Open hmaarrfk opened 3 months ago
It might be something like this patch:
diff --git a/recipe/conda_forge_ci_setup/upload_or_check_non_existence.py b/recipe/conda_forge_ci_setup/upload_or_check_non_existence.py
index bd8c0d2..8188ed6 100644
--- a/recipe/conda_forge_ci_setup/upload_or_check_non_existence.py
+++ b/recipe/conda_forge_ci_setup/upload_or_check_non_existence.py
@@ -105,7 +105,7 @@ def built_distribution_already_exists(cli, name, version, fname, owner):
def upload(token_fn, path, owner, channels, private_upload=False, force_metadata_update=True):
cmd = ['anaconda', '--quiet', '--show-traceback', '-t', token_fn,
'upload', path, '--user={}'.format(owner),
- '--channel={}'.format(channels)]
+ '--channel={}'.format(channels), '--skip-existing']
if private_upload:
cmd.append("--private")
if force_metadata_update:
I wonder if we rely on this to fix partial uploads that didn't finish due to network errors. Regardless, it could be added as a conda-forge.yml option via conda-smithy? We only need to support that optional flag here.
Comment:
I'm trying to build a few different combinations of split packages.
However, some outputs are the same between builds which is causing them to fail upload as shown in the message bolow.
I know that anaconda upload has an option
--skip-existing
would it be acceptable to add it?The alternative is to add the package to every output, which is both redundant, and "unecessary". Example logs:
xref: https://github.com/conda-forge/openvino-feedstock/pull/86
cc: @ilya-lavrenov