This makes the setup script fail on the first command that has an unchecked error, and makes tracing what's happening while debugging easier via the TRACE environment variable.
Also, while testing, the cudatoolkit=12.1 package failed (it looks like only up to 11.8 is on conda-forge and the nvidia channels), so this instead uses the cuda-version meta package to grab version 12.1. This part could be a mixup in my setup - if there's a channel with cudatoolkit=12.1 let me know!
NOTE: For a future pull request, regardless of the answer, it looks like we still install cuda relevant packages even if $INSTALL_CUDA is not specified. Is that intended?
Tested by:
Running ./setup_octopi.sh without conda installed and making sure it fails with a non-zero error code, and prints a message.
Running TRACE=1 ./setup_octopi.sh and making sure the + trace messages are shown
Running ./setup_octopi.sh on an Ubuntu 22.04 system and making sure it runs successfully (after installing conda).
This makes the setup script fail on the first command that has an unchecked error, and makes tracing what's happening while debugging easier via the
TRACEenvironment variable.Also, while testing, the
cudatoolkit=12.1package failed (it looks like only up to 11.8 is on conda-forge and the nvidia channels), so this instead uses thecuda-versionmeta package to grab version 12.1. This part could be a mixup in my setup - if there's a channel with cudatoolkit=12.1 let me know!NOTE: For a future pull request, regardless of the answer, it looks like we still install cuda relevant packages even if
$INSTALL_CUDAis not specified. Is that intended?Tested by:
./setup_octopi.shwithout conda installed and making sure it fails with a non-zero error code, and prints a message.TRACE=1 ./setup_octopi.shand making sure the+trace messages are shown./setup_octopi.shon an Ubuntu 22.04 system and making sure it runs successfully (after installing conda).