Improvements to corrosion_link_libraries() #508
Conversation
- Allow linking by full path and non-target library name - Propagate linking dependencies to artifacts
| corrosion_link_libraries(${target_name} ${libs}) | ||
| elseif(IS_ABSOLUTE ${library}) | ||
| # Linking via full path | ||
| corrosion_add_target_local_rustflags(${target_name} "-Clink-arg=${library}") |
There was a problem hiding this comment.
Could you add
- a link to the relavant rustc documentatation (https://doc.rust-lang.org/rustc/command-line-arguments.html#linking-modifiers-verbatim)
- ideally a test project that utilizes absolute path linking.
| set_property( | ||
| TARGET _cargo-build_${target_name} | ||
| APPEND | ||
| PROPERTY INTERFACE_LINK_LIBRARIES |
There was a problem hiding this comment.
_cargo-build_${target_name} is a custom target, so I don't believe this property will have any effect.
Is your intention here to add a rule to trigger rebuilds, or to propogate the dependencies to C/C++ consumers of a Rust static library ? If it's the latter, then #506 should be sufficient.
| "-l$<TARGET_LINKER_FILE_BASE_NAME:${library}>" | ||
| ) | ||
| add_dependencies(_cargo-build_${target_name} ${library}) | ||
| get_property(libs TARGET ${library} PROPERTY INTERFACE_LINK_LIBRARIES) |
There was a problem hiding this comment.
INTERFACE_LINK_LIBRARIES in CMake is evaluated at build time. This means simply reading this property here would cause us to miss any libraries that are added to INTERFACE_LINK_LIBRARIES after corrosion_link_libraries().
We could work around this by making a deferred function call for TARGETs, which reads the property at a later point in time.
This is just a nit, it's not required for merging this PR.
| ) | ||
| add_dependencies(_cargo-build_${target_name} ${library}) | ||
| get_property(libs TARGET ${library} PROPERTY INTERFACE_LINK_LIBRARIES) | ||
| corrosion_link_libraries(${target_name} ${libs}) |
There was a problem hiding this comment.
Can you add a recursion check here? While ideally there would be no cyclic dependencies, I don't think that assumption holds in practice.
The implementation could probably use something like this (not tested)
set(recursion_list "${recursion_list};${library}")
if( "${library}" IN ${recursion_list})
return
endif()
# do the things
# the pop happens implicitly, due to scoping of variable names in functions.
As mentioned in #505,