## What changes were proposed in this pull request?
This PR updates dev/create-release/known_translations to add more contributor name mapping. It also fixes a small issue in translate-contributors.py
## How was this patch tested?
manually tested
Author: Yin Huai <yhuai@databricks.com>
Closes#16423 from yhuai/contributors.
This bypasses unnecessary calls to the Github and JIRA API.
Additionally, having a local cache allows us to remember names
that we had to manually discover ourselves.
This commit introduces several major improvements to the script
that generates the contributors list for release notes, notably:
(1) Use release tags instead of a range of commits. Across branches,
commits are not actually strictly two-dimensional, and so it is not
sufficient to specify a start hash and an end hash. Otherwise, we
end up counting commits that were already merged in an older branch.
(2) Match PR numbers in addition to commit hashes. This is related
to the first point in that if a PR is already merged in an older
minor release tag, it should be filtered out here. This requires us
to do some intelligent regex parsing on the commit description in
addition to just relying on the GitHub API.
(3) Relax author validity check. The old code fails on a name that
has many middle names, for instance. The test was just too strict.
(4) Use GitHub authentication. This allows us to make far more
requests through the GitHub API than before (5000 as opposed to 60
per hour).
(5) Translate from Github username, not commit author name. This is
important because the commit author name is not always configured
correctly by the user. For instance, the username "falaki" used to
resolve to just "Hossein", which was treated as a github username
and translated to something else that is completely arbitrary.
(6) Add an option to use the untranslated name. If there is not
a satisfactory candidate to replace the untranslated name with,
at least allow the user to not translate it.
This commit involves three main changes:
(1) It separates the translation of contributor names from the
generation of the contributors list. This is largely motivated
by the Github API limit; even if we exceed this limit, we should
at least be able to proceed manually as before. This is why the
translation logic is abstracted into its own script
translate-contributors.py.
(2) When we look for candidate replacements for invalid author
names, we should look for the assignees of the associated JIRAs
too. As a result, the intermediate file must keep track of these.
(3) This provides an interactive mode with which the user can
sit at the terminal and manually pick the candidate replacement
that he/she thinks makes the most sense. As before, there is a
non-interactive mode that picks the first candidate that the
script considers "valid."
TODO: We should have a known_contributors file that stores
known mappings so we don't have to go through all of this
translation every time. This is also valuable because some
contributors simply cannot be automatically translated.
