If a partial message is flushed into the input pipe for CMake Server,
the parser will try and parse it as a full message because of some bad
loop checks. This was introduced accidentally in commit
v3.10.0-rc1~365^2~2 (server: Refactor to make the event loop owned by
server object, 2017-03-24).
We used to detect the `rcc` features before every `rcc` list invocation
wich resulted in `rcc` be called twice for every listing operation.
Now we detect the `rcc` list capabilities once during configuration and
pass it to the cmake_autorcc target in the info file.
Instead of processing all `rcc` invocation requests in the
_autogen target that calls `cmake -E cmake_autogen ...` once,
use a dedicated custom command that calls
`cmake -E cmake_autorcc ...` for each `.qrc` file.
This allows parallel `.qrc` file processing and reduces the
workload (and complexity) in the _autogen target.
If only `AUTORCC` is enabled, the _autogen target won't be created
at all since it is now used for `AUTOMOC` and `AUTOUIC` only.
For `.qrc` files that are GENERATED a custom target is used
instead of a custom command.
Closes#17161
Introduces the standalone RCC generator class `cmQtAutoGeneratorRcc`.
Every instance of `cmQtAutoGeneratorRcc` class handles the
`rcc` invocation for a single `.qrc` file.
The class will be used in the future to allow parallel `.qrc` file
processing by calling `cmake -E cmake_autorcc <INFO_FILE> <CONFIG>`.
When rejecting the `COMPILE_LANGUAGE` generator expression on include
directories and compile definitions with Xcode, add `file(GENERATE)` to
the allowed set in the message. It is allowed and already covered
by the `RunCMake.File_Generate` test `COMPILE_LANGUAGE-genex` case.
This reintroduces the change from commit v3.10.0-rc1~69^2 (Performance:
Improve efficiency of source file lookup in cmMakefile, 2017-08-17) with
some corrections. The original was rolled back by commit
v3.10.0-rc1~52^2~1 (Revert "Performance: ...", 2017-09-25) due to
incompatibilities found. The rollback was followed-up by addition of a
test for the offending case, and this revision passes the test.
After the refactor to make CTest use std::chrono::steady_clock for the
keeping of time for test duration, there are still references to
cmSystemTools::GetTime() left.
To further adopt std::chrono for time related activities, this commit
changes those remaining references to std::chrono::system_clock::now()
calls and alters the storage from either unsigned int or double to
std::chrono::system_clock::time_point.
For ease of conversion, a converter method is added to cmXMLWriter that
converts from a std::chrono::system_clock::time_point to the number of
seconds since the UN*X epoch as that is expected behaviour. This means
no more casts as required.
Functionally should be no difference as the system_clock is implemented
in the same terms.
It was reported in issue #17345 that CTest does not use monotonic time
to report test duration. Monotonic clocks are not affected by large NTP
adjustments or things like daylight savings time.
As CMake 3.10 requires C++11, which introduced std::chrono, this commit
moves the time keeping in CTest from cmSystemTools::GetTime() to
std::chrono::steady_clock.
Fixes: #17345
A diagnostic message added in commit v3.10.0-rc1~59^2 (Windows: Improve
link-time error messages when rc or mt fail, 2017-09-22) incorrectly
reports the `mt /notify_update` special return code as a failure.
Fix the logic to consider the special return codes as success.
Fixes: #17444
The changes are mostly converting old-style explicit iterator for loops
into ranged for statements. A number of for loops had already been
changed over, but local variables had been left behind instead of being
absorbed into the ranged for statement, so these have been cleaned up
too. A couple of minor improvements were made in areas already being
updated by the for loop changes to slightly simplify the code or to
avoid unnecessary conversions between `const char*` and `std::string`.
Brad King