Refactoring in commit 5ff0b4ed57 (cmCTestMultiProcessHandler:
Consolidate test readiness checks, 2023-10-20, v3.29.0-rc1~378^2~4)
accidentally broke `RESOURCE_LOCK`. Fix it and replace the previous
test with one that would have caught this.
Fixes: #25843
Under job server integration, added by commit 80fe56c481 (ctest: Add
support for running under a make job server on POSIX systems,
2023-11-15, v3.29.0-rc1~324^2), use a very high default so that
parallelism is effectively limited only by available job server tokens.
Otherwise, choose a default limit based on the number of processors.
Also allow passing `0` to specify unbounded parallelism.
Fixes: #25739
Once a test is ready to run, count it against the number of tests to
start in the current batch whether or not the test process actually
starts successfully. If a test process does fail to start, simply
schedule a new startup batch on the next loop iteration.
When a test finishes, defer starting new tests until the next loop
iteration. That way, if multiple tests finish in a single loop
iteration, we can free all of their resources first, and then start
a new batch of tests.
The `TestRunningMap` and `TestFinishMap` members have not been used
since commit 44017a4767 (CTest: handle dependent and non dependent test
requirements equally, 2013-10-17, v3.0.0-rc1~451^2~7).
#pragma once is a widely supported compiler pragma, even though it is
not part of the C++ standard. Many of the issues keeping #pragma once
from being standardized (distributed filesystems, build farms, hard
links, etc.) do not apply to CMake - it is easy to build CMake on a
single machine. CMake also does not install any header files which can
be consumed by other projects (though cmCPluginAPI.h has been
deliberately omitted from this conversion in case anyone is still using
it.) Finally, #pragma once has been required to build CMake since at
least August 2017 (7f29bbe6 enabled server mode unconditionally, which
had been using #pragma once since September 2016 (b13d3e0d)). The fact
that we now require C++11 filters out old compilers, and it is unlikely
that there is a compiler which supports C++11 but does not support
#pragma once.
Renaming these ahead of other refactoring which will use the
previous names. The previous names more accurately reflect their
purpose after this commit anyway (talking about locking and
unlocking rather than allocating and deallocating).
format can be 'human' to print the current text format or 'json-v1' to
print the test object model in json format and is useful for IDEs who
want to gather information about the tests. Defaults to 'human' format.
The CTestTestStopTime test has been failing sporadically because the
stop time causes the first internal test to have a timeout short enough
that we might hit it and start the second test just before the stop time
is reached. Instead we should track when a timeout is shortened in
order to stay within the stop time. If a test times out for this reason
then we should consider the stop time reached and not start any more
tests.
The `ctest --test-load` option is implemented in `StartNextTests` by not
starting any tests when the load is too high and instead sleeping and
then returning. Prior to commit v3.11.0-rc1~117^2 (CTest: Re-implement
test process handling using libuv, 2017-12-10) our outer loop in
`RunTests` would immediately call `StartNextTests` again. However, now
the `uv_run` loop may simply terminate if there are no tests running
because no events are left pending.
Fix this by converting the sleep in `StartNextTests` into a libuv timer
that it starts instead. This avoids leaving `uv_run` with no pending
events. In the case that there are other running tests this also allows
CTest to detect when they finish even if it during the wait period where
we previously slept.
This regression was not caught by the test suite because it only
verified that we do not start new tests when the load was too high and
not that we proceed to start tests when the load drops. Revise the test
suite to cover both.
Fixes: #18338
Run the `clang-format.bash` script to update all our C and C++ code to a
new style defined by `.clang-format`. Use `clang-format` version 6.0.
* If you reached this commit for a line in `git blame`, re-run the blame
operation starting at the parent of this commit to see older history
for the content.
* See the parent commit for instructions to rebase a change across this
style transition commit.
In commit v2.8.0~170 (ENH: Added ctest test options PROCESSORS and
RUN_SERIAL, 2009-09-07) CTest learned to track the number of processors
allocated to running tests in order to balance it against the desired
level of parallelism. Extend this idea by introducing a new
`PROCESSOR_AFFINITY` test property to ask that CTest run a test
with the CPU affinity mask set. This will allow a set of tests
that are running concurrently to use disjoint CPU resources.
If `StartTestProcess` does not start a test, propagate this information
back up to the `StartNextTests` loop so that it can move on to another
candidate without allocating processors to a test that didn't run.
Otherwise we have to wait for the next time `RunTests` loops around and
calls `StartNextTests` again.
Per-source copyright/license notice headers that spell out copyright holder
names and years are hard to maintain and often out-of-date or plain wrong.
Precise contributor information is already maintained automatically by the
version control tool. Ultimately it is the receiver of a file who is
responsible for determining its licensing status, and per-source notices are
merely a convenience. Therefore it is simpler and more accurate for
each source to have a generic notice of the license name and references to
more detailed information on copyright holders and full license terms.
Our `Copyright.txt` file now contains a list of Contributors whose names
appeared source-level copyright notices. It also references version control
history for more precise information. Therefore we no longer need to spell
out the list of Contributors in each source file notice.
Replace CMake per-source copyright/license notice headers with a short
description of the license and links to `Copyright.txt` and online information
available from "https://cmake.org/licensing". The online URL also handles
cases of modules being copied out of our source into other projects, so we
can drop our notices about replacing links with full license text.
Run the `Utilities/Scripts/filter-notices.bash` script to perform the majority
of the replacements mechanically. Manually fix up shebang lines and trailing
newlines in a few files. Manually update the notices in a few files that the
script does not handle.
Use clang-tidy's performance-unnecessary-value-param checker to find
value parameter declarations of expensive to copy types that are not
modified inside the function. Ignore findings in kwsys.
After applying the fix-its, manually change `const T&` to `T const&`.
Run the `Utilities/Scripts/clang-format.bash` script to update
all our C++ code to a new style defined by `.clang-format`.
Use `clang-format` version 3.8.
* If you reached this commit for a line in `git blame`, re-run the blame
operation starting at the parent of this commit to see older history
for the content.
* See the parent commit for instructions to rebase a change across this
style transition commit.
Add a TestLoad setting to CTest that can be set via a new --test-load
command-line option, CTEST_TEST_LOAD variable, or TEST_LOAD option to
the ctest_test command. Teach cmCTestMultiProcessHandler to measure
the CPU load and avoid starting tests that may take more than the
spare load currently available. The expression
<current_load> + <test_processors> <= <max-load>
must be true to start a new test.
Co-Author: Zack Galbreath <zack.galbreath@kitware.com>
The original implementation of the RUN_SERIAL test property worked by
having such a test consume all available processors. Instead use an
explicit flag to indicate that a serial test is running. This avoids
artificially inflating the number of processors a test is expected to
consume.
Casts from std::string -> cmStdString were high on the list of things
taking up time. Avoid such implicit casts across function calls by just
using std::string everywhere.
The comment that the symbol name is too long is no longer relevant since
modern debuggers alias the templates anyways and the size is a
non-issue since the underlying methods are generated since it's
inherited.
The first regression resulted in endless looping due to unrun test
dependencies. The second regression prioritized all tests with dependencies
in serial test runs.
