This new structure provides several benefits:
* Allows us to install multiple CMake projects in the same destination
dir (without include path conflicts).
- This makes shipping the trading algo sdk a trivial excercise. Simply
copy the lib and include dirs from the install dest. That's it!
* Puts us in a position to break up the monolithic libspark into smaller
modules, which are easier to manage.
Windows specific notes:
* Get rid of unnecessary script vs_filters.cmake
* Define WebsocketConnectionGateway directly in cs (instead of proto)
- The proto would no longer be in the include path passed to protoc.
Moving it to c# because there wasn't a compelling reason for it to
be a proto in the first place.
* Set spark output prefix to "lib" on windows
- So we get "libspark.dll" instead of "spark.dll", which conflicts
with the csharp lib, "Spark.dll".
* Copy libspark.dll to net8.0 & net8.0-windows in a single cmake build
target
fix build: Copy libspark to both dirs net8.0, net8.0-windows
** Migrate nlohmann-json over to vcpkg in the process **
Make it so databento is built with the same nlohmann-json lib that our
code is built with. Recently, I started getting undefined symbol errors
in databento code from nlohmann headers - presumably a conflict between
their nlohmann dep and ours.
So we don't have to worry about keeping it synchronized with the actual
targets in the build (just one more thing that can break). You can build
specific targets via the cmake cli anyways.
All other builds follow this pattern. Given the fact that we have so
many different builds now (spark, sparkd, data-streams, api-server,
etc.), it behooves us to adopt consistent naming conventions.
It may use more disk space, but we really don't have a choice. CMake
projects with different parameters, preprocessors, source files, etc.
are not designed to share binary dirs. For example, the libspark.so file
that the Spark front-end is supposed to use is not necessarily the same
as the libspark.so file the Api Server is supposed use.
* BUILD_API_SERVER (cmake cache var)
* SPARK_ENABLE_API_SERVER (preprocessor defined conditionally w/ BUILD_API_SERVER)
They were originally added to control whether or not MD Consumer extern
fns were built. They have been made redundant by the more appropriately
named EXPORT_MD_CONSUMER cache var/macro (both have same name now).
Now that we have all these different builds (sparkd, spark, api server,
data streams) - many of which are cross platform - we need a proper
toggle for this (relying on 'IF WIN32' preprocessors doesn't cut it)
Not sure how this slipped through the cracks, but I'm guessing the build
broke when we upgraded the cpp compilers (to clang 17 and gcc 13).
* Include <cstdint> in a few places
* Suppress -Wstringop-overflow in sparkd-gcc-release build
* Change const char * param to explicit std::string for proper type
deduction in call to 'std::make_tuple'. For some reason, the gcc build
was failing with an array-bounds werror (changing the arg fixed it).
Libspark's inclusion in the sln file has become redundant now that
its bins are copied to the dotnet build dir as part of the CMake build
process. Unlike msbuild, dotnet doesn't integrate nicely with c++. It
seems better to keep c++ projects out of the sln.
The path order represents the value-inheritance hierarchy. This way,
values from files earlier in the path take precedence over those from
later files.
Restructure build configurations so 'dev' builds target venue test urls,
while 'prod' builds target production urls (technically, there is only
one prod build at the moment; it's compiled w/ gcc/release). This is
necessary because the urls ccapi uses are defined at compile time (all
seem to be pre-processor directives defined in ccapi_macro.h).
For some reason, this warning isn't suppressed in ccapi_date.h, which is
included as a system header (like all headers in externals). According to GCC's
documentation, compiler warnings shouldn't be emitted from system headers.
Disabling the warning for now so we can focus on bigger fish.