- Improved error reporting from LTD Keeper API responses for common scenarios and direct the user to contact dm-docs-support on the LSSTC Slack.
- Removed usage of
pkg_resources
for better compatibility with Python 3.12 environments. - Fixed internal typing issues.
- Updated the GitHub Actions workflows.
- Fix parsing of the
GITHUB_HEAD_REF
environment variable in GitHub Actions.
- The
ltd upload
command now works in pull requests workflows on GitHub Actions.
-
This release focuses on infrastructure improvements, and should not have a significant impact on users.
-
This tool is officially tested and supported on Python 3.7 and 3.8. Python 3.6 is no longer officially supported.
-
Removed pins on third-party dependencies, including boto3, requests, uritemplate, and click. In doing so, LTD Conveyor is easier to integrate into your existing Python environment.
-
We have migrated from Travis CI to GitHub Actions for testing, linting, documentation builds and deployment, and deployments to PyPI. In addition to on-demand testing, GitHub Actions tests the package on a daily basis to ensure compatibility with third-party dependencies.
-
Internally, we use tox for running tests, linters, and to build documentation. tox makes it possible for local test runs to use the same set up as CI.
-
The codebase is now automatically formatted with black and isort. By using pre-commit, we ensure that formatting is always applied.
-
The codebase now has type annotations, which are tested with mypy. Type annotations help ensure that APIs are explicitly defined and correctly used.
- The
--gh
option forltd upload
now supports tag and PR events, in addition to branches.
- Added a
--gh
option to theltd upload
command to support usage in GitHub Actions.
- LTD Conveyor can now upload new builds to S3 using Amazon's S3's presigned POST URL feature. This means that clients no longer need credentials for S3 — the LTD Keeper API server generates presigned POST URLs as part of the
POST /products/<product>/builds/
build registration step. - The
ltd
command now uses the presigned POST-based URLs. The--aws-id
and--aws-secret
options have been removed, but the--user
and--password
options remain. If you use environment variables, there should be no change in how you useltd upload
(aside from not needing the$LTD_AWS_ID
and$LTD_AWS_SECRET
environment variables).
- In INFO-level and higher logging, the module path isn't displayed. Module paths are only shown in DEBUG-level logging where it's most useful.
- Packaging is now done through a
setup.cfg
file and apyproject.toml
file (PEP 518). - Updated Travis CI configuration to test with Python 3.8; also refactored the CI Pipeline with Travis CI stages.
- Fixed a bug where
ltdconveyor.s3.delete_dir
, since 0.4.1, would raise aTypeError
while deleting an empty directory (no objects in the S3 prefix).
- Fixed a bug where
ltdconveyor.s3.delete_dir
would fail if there are more than 1000 objects under a path prefix that is being deleted. - Fixed title of the project in the documentation by updating to use a Documenteer-based Sphinx set up.
-
Added a new command-line app,
ltd
, that provides subcommands for clients to work with the LSST the Docs API. This app is implemented with Click, and its documentation is automatically generated from the command-line help withsphinx-click
.The first subcommand is
ltd upload
, which lets a client upload a site to LTD as a new build. This command includes special features for using the client from Travis CI (ltd upload --travis
option) to populate the version information from the Travis environment. Other flags allowltd upload
to become a no-op under certain circumstances (for example, skip uploads on pull requests) or arbitrarily (setltd upload --skip
orexport LTD_SKIP_UPLOAD=true
).With this feature, LTD Conveyor effectively deprecates LTD Mason. LTD Mason was the original uploader for LSST the Docs, but it was also designed around the idea of building the
LSST Science Pipelines
documentation as well (hence the term "mason"). Over time, we realized it is better to have a general purpose client and uploader for LSST the Docs (LTD Conveyor) and a dedicated tool for assembling the multi-package Sphinx documentation site (Documenteer). -
Added the
ltdconveyor.keeper
subpackage that widens the scope of LTD Conveyor to be a full-service library for building LTD clients, not just an S3 upload client.-
The
ltdconveyor.keeper.login.get_keeper_token
function lets you obtain a temporary auth token for the LTD Keeper API. -
The
ltdconveyor.keeper.build
module includes functions for performing the build upload handshake with the LTD Keeper API.
-
-
Removed Python 2.7 and 3.4 compatibility. Now the codebase is entirely Python 3 (3.5, 3.6)-oriented.
-
Completely reorganized Python namespace. Now all S3 APIs are in
ltdconveyor.s3
. Fastly APIs are available fromltdconveyor.fastly
. -
Switched to using
setuptools_scm
to generate version strings for releases. -
Switched to using
extras_require
for development dependencies (pip install -e ".[dev]"
). This lets us exclusively coordinate dependencies insetup.py
. -
Enable testing via
python setup.py test
. Also upgrade the testing stack topytest
3.5 andpytest-flake8
1.0. -
Compatibility updates to the Sphinx documentation infrastructure.
-
Default to
acl=None
to support more AWS IAM users. It turns out that not all IAM users withPutObject
permissions also have permissions to set theACL
for an object. We want to make it possible for many lightweight IAM users to upload to restricted sub-directories of thelsst-the-docs
bucket, but it seems hard to make these users ACL grantees too. We now seek an alternative:- The
lsst-the-docs
bucket now has a PublicRead bucket policy - No ACL is set on individual objects.
- The
- Changed assertions to
RuntimeErrors
. Assertions shouldn't be used to raise exceptions in production code.
- Add
open_bucket
function. This provides a convenient API for getting a boto3 bucket resource, particularly for clients that use the upload_object and upload_file APIs directly that take a bucket only.
-
Added
content_type
parameter toupload_object
so that a user can specifytext/html
for an HTML upload. Theupload_file
/upload_dir
functions avoid this problem by using Python'smimetypes
library to guess the encoding, butupload_object
is lower-level and warrants having the user explicitly provide the content type. -
Refactored new function
create_dir_redirect_object
. This code used to be insideupload_dir
, but in LTD Dasher it seems necessary to upload files one at a time, and thus it's necessary to directly create these directory redirect objects.
- Fix issue where
'..'
shows up in directory names (seen when doingupload_dir
withLTD Dasher
asset directories).
- Ported Fastly
purge_key
function from LTD Keeper. Any LTD application that uploads objects to S3 might also need to purge Fastly CDN caches. This purge capability is presented as a simple function.
- Port S3 codebase from LTD Mason. The purpose of LTD Conveyor is to provide a set of common S3 APIs that can be used from both client (uploading docs from CI) and server (LTD Keeper) settings.