You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The way that we fill the CPM cache with nvcomp during testing does not work properly if nvcomp is already installed on the host system. The problem is that since we don't actually use the CPM cache to establish nvcomp source but rather download a binary directly (because official nvcomp releases are binary-only, the project isn't open-source), and to avoid redundant downloads we handle this internally by searching for nvcomp packages directly within rapids_cpm_nvcomp before we download the binary. As a result, if nvcomp is already installed the download doesn't happen when populating the cache. Moreover, when later tests are run that invoke rapids_cpm_nvcomp the same logic leads to the local copy being found first, so even if the cache were populated it would have no effect on what is found. To fix this we would have to use a custom override.json to ensure the download and propagate the associated nvcomp root through to all the tests. This is doable, but it is a lot of effort for handling an edge case (testing in environments where nvcomp is installed) so I wouldn't prioritize fixing this anytime soon.
The text was updated successfully, but these errors were encountered:
The way that we fill the CPM cache with nvcomp during testing does not work properly if nvcomp is already installed on the host system. The problem is that since we don't actually use the CPM cache to establish nvcomp source but rather download a binary directly (because official nvcomp releases are binary-only, the project isn't open-source), and to avoid redundant downloads we handle this internally by searching for nvcomp packages directly within
rapids_cpm_nvcomp
before we download the binary. As a result, if nvcomp is already installed the download doesn't happen when populating the cache. Moreover, when later tests are run that invokerapids_cpm_nvcomp
the same logic leads to the local copy being found first, so even if the cache were populated it would have no effect on what is found. To fix this we would have to use a custom override.json to ensure the download and propagate the associated nvcomp root through to all the tests. This is doable, but it is a lot of effort for handling an edge case (testing in environments where nvcomp is installed) so I wouldn't prioritize fixing this anytime soon.The text was updated successfully, but these errors were encountered: