[pull] master from scrapinghub:master

.github/workflows/cifuzz.yml

@@ -28,7 +28,7 @@ jobs:
         path: ./out/artifacts
     - name: Upload Sarif
       if: always() && steps.build.outcome == 'success'
-      uses: github/codeql-action/upload-sarif@v2
+      uses: github/codeql-action/upload-sarif@v4
       with:
         # Path to SARIF file relative to the root of the repository
         sarif_file: cifuzz-sarif/results.sarif

.github/workflows/main.yml

@@ -6,6 +6,12 @@ on:
   pull_request:
     branches:
       - master
+  # Called by the Nightly workflow, which re-runs these same legs daily. The
+  # schedule lives there and not here because GitHub disables a workflow that
+  # has a schedule after 60 days without repository activity, and a disabled
+  # workflow answers none of its triggers, which would take pull request runs
+  # down with it.
+  workflow_call:
 jobs:
   build:
     name: Build
@@ -20,6 +26,7 @@ jobs:
             toxenv: min
           - python-version: "3.10"
             toxenv: min-all
+          - python-version: "3.10"
           - python-version: "3.11"
           - python-version: "3.12"
           - python-version: "3.14"
@@ -53,6 +60,10 @@ jobs:
           TZ=${{ matrix.timezone || 'UTC' }} echo "Running tests with timezone: $TZ"
           TZ=${{ matrix.timezone || 'UTC' }} tox -e ${{ matrix.toxenv || 'py' }}
       - name: Upload coverage.xml to codecov
+        # A nightly run, scheduled or dispatched, re-tests a commit that already
+        # has a coverage report, so uploading again only repeats the Codecov
+        # notification for that SHA.
+        if: github.event_name == 'push' || github.event_name == 'pull_request'
         uses: codecov/codecov-action@v5
         with:
           token: ${{ secrets.CODECOV_TOKEN }}

.github/workflows/nightly.yml

@@ -0,0 +1,79 @@
+name: Nightly
+on:
+  # Dependency, time zone database and calendar drift arrives with no commit
+  # behind it, and tests that only fail on the 31st of a month need a run on
+  # every date; a fixed weekday lands on a 31st about once a year. Off the hour,
+  # which GitHub names as a high-load window, and early in the UTC day so that
+  # the Pacific/Auckland and Pacific/Fiji legs stay on the same calendar date as
+  # the rest of the matrix.
+  schedule:
+    - cron: "17 4 * * *"
+  # Runs the same legs on demand, which is also the only way to exercise the
+  # reporting below before it is needed. GitHub disables this workflow after 60
+  # days without repository activity; re-enable it from the Actions tab or with
+  # `gh workflow enable nightly.yml`, since a disabled workflow answers no
+  # triggers at all.
+  workflow_dispatch:
+
+# The called matrix only checks out and tests; the report job below asks for
+# the one extra scope it needs.
+permissions:
+  contents: read
+
+jobs:
+  build:
+    uses: ./.github/workflows/main.yml
+    secrets: inherit
+
+  report:
+    name: Report failure
+    needs: build
+    # A push or pull request failure is already in front of whoever caused it.
+    # A failure with no commit behind it has nowhere to land but a single
+    # notification email to whoever last edited the schedule. A cancelled build
+    # counts, since a leg that hangs until GitHub kills it is a failure too.
+    # Master only, so that a dispatch on a work branch can neither file a
+    # report nor mask the real one.
+    if: >-
+      (needs.build.result == 'failure' || needs.build.result == 'cancelled')
+      && github.ref == 'refs/heads/master'
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+    steps:
+      - name: File a tracking issue unless one is already open
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+          TITLE: The nightly build is failing
+          LABEL: nightly-failure
+          RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+          SHA: ${{ github.sha }}
+        run: |
+          set -euo pipefail
+          # One issue per outage: a daily comment stream would only be one more
+          # signal to ignore. The label, not the title, identifies the tracking
+          # issue, so retitling it to record the cause does not defeat the
+          # check; --force makes creating the label idempotent. A transient
+          # listing failure must not stop the report, since a duplicate issue
+          # is cheaper than silence.
+          gh label create "$LABEL" --force --color d93f0b \
+            --description "Filed by the Nightly workflow"
+          existing="$(gh issue list --state open --label "$LABEL" --limit 1 \
+            --json number --jq '.[0].number // empty' || true)"
+          if [ -n "$existing" ]; then
+            echo "Already tracked in #$existing."
+            exit 0
+          fi
+          gh issue create --title "$TITLE" --label "$LABEL" \
+            --label "Type: Maintenance" --body-file - <<EOF
+          The nightly run of the Build workflow failed on master at $SHA:
+          $RUN_URL
+
+          It tested a commit that was already on master, so this is most likely
+          drift in an unpinned dependency, drift in the CI toolchain, or a test
+          that only fails on this date.
+
+          Filed once per outage, with no follow-up comments. Close it once a
+          nightly run is green again.
+          EOF

CONTRIBUTING.rst

@@ -91,6 +91,10 @@ installed, this is how you set up your fork for local development::
 
    Bare ``pytest`` runs the ``tests/`` directory. Optional calendar and language
    detection tests require ``pip install -e '.[calendars,langdetect]'``.
+   The suite turns warnings into errors (``filterwarnings`` in ``pytest.ini``),
+   so a new deprecation warning fails the run: use ``pytest -W default`` to see
+   such a warning without failing, fix it when it comes from dateparser itself,
+   and otherwise add a narrowly scoped ``ignore:`` line to ``pytest.ini``.
    The ruff formatter uses its default line length of 88; the linter does not
    enforce a maximum line length.
 

pyproject.toml

@@ -80,6 +80,11 @@ serialize = ["{major}.{minor}"]
 [tool.ruff]
 exclude = ["date_translation_data"]
 
+[tool.ruff.lint]
+# An invalid string escape warns at compile time, which pip's byte-compilation
+# absorbs before the test suite's warnings-as-errors setting can see it.
+extend-select = ["W605"]
+
 [tool.ruff.lint.per-file-ignores]
 "dateparser/data/__init__.py" = ["F401"]
 "dateparser/languages/__init__.py" = ["F401"]

pytest.ini

@@ -1,2 +1,16 @@
 [pytest]
 testpaths = tests
+# Warnings are errors: warning-class problems have repeatedly reached users
+# instead of CI (#1246, #1013, #506). The suite emits no warnings at all today.
+# When an upstream deprecation turns the suite red, fix it if it comes from
+# dateparser, and otherwise add the narrowest possible ignore *after* `error`,
+# since the last matching filter wins.
+#
+# The exemptions below are environmental, not code: tzlocal warns when the host
+# has no usable time zone configuration or a stale one, which CI never sees
+# because it always sets TZ, and pytest warns when it cannot write its cache.
+filterwarnings =
+    error
+    ignore:Can not find any timezone configuration:UserWarning
+    ignore:Timezone offset does not match system offset:UserWarning
+    ignore::pytest.PytestCacheWarning

tox.ini

@@ -29,6 +29,14 @@ deps =
     pytz==2024.2
     regex==2024.9.11
     tzlocal==0.2
+# tzlocal 0.2 contains invalid string escapes, which pytest.ini turns into an
+# error. Installers that byte-compile, pip included, absorb that warning before
+# pytest starts, but the ones that do not turn it into an import-time
+# SyntaxError, so ignore that one message for the environments pinned to it. No
+# category: it is a DeprecationWarning up to Python 3.11 and a SyntaxWarning
+# from 3.12 on.
+set_env =
+    PYTEST_ADDOPTS = -W "ignore:invalid escape sequence"
 
 [testenv:min-all]
 basepython = {[testenv:min]basepython}
@@ -38,6 +46,7 @@ deps =
     convertdate==2.2.1
     hijridate==2.3.0
     langdetect==1.0.0
+set_env = {[testenv:min]set_env}
 
 [testenv:scripts]
 deps =
@@ -87,5 +96,8 @@ passenv =
 # instruction counts CodSpeed compares.
 set_env =
     PYTHONHASHSEED = 0
+# pytest-codspeed reports a measurement hook that failed to start or stop as a
+# RuntimeWarning; pytest.ini would turn that runner glitch into a red CodSpeed
+# check that says nothing about dateparser.
 commands =
-    pytest tests/benchmarks/ --no-cov --codspeed
+    pytest -W "ignore::RuntimeWarning:pytest_codspeed" tests/benchmarks/ --no-cov --codspeed