Closed jcapona closed 1 year ago
Base: 45.90% // Head: 60.79% // Increases project coverage by +14.89%
:tada:
Coverage data is based on head (
d5e1755
) compared to base (e291cd0
). Patch coverage: 78.38% of modified lines in pull request are covered.
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
Yeah, on the deb side, we're pretty much stuck with numpy 1.19.5
and opencv 4.5.1
, and both of them work together okay.
If you pip install the same versions, sadly, they're not compatible with each other (module compiled against API version 0xf but this version of numpy is 0xd
). My guess is that the version of opencv
distributed via pypi was compiled using a different version of numpy
compared to the one distributed via apt
.
The main issue we need to solve here is to make sure that the version of numpy
is >1.19.5
, at least when pip-installing. The limitation on the version of opencv
is not really necessary, since the latest version available in pypi, 4.6.0
, is still compatible with the version of numpy
that gets installed via pip, 1.23.0
.
Main changes
When pip installing a subpackage in Raspberry Pi OS, we used the already installed
numpy
package, version1.19.5
, which was the version we wanted to install. However, this error came up when trying to importcv2
:After reading this useful explanation, it seems that we need to use a newer version of
numpy
.Note: This is not an issue when installing the subpackages using the
deb
files.Manual testing tips