We check them out from a git tag, so this *should* be a noop. However,
cmake seems to want to rebuild these dependencies every time you run
make as it assumes the dependency *might* have been updated. (In
practice, this isn't completely awful, as make notices the files haven't
changed, so files don't actually get recompiled. But the configure step
is still re-run and all the files still need to be rescanned.)
Skipping the update step speeds up the build considerably.
On my linux box, running:
cmake .. && make -j && time make -j
takes ~8.5s prior to this CL and ~6.5 afterwards. (6s is used by the
test suite.) The upcoming protobuf addition would otherwise have made
this much worse. (It takes a long time to ./configure.)
We originally had this as it was thought that cmake would spawn n*m jobs
(where n is the number of jobs it was instructed to create, ie. make
-j8, and m is the number of external projects.) However, it isn't
supposed to do that, and doesn't appear to be doing that right now. So
we'll remove this.
If the problem re-appears, we'll add these back in. (Symptom was being
unable to spawn /bin/sh.)
* Import iterator_adapters from google3
* Remove -Wconversion which is annoyingly hard to satisfy
* Strip dependency on absl_container from iterator_adapters_test
* Format and lint iterator_adaptors
* More flexible copyright checking in Travis
* Move -fvisibility-inlines-hidden from common_flags to cxx_flags
This option isn't supported by C (and causes the build to fail under
at least rodete).
Manifested error was:
```
-- Looking for include file openssl/rand.h
-- Looking for include file openssl/rand.h - not found
CMake Error at core/src/firebase/firestore/util/CMakeLists.txt:94 (message):
No implementation for SecureRandom available.
-- Configuring incomplete, errors occurred!
```
which is completely misleading. :(
* Fix exception include for std::logic_error
Was causing compiler error on (at least) rodete.
(http://en.cppreference.com/w/cpp/error/logic_error suggests that
stdexcept is the correct header.)
* Remove 'const' from vector contents.
vectors cannot contain const objects as the contained objects are
required to be assignable and copy constructable. (Not *quite* strictly
true; since c++11, this requirement now depends on the operations
performed on the container. But my compiler objects at any rate.)
http://en.cppreference.com/w/cpp/container/vectorhttps://stackoverflow.com/questions/8685257/why-cant-you-put-a-const-object-into-a-stl-container
* Add brackets as suggested (required) by my compiler.
* Add missing include
For LLONG_MIN, etc.
* Enable gnu extension to c++11 for firestore/util *test*.
We disable them by default (in cmake/CompilerSetup.cmake) but our tests
requires hexidecimal floating point, which is not supported in c++11
(though is supported in C++17, gnu++11, and others).
http://en.cppreference.com/w/cpp/language/floating_literalhttps://bugzilla.redhat.com/show_bug.cgi?id=1321986
* Fix printf format for uint64_t
http://en.cppreference.com/w/cpp/types/integerhttps://stackoverflow.com/questions/9225567/how-to-print-a-int64-t-type-in-c
* Allow 0 length printf template strings in tests.
* Get rid of a multi line comment.
The backslash is apparently interpreted by the compiler cause the next
line to be ignored too. In this case, the next line is (currently) a
comment, and would be ignored anyways. (But that doesn't stop the
compiler from yelling.)
* Run ./scripts/style.sh
* Rewrite cc_test to take named arguments
Cut down on build file verbosity by having cc_test take SOURCES and
DEPENDS. The separate invocation of target_link_libraries is no longer
necessary.
* Add a cc_library rule to parallel cc_test
This cuts down on build file verbosity.
* Automatically add OBJC_FLAGS to cc_libraries if applicable
* Exclude platform-specific libraries from 'all'
This is makes it possible to declare this kind of library
unconditionally. Usage within a test or as a dependency will actually
trigger building.
* Restore secure_random_test.cc; clean-up comments
* Clean up quoting and other minor issues
* Reorganize CMake build output
Make it clearer which parts of the output pertain to external projects.
* Use a consistent ordering of ExternalProject arguments
* Prevent the top-level build from running in parallel
This prevents spurious failures when running make -j.
* Actually parse arguments in the xcodebuild function
* Use ExternalProject features when available
* submodule limits from CMake 3.0
* shallow clones from CMake 3.6
* git progress output from CMake 3.8
* Only build the parts of leveldb we need
Skip building the tools and other libraries
* Avoid installing ExternalProjects
Consume build output directly so that we can build just the targets we
need. Installing causes all targets to be built.
This doesn't matter as much for these targets but the gRPC build
includes a ton of stuff we don't need so it's worth adopting this as a
general strategy.
* Define an external build for grpc
* Test that grpc can link successfully.
* Add a FindGRPC CMake module
* Actually comment ExternalProjext_GitSource
* Port StringPrintf from //base.
Prefer this to approaches based on variadic templates. While the variadic
template mechanisms are strictly safer, they result in binary bloat we
can't afford.
This is essentially the same StringPrintf previously open sourced as a
part of protobuf, though updated for C++11 which saves a copy and a
temporary buffer on the heap.
* Add abseil as a subdirectory of Firestore
This saves having to redefine all the libraries that abseil defines as
imported libraries.
* Rename firebase_firesture_util_log_* targets
Cut the log out of the name to reflect that these will get more
components besides just logging.
* Rename FIREBASE_BINARY_DIR to FIREBASE_INSTALL_DIR
Make this consistent with the outer superbuild and also make the
association with CMAKE_INSTALL_PREFIX more obvious.
* Build and test log_stdio separate from the rest of util
This is in preparation for adding a test for log_apple
* Build and test log_apple under CMake
Also add notes about how FIRLogger's debug mode can break this test
* Refactor log_apple to cut down duplicate switch statements
There's also a slight reduction in final binary size.
* Deprecate FIRAppEnvironmentUtil sharedApplication
* Remove sharedApplication private API.
This also allows us to remove an AppKit dependency.
* Disable autolink when building FirebaseCore (#606)
This prevents the dependencies of FirebaseCore's dependencies from
bleeding into its interface.
The proximate benefit of avoiding this is that FirebaseCore no longer
ends up depending on ColorSync.framework which became a top-level system
framework in 10.13. This makes it possible to build against the
resulting FirebaseCore.framework on macOS 10.12 using Xcode 9 (with the
macOS 10.13 SDK).
* Don't bother specifying a download directory to CMake ExternalProject
* Teach CMake to build pure Xcode projects as dependencies
This allows downstream code (like log_apple.mm) to consume this for
testing within the CMake build without requiring a CMake-native build
for these components.
This makes integrating these components into the cmake build essentially
free from the point of view of the consumed component.
* Get the CMake build semi-working on Linux again
Many prebuilt versions of cmake on Linux lack the ability to download
over https so use git to get googletest.
Don't attempt to build FirebaseCore on Linux; there's no xcodebuild.
Note the build is still ultimately broken because we don't yet have an
alternative to arc4random on Linux but at least this is no more broken
than it was before.
* Use Debug as the default build type if unspecified
* Add a leveldb ExternalProject, FindLevelDB module, and use it in Firestore
* Accept an externally built leveldb via LEVELDB_ROOT
* Clone leveldb via https: instead of git:
* Remove extra DOWNLOAD_DIR which doesn't work with GIT sources
Rich Gowman