Open sammyne opened 4 years ago
Thanks, I suggest we can discuss this proposal in several aspects:
I also added test in your proposal:
|-teaclave
|-...
|-examples // A CARGO WORKSPACE
|-CMakeLists.txt // passing env variables to help building examples
|-Cargo.toml // just list the examples of interest, and exclude any dislikes
|-hello-world // examples
|-...
|-...
|-services
|-apps // A CARGO WORKSPACE, gathering all untrusted applications driven by enclaves
|-CMakeLists.txt // passing env variables to help building apps
|-Cargo.toml // lists all apps
|-access_control // the app part of the access control service
|-authentication // the app part of the authentication service
|-execution // the app part of the execution service
|-frontend // the app part of frontend service
|-management // the app part of the management service
|-scheduler // the app part of the scheduler service
|-storage // the app part of the storage service
|-enclaves // A CARGO WORKSPACE, gathering all trusted enclaves
|-CMakeLists.txt // passing env variables to help building enclaves
|-Cargo.toml // lists all enclaves
|-access_control // the enclave part of the access control service
|-authentication // the enclave part of the authentication service
|-execution // the enclave part of the execution service
|-frontend // the enclave part of frontend service
|-management // the enclave part of the management service
|-scheduler // the enclave part of the scheduler service
|-storage // the enclave part of the storage service
|-proto
|-tests
|-apps
|-CMakeLists.txt
|-Cargo.toml
|-functional
|-integration
|-unit
|-enclaves
|-CMakeLists.txt
|-Cargo.toml
|-functional
|-integration
|-unit
|-fixtures
|-...
The only draw back for me is code of same service are separated in two different places. Another important issue we need to take consideration is the compilation performance. We don't want to see any downgrade on the compilation time caused by multiple compilation. For the current proposal, I believe the tests apps/enclaves will be compiled separately, resulting compiling same crates twice.
In terms of our building system with cmake, I'm not sure whether this will make it more simpler. I can imagine that our previous workarounds with symlinks can be removed.
Overall, I'm okay with the new directory organization. However, since this is a major changes of a lot of files, I'd like to suggest to put this change on hold before we finished all service implementations to avoid potential merging conflicts.
Under every relevant folders (services/tests), the cmake targets to build enclaves should depend on the ones to build apps, and this dependency is achieved through CMakeLists.txt in the services/tests folders.
So I don't get the following point~
For the current proposal, I believe the tests apps/enclaves will be compiled separately, resulting compiling same crates twice
Help me out, please 😄 @mssun
I see. Okay, then it seems that the compilation time won't be changed.
Currently, compiling the project relies heavily on the cmake, rendering CMakeLists.txt so large. That's not so friendly for starters to play with it.
It would be nice if the the CMakeLists.txt is slimmed down, delegating part of its job to cargo workspaces, which would make the CMakeLists.txt easier to understand.
Therefore comes the proposal of arranging the project structure as follows (only
examples
,services
folders is of interest)Pro:
add_cargo_build_target
can be removedsetup_cmake_tomls
andparse_cargo_package.py
can be removedJust some personal ideas ~