CI: Adding and testing Big-Endian support for quaddtype #152
Conversation
|
I am guessing there might be more failures, eg: str and repr methods of dtype that uses dragon4, we can extensively test them later or maybe in this PR, whatever sounds good |
|
Self note: But the main purpose of this PR is CI and that is working |
|
@ngoldbaum this is ready now, let me know your review |
SwayamInSync
changed the title
|
Thanks @SwayamInSync for your work on this! |
| #if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) | ||
| npy_uint64 hi; | ||
| npy_uint64 lo; | ||
| #else |
There was a problem hiding this comment.
Nice! glad you chased this down. While BigEndian machines are rare these days it's nice to have fully correct code and it's important for it to be reflected in the source code that this representation is endian-dependent.
| path: ~/docker_${{ matrix.BUILD_PROP[1] }} | ||
| key: container-quaddtype-${{ runner.os }}-${{ matrix.BUILD_PROP[1] }}-${{ matrix.BUILD_PROP[2] }}-${{ hashFiles('quaddtype/pyproject.toml') }} | ||
|
|
||
| - name: Create cross-compilation container |
There was a problem hiding this comment.
This step takes 10 minutes. It might be possible to add some caching to make that quicker. Just a note for a future improvement.
|
Thanks @SwayamInSync! |
closes #119
I took inspiration from NumPy's workflow for this https://github.com/numpy/numpy/blob/main/.github/workflows/linux_qemu.yml
Since little-endian are already available in our current CI so here only big-endian testing.
Things seem to work and one test case is failing. Making the PR because CI is seems to work as expected