Adopt native-Python code generation convention #378
Conversation
kostmo
force-pushed
the
adopt-native-codegen
branch
4 times, most recently
from
220c15e to
f559241
Compare
| # XXX This logic is suspect... | ||
| upload_job_filter_branch = not is_py3_linux and filter_branch |
There was a problem hiding this comment.
Not sure if this was a bug introduced at some point in the in the existing logic. Currently, the filters: yaml chunk is excluded from *_upload jobs on Linux with Python 3, so I've crafted my logic to match this behavior. Not sure what the rationale is, though, or whether that's intentional.
There was a problem hiding this comment.
Is this code generation (with exception) the same across domains? What are the differences?
There was a problem hiding this comment.
There are some minor differences with the vision repo's config.yml file, and by extension the codegen. To name a few:
visionexercises an extra dimension of configuration with different versions of CUDA + cpu- it does not have
smoke_testjobs - it lacks
filters:sections in the YAML
.circleci/sort-yaml.py
Outdated
| @@ -0,0 +1,13 @@ | |||
| #!/usr/bin/env python3 | |||
|
|
|||
There was a problem hiding this comment.
I have added this little script to prove that my change yields identical output, modulo ordering of YAML keys, whitespace, and quoting style (single- vs. double-quotes).
There was a problem hiding this comment.
So this is a test file, right? How about adding it as .circleci/test/test_sort_yaml.py ?
There was a problem hiding this comment.
Will do, though it shouldn't be considered a standalone test, but rather a utility to facilitate a manual comparison for this somewhat extensive code change.
.circleci/sort-yaml.py
Outdated
| @@ -0,0 +1,13 @@ | |||
| #!/usr/bin/env python3 | |||
|
|
|||
There was a problem hiding this comment.
So this is a test file, right? How about adding it as .circleci/test/test_sort_yaml.py ?
| # XXX This logic is suspect... | ||
| upload_job_filter_branch = not is_py3_linux and filter_branch |
There was a problem hiding this comment.
Is this code generation (with exception) the same across domains? What are the differences?
kostmo
force-pushed
the
adopt-native-codegen
branch
from
f559241 to
508522f
Compare
Closes pytorch#304 See rationale writeup: pytorch/vision#1321 (comment)
kostmo
force-pushed
the
adopt-native-codegen
branch
from
508522f to
0ef4742
Compare
Compare to application to vision repo in pytorch/vision#1321
See rationale writeup: pytorch/vision#1321 (comment)
Fixes #304