Conversation
Instead of using add_custom_command for a TARGET which does not specify the required PRE_BUILD/PRE_LINK/POST_LINK argument, use add_custom_command with OUTPUT to perform the copy, record files coped and create a custom target that depend on those. All Cython files then depend on that target.
Ninja parallalizes the build and hence works faster to build the project, it also does not have the bug that make has cause it to unconditionally recompile and relink extension libraries saving significant iterative development time.
|
View rendered docs @ https://intelpython.github.io/dpctl/pulls/766/index.html |
-- Add an option to specify custom gtest installation.
-- FindLLVMCov offers a way to specify where to look
for llvm-cov. If `bin-llvm` is provided, set the
`LLVM_TOOL_DIR` envar as well to help FindLLVMCov.
Otherwise, just adding to PATH is not enough.
-- Some formatting fixes to coverage workflow.
|
@oleksandr-pavlyk I was looking at why the coverage generation workflow is failing. I have not identified it, but made some clean ups to the |
|
Following is the output from |
|
The issue was that |
|
Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞 |
Cleaned up cmake scripts, and transitioned from using Unix makefiles generator on Linux, and Ninja on Windows to using Ninja for both platforms.
This PR accomplishes that incremental rebuilds is very fast if no native source files changed.
Generator is changed in all
scripts/drivers, inconda-recipeand in all workflows.