ci: Don't remove .coverage file & set empty suffix.

Fixes #14962
* codecov needs `.coverage` file in the pwd to upload coverage
results.
* concurrency='multiprocessing' forces `.coverage` file to have a
data_suffix, we explicity set it to "" for the file to have no suffix.
Filed issue as https://github.com/nedbat/coveragepy/issues/989

.circleci/config.yml

@@ -95,11 +95,20 @@ aliases:
     run:
      name: upload coverage report
      command: |
+       # codecov requires `.coverage` file to be stored in pwd for
+       # uploading coverage results.
+       mv /home/circleci/zulip/var/.coverage /home/circleci/zulip/.coverage
+
        . /srv/zulip-py3-venv/bin/activate
-       # codecov version is fixed here, since future versions of it
-       # use "find" for locating files which is buggy on some platforms.
-       # See https://github.com/codecov/codecov-python/issues/250
-       pip install codecov==2.0.15 && codecov \
+       # TODO: Check that the next release of codecov doesn't
+       # throw find error.
+       # codecov==2.0.16 introduced a bug which uses "find"
+       # for locating files which is buggy on some platforms.
+       # It was fixed via https://github.com/codecov/codecov-python/pull/217
+       # and should get automatically fixed here once it's released.
+       # We cannot pin the version here because we need the latest version for uploading files.
+       # see https://community.codecov.io/t/http-400-while-uploading-to-s3-with-python-codecov-from-travis/1428/7
+       pip install codecov && codecov \
          || echo "Error in uploading coverage reports to codecov.io."
 
   - &build_production

tools/ci/backend

@@ -12,7 +12,7 @@ set -x
 # docker setup means the auto-detection logic sees the ~36 processes
 # the Docker host has, not the ~2 processes of resources we're
 # allocated.
-./tools/test-backend --coverage --include-webhooks --parallel=6
+./tools/test-backend --coverage --include-webhooks --no-cov-cleanup --parallel=6
 
 # We run mypy after the backend tests so we get output from the
 # backend tests, which tend to uncover more serious problems, first.

tools/test-backend

@@ -216,6 +216,10 @@ def main() -> None:
     parser.add_argument('--verbose-coverage', dest='verbose_coverage',
                         action="store_true",
                         default=False, help='Enable verbose print of coverage report.')
+    parser.add_argument('--no-cov-cleanup', dest='no_cov_cleanup',
+                        action='store_true',
+                        default=False,
+                        help="Do not clean generated coverage files.")
 
     parser.add_argument('--parallel', dest='processes',
                         type=int,
@@ -354,10 +358,12 @@ def main() -> None:
     assert_provisioning_status_ok(options.force)
 
     if options.coverage:
-        import atexit
         import coverage
-        cov = coverage.Coverage(config_file="tools/coveragerc", concurrency='multiprocessing')
-        atexit.register(lambda: cov.erase())  # Ensure the data file gets cleaned up at the end.
+        cov = coverage.Coverage(data_suffix="", config_file="tools/coveragerc", concurrency='multiprocessing')
+        # Do not clean .coverage file in CircleCi job so that coverage data can be uploaded.
+        if not options.no_cov_cleanup:
+            import atexit
+            atexit.register(lambda: cov.erase())  # Ensure the data file gets cleaned up at the end.
         cov.start()
     if options.profile:
         import cProfile
@@ -407,7 +413,6 @@ def main() -> None:
         cov.stop()
         cov.save()
         cov.combine()
-        cov.data_suffix = False  # Disable suffix so that filename is .coverage
         cov.save()
         if options.verbose_coverage:
             print("Printing coverage data")