svn commit: r1905756 - /subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py

[ko...@apache.org]

Author: kotkov
Date: Mon Dec  5 12:35:54 2022
New Revision: 1905756

URL: http://svn.apache.org/viewvc?rev=1905756&view=rev
Log:
On the 'pristines-on-demand-on-mwf' branch: Skip the revert_reexpand_keyword()
test for working copies that don't store local pristines.

The test expects two different things to happen during revert, depending
on whether an `svn cleanup` or any other command that repairs the timestamps
was called in between:

- In the first part, we expect a revert to change file contents and
  re-expand its keywords, because a revert happens right after editing
  the file.

- In the second part, we expect a revert to skip the file with exactly
  the same contents, because there's an in-between operation that has
  recorded the new unmodified timestamp in the db.

This is a problem, because we expect two different things to happen for
two identical on-disk file states.  The only difference is whether any
of the commands that perform the internal timestamp bookkeeping was
called in between.  For example, calling `svn cleanup` after editing the
file makes the first and the second parts of the test behave identically.

This problem currently prevents us from testing the `--store-pristine=no`
case, because when we walk the text-bases we use that opportunity to repair
timestamps, but that also makes the first and the second reverts in this
test behave identically.  So let's postpone testing this case until we
resolve the described problem with opposite expectations.

* subversion/tests/cmdline/revert_tests.py
  (revert_reexpand_keyword): Remove XFail() marker.  Skip for working copies
   that don't store local pristines.  Adjust comments.

Modified:
    subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py

Modified: subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py
URL: http://svn.apache.org/viewvc/subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py?rev=1905756&r1=1905755&r2=1905756&view=diff
==============================================================================
--- subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py (original)
+++ subversion/branches/pristines-on-demand-on-mwf/subversion/tests/cmdline/revert_tests.py Mon Dec  5 12:35:54 2022
@@ -243,15 +243,6 @@ def revert_from_wc_root(sbox):
   svntest.actions.run_and_verify_status('', expected_output)
 
 @Issue(1663)
-# Walking the text-bases automatically repairs timestamps, so now the
-# first and the second reverts in this test behave identically, as if
-# 'svn cleanup' or any other command that repairs the timestamps had been
-# called beforehand.  Judging by the second part of the test, we're fine
-# with revert doing nothing in that case, but that essentially contradicts
-# the expectation in its first part.
-#
-# I temporarily mark the test XFail.  See r1101730 and r1101817 for details.
-@XFail()
 def revert_reexpand_keyword(sbox):
   "revert reexpands manually contracted keyword"
 
@@ -269,6 +260,34 @@ def revert_reexpand_keyword(sbox):
 
   sbox.build()
   wc_dir = sbox.wc_dir
+
+  # The test expects two different things to happen during revert, depending
+  # on whether an `svn cleanup` or any other command that repairs the
+  # timestamps was called in between:
+  #
+  # - In the first part, we expect a revert to change file contents and
+  #   re-expand its keywords, because a revert happens right after editing
+  #   the file.
+  #
+  # - In the second part, we expect a revert to skip the file with exactly
+  #   the same contents, because there's an in-between operation that has
+  #   recorded the new unmodified timestamp in the db.  (See r1101730 and
+  #   r1101817.)
+  #
+  # This is a problem, because we expect two different things to happen for
+  # two identical on-disk file states.  The only difference is whether any
+  # of the commands that perform the internal timestamp bookkeeping was
+  # called in between.  For example, calling `svn cleanup` after editing the
+  # file makes the first and the second parts of the test behave identically.
+  #
+  # This problem prevents us from properly testing the `--store-pristine=no`
+  # case, because when we walk the text-bases we use that opportunity to
+  # repair timestamps, but that also makes the first and the second reverts
+  # in this test behave identically.  Let's skip testing this case for now,
+  # until we resolve the described problem with opposite expectations.
+  if not svntest.actions.get_wc_store_pristine(wc_dir):
+    raise svntest.Skip()
+
   newfile_path = os.path.join(wc_dir, "newfile")
   unexpanded_contents = "This is newfile: $Rev$.\n"