, but used ExecTest's __init__ method, which
has implications for changing either of those classes.
v2: - remove shader_test_tests. This only had an initializer test, and
that no longer applies since ShaderTest is now a function
v3: - Don't remove shader_test_tests
- Keep ShaderTest as a class rather than function
- make looping over test files more efficient
- make errors more specific
- make tests more precise
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
Shader_test has a bizarre capability to be run as a standalone script as
well as be used as a module. This was probably useful when developing
the original shader_test implementation, but at this point it's unused
code just waiting to bit-rot adding needless complexity
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
One of the biggest changes between python2 and python3 is that print
changed from a statement to a function. This change was backported to
python2 in the __future__ module, and this patch makes that conversation
to ease the transition from python2 to python3.
v2: - Replace dict comprehension with dict(generator). This is for
python 2.6 compatability
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> (v1)
|
|
This actually makes use of the new dmesg class rather than the dmesg
functions. It touches a lot of files, but almost all of these changes
are code removal rather than code addition.
v5: - Fix dmesg handling in resume
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu> (v4)
|
|
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
This reverts commit ab2eb660925e3680e868c5f469a4ef32e6dd9a59.
This patch has a bad interaction with the locking in the json writing
library, running concurrently with this patch enabled can cause json
corruption that cannot easily be untangled, and even when it is is the
result is of low trustworthiness.
v2: - Add message explaining why this patch is to be reverted
- fix framework/status.py docstring that was mangled after the
revert
Acked-by: Daniel Vetter <daniel@ffwll.ch>
Reviewed-by: Kenneth Gruanke <kenneth@whitecape.org>
Signed-off-by: Dylan Baker <baker.dylan.c@mgmail.com>
|
|
This patch fixes a minor bug. An exception catch calls a function with a
single leading underscore instead of two leading underscores.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
i-g-t tests can take a long time, and for a bunch of reasons (bad
tuning on disparate set of platforms, stuck in the kernel which often
can be recovered by interrupting with a signal, ...) that sometimes
extends to a good approximation of forever.
Hence this adds a per-test timeout value and a bit of infrastructure
to adjust the results. Test results adjusting is done after calling
interpretResult so that we don't have to replicate this all over the
place. This might need to be adjusted for the piglit-native subtest
stuff, but otoh igt is a bit special with it's crazy long-running
tests. So I think we can fix this once it's actually needed.
The default timeout is None, so this is purely opt-in.
Note on the implementation: SIG_ALARM doesn't exists on Windows and
stackoverflow overwhelmingly recommended to go with this thread-based
approach here. But only tested on my Linux box here.
I've also timed quick.tests run a bit and the overhead due to the
additional thread we launch seems to be in the wash. So I didn't opt
to make the thread launching optional if there's no timeout limit.
v2: Also add all the boilerplate needed to handle the new test status
in summaries. For the color I've opted for two shades of light blue,
they nicely stick out from the current selection.
v3: Fix GLSLParserTest and ShaderTest. They both derive from
PlainExecTest but only call the __init__ function of the Test
baseclass. I haven't really figured why this is and also not really
what command I should pass to PlainExecTest.__init__, so just
replicated the init code for now.
v4: Initialize timeout earlier for tests where we use the ENOENT
handling to skip them.
Fix up the out, err passing from the thread. Apparently my idea of how
python closures work was completely misguided - like with a function
call a closure captures a copy of of a reference pointing at the
original object. So assinging anything to them won't have any effect
outside of the lambda. Hence we need to jump through hoops and pass an
array around. Nicer fix would be to create a class or something, but
that seems like overkill.
v5: Fixup the fraction setting in status.py
Reviewed-by: Dylan Baker <baker.dylan.c@gmail.com>
Cc: Dylan Baker <baker.dylan.c@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
|
|
simplejson is a native code version of the json module in python
(the json module in core python implements the simplejson api), which
means that simplejson can be used in place of the json module, but it
performs much better. This patch attempts to load simplejson, and falls
back to normal json if it fails.
In the summary path on a single run of quick.tests, simplejson is about
3-4 seconds faster than json, ~7 seconds vs ~11 seconds.
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
|
|
The Radeon driver writes GPU page faults to dmesg and we need to know which
tests caused them.
If there is any change in dmesg during a test run, the test result is changed
as follows:
* pass -> dmesg-warn
* warn -> dmesg-warn
* fail -> dmesg-fail
Dmesg is captured before and after each test and the difference between the two
is stored in the test summary.
The piglit-run.py parameter which enables this behavior is --dmesg. It's also
recommended to use -c0.
v2: - address some of Dylan's remarks, mainly bug fixes
- fix get_dmesg_diff
Reviewed-by: Dylan Baken <baker.dylan.c@gmail.com>
|
|
v3: - revert some things that were falsely marked as failures by an
older version of the pep8 tool
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
|
|
This fixes resume failures when paths are interpreted as unicode when the
script expects ascii.
[chadv]: Replace `if (!X and !Y) assert()` with `assert(X or Y)`.
Reviewed-and-tested-by: Chad Versace <chad.versace@linux.intel.com>
|
|
With commit dadf56dad25730556e36e9446aabfc80458cc12f the syntax
for the [require] section in the shader test data files changed for
GL ES. This patch, updates the shader_runner.py regex to also
understand the new syntax since this code parses the [require]
section of the shader test data file to determine which
shader_runner binary to run.
When dispatch is implemented for GLES all this can go away.
v2: formatting, better description, no GLSL ES, GLSL detection.
Signed-off-by: Tom Gall <tom.gall@linaro.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
|
|
Fix shader_test.py so it uses shader_runner_gles2 when running
a shader test with GL 2.0 es in the [required] section.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Signed-off-by: Tom Gall <tom.gall@linaro.org>
|
|
Commit 260f211 caused some shader tests to regress with the message:
"Failed to find [require] block".
This patch changes the python script that drives shader_runner to assume
that the test's API is GL if no GL requirement is found in the test file.
This behavior matches the behavior of the shader_runner executable, whose
default requirements are GL >= 1.0 and GLSL >= 1.10.
I tested this patch with a full Piglit run. The delta against
master-ed3903c consists only of fixes.
Reviewed-by: Tom Gall <tom.gall@linaro.org>
Tested-by: Tom Gall <tom.gall@linaro.org>
Reported-by: Ian Romanick <idr@freedesktop.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
|
|
Add a new class to shader_test.py, ShaderTest. ShaderTest.run() parses the
[require] block of the shader test to determine which shader_runner
executable should be used to run the test.
This is needed because GLES3 shader tests must be run by
shader_runner_gles3 and the GL tests by shader_runner.
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
|
|
This patch is a small refactor. It moves add_shader_test_dir() from
all.tests into a new module, shader_test.py. Follow-on commits will expand
the role of the new module.
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
|
