Repository navigation
Fix: use _BUILD_TIME_PYTHON when cross-compiling - #187
Conversation
686e3e9 to
a2a1f4c
Compare
|
Can you quickly explain the rationale for this? |
|
yep. at the moment all cross-compiles include a Though, open question you may be able to answer that I don't know: is this cross-compile actually used in CI right now? probably worth fixing regardless |
|
We don't cross-compile anymore, and it's quite likely that cross-compilation is broken by now. Maybe we should remove deprecated cross-compilation code instead? |
EzraBrooks
changed the title
|
I can put together an alternate PR that removes cross-compile piping entirely to patch this bug by removing it 😀 |
|
Though I will say that I'm deeply commercially interested in RISC-V and right now it's a lot easier to cross compile to RISCV64 than natively build in the cloud |
|
Good point about RISC-V. I think @wolfv and the prefix folks are interested in this, too - in this case, it makes sense to keep cross-compiling code (which can be fixed if it doesn't work properly anymore). In this case, I'm supportive of the PR, but will wait for @traversaro to take another look. |
|
Without #186 this got a bit more complicated. I don’t like this multi-step re-pinning process that the agent added.. going to noodle a bit more |
…n included With conditional_python_dependency, recipes without Python content get a build python pinned to python_min, so they aren't built per Python. The cross-compilation block's bare python escaped that pin and put them back into the Python variant matrix when cross-compiling. Packages built per Python, on the other hand, got the pin too: cross-compiling, their build environment also gets the variant's python (for cross-python), so a cp313 build asked for python 3.12.* and 3.13 at once. Pin both build pythons in one place, after dependencies.yaml (which can add python to a package without Python dependencies, e.g. one that embeds the interpreter), and only when host ends up without python.
ae7e910 to
d2ffd7a
Compare
|
Okay, this now does the pinning rewrite after the dependencies.yaml is merged in, fixing the plotjuggler case. |
|
I realized overnight that this isn't actually blocking my RoboStack PR and it was just blocking my testing of it on my big beefy build workstation, where I was doing cross builds in testing. Still worth merging though imo |
|
Could you please also take a look @traversaro |
No description provided.