tensorrt sdist install fails

[ddelange]

Hi 👋

tensorrt sdist install runs pip install inside pip install, which is undefined/unsupported behaviour from the PyPA side (just like running multiple concurrent pip commands).

On my local machine (and probably on many other machines), it happens to work, but it is not robust and does not comply with the PyPA index url mechanic design.

The proper way to employ external index urls (see e.g. pytorch), is by having the user set the --extra-index-url flag on the top level installation command, or exporting PIP_EXTRA_INDEX_URL before running top level installation command.

Here is an example where this hack fails in the wild:

• tensorrt install errors with /home/ci/opt/venv/bin/python3: No module named pip
  • command ['/home/ci/opt/venv/bin/python3', '-m', 'pip', 'install', 'tensorrt_libs==8.6.1', '--index-url', 'https://pypi.nvidia.com/']
• yet autogluon ci runs python3 -m pip install --upgrade pip
  • which upgrades /home/ci/opt/venv/lib/python3.8/site-packages
• the venv in question is set up and activated properly, as can be seen from the environment variables in the logs linked: VIRTUAL_ENV env var is set and VIRTUAL_ENV/bin is prepended to the PATH env var.

full error log for future reference: logs.txt

Originally posted by @ddelange in #2933 (comment)

[ddelange]

Hi @zeroepoch 👋

I've opened #3080 to switch to the best practice implementation (similar to how torchvision 0.15.2 pins torch==2.0.1 and allows the end user to pick the cuda/cpu flavour using --extra-index-url).

How does that look to you?

[zerollzeng]

@pranavm-nvidia ^ ^

[ddelange]

closed in #3080