Compared to Corrade, the improvement in compile time is about a minute
cumulative across all cores, or about 8 seconds on an 8-core system (~2
minutes before, ~1:52 after). Not bad at all. And this is with a
deprecated build, the non-deprecated build is 1:48 -> 1:41.
So people new to the plugin stuff can quickly get to usage introduction
and code snippets. The plugins alone don't list anything like that and
it may be *very* confusing otherwise.
When for example only CMAKE_RUNTIME_OUTPUT_DIRECTORY is set, but not the
others, the original code skipped overriding the locations altogether.
This is a valid use case, as e.g. ARCHIVE and LIBRARY_OUTPUT_DIRECTORY
tend to mess with the way Visual Studio produces and consumes *.lib
files.
Furthermore, this now also handles CMAKE_*_OUTPUT_DIRECTORY_<CONFIG> in
a similar way, which is what Conan uses for example.
Similar to the change done in Corrade, see the commit for details:
878624ac36
Wow, this is probably the most backwards-compatibility code I've ever
written. Can't wait until I can drop all that.
It limits the support for CMake 3.12+, but it's much less verbose and I
don't expect people to use ancient CMake versions with IDEs like Xcode
or VS anyway, so this should be fine.
It doesn't really work for tests that depend on more than one plugin
(because there i would need to handle all combinations, somehow), but it
does the job when the end user has such use case.
Bloaty says it saved 10 kB in Debug build of MagnumGL:
VM SIZE FILE SIZE
-------------- --------------
[ = ] 0 .debug_info +1.59Ki +0.0%
+0.4% +1.50Ki .text +1.50Ki +0.4%
[ = ] 0 .debug_str +409 +0.0%
[ = ] 0 .debug_line +276 +0.1%
[ = ] 0 .debug_abbrev +20 +0.0%
-28.6% -2 [LOAD [RX]] -2 -28.6%
[ = ] 0 [Unmapped] -4.28Ki -41.0%
-22.7% -9.23Ki .rodata -9.23Ki -22.7%
-0.8% -7.73Ki TOTAL -9.73Ki -0.1%
And 4 kB in Release:
VM SIZE FILE SIZE
-------------- --------------
+1.1% +3.44Ki .text +3.44Ki +1.1%
+1.7% +1.39Ki .eh_frame +1.39Ki +1.7%
[ = ] 0 [Unmapped] +656 +51%
-25.5% -9.47Ki .rodata -9.47Ki -25.5%
-0.7% -4.64Ki TOTAL -4.00Ki -0.4%
That's not negative, so I guess that's good. This change is of course
more significant in the context of a minimal WebGL build, where the exe
can be as little as 50 kB -- there 4 kB is almost 10% of the size.
This makes it possible to:
- finally use Magnum as a CMake subproject on Windows and have your
executables not fail to run with a "DLL missing" error (and the
setting is put to cache so superprojects just implicitly make use of
that)
- run tests on Windows without having to install first
- use dynamic plugins from a CMake subproject on any platform without
having to install first or load them by filename --- and the plugin
directory is now easily discovered as relative to
libraryLocation() of the library implementing given plugin interface
No matter how broken iOS is in CMake 3.6, $<CONFIG> seems to work there,
so reducing the amount of code and putting the configure into a single
place independently of what generator or what system/build is used.
Compared to current state it always adds Debug/configure.h instead of
putting it directly to the ${CMAKE_CURRENT_BINARY_DIR}, but the
alternative would be some CMake branching again and I just removed that,
so no.
This also prepares everything for plugin libraries being put into a
central place -- the config files don't depend on their location
anymore.
On this day, MSVC linker started complaining about missing symbols like
WavImporter::doOpenData() in the WavHeaderTest. Mind you, there's
NOTHING the WavHeader.h would use from there. Neither the test. The only
thing that connects those two together is that WavHeader.h includes
WavImporter.h (probably just a leftover from the time where those two
actually depended on each other). My suspicion is that this got
triggered due to recent changes in AbstractPlugin (it's movable now) and
MSVC attempts to instantiate the destructor or whatnot, needing
references to the privately defined virtual functions. Or something. All
that while nothing from there is EVER used.
Removing the header dependency, hopefully this fixes it. Ugh.
More consistent with what's done elsewhere, reduces header dependencies,
allows me to (later) make this independent on the AL library and also
works around a Doxygen bug. Win win!
The current testing workflow had quite a few major flaws and it was no
longer possible after the move of Any* plugins to core. Among the flaws
is:
* Every plugin was basically built twice, once as the real plugin and
once as a static testing library. Most of the build shared common
object files, but nevertheless it inflated build times and made the
buildsystem extremely complex.
* Because the actual plugin binary was never actually loaded during the
test, it couldn't spot problems like:
- undefined references
- errors in metadata files
- mismatched plugin interface/version, missing entry points
- broken static plugin import files
* Tests that made use of independent plugins (such as TgaImageConverter
test using TgaImporter to verify the output) had a hardcoded
dependency on such plugins, making a minimal setup very hard.
* Dynamic loading of plugins from the Any* proxies was always directed
to the install location on the filesystem with no possibility to
load these directly from the build tree. That caused random ABI
mismatch crashes, or, on the other hand, if no plugins were
installed, particular portions of the codebase weren't tested at all.
Now the workflow is the following:
* Every plugin is built exactly once, either as dynamic or as static.
* The test always loads it via the plugin manager. If it's dynamic,
it's loaded straight from the build directory; if it's static, it
gets linked to the test executable directly.
* Plugins used indirectly are always served from the build directory
(if enabled) to ensure reproducibility and independence on what's
installed on the filesystem. Missing presence of these plugins causes
particular tests to be simply skipped.
* Plugins that have extensive tests for internal functionality that's
not exposed through the plugin interface are still built in two
parts, but the internal tests are simply consuming the OBJECT files
directly instead of linking to a static library.
Vladimír Vondruš