This used to generate an extra entry for @EditParagraphs which in turn
prevented the page from ending with an edit link if it ended with
multiple newlines.
Also made $EditParagraphPencil settable in the config file.
On my website, I noticed that searching for matching pages with a
string containing a non-ascii character worked, where as doing a
normal search resulted in an encoding error. The difference was this:
# fix match parameter
my $match = GetParam('match', '');
SetParam('match', FreeToNormal($match)) if $match;
This may be necessary to remove underscores, but it should not be
necessary to fix encoding errors.
When calling clean_pages on OSX, the mac.pl module is installed via
add_module, which in turn calls mkdir. On other systems, however,
installing the old namespace module will not work since the module
directory will not exist. This change ensures that mkdir is called
before symlinking.
Changed the tests to no longer assume the default value for
$CookieName. This is important when running the tests on a system
where Apache starts a number of wikis for various virtual hosts and
one of them will end up as the host to use for http://localhost/. That
wiki might very well have a cookie name set.
It turns out that one of the tests in lock.t relied on the exact wording
of the error message. I decided to use the same wording for all
instances: "Editing not allowed: %s is read-only" and no longer using
"Editing not allowed for %s".
Closes bug #41625 "ban-quick-editors.pl can hide the error message".
DoPost now uses the same code as DoEdit and thus it will display the
error message, which solves the issue for ban-quick-editors.pl. Added a
test to demonstrate it.
I find that removing " space" and "_underscore" made it easier for me to
visually compare the XPath expressions. I also wanted to keep the
[[foo bar|text]] test (and keep the XPath expression as similar as possible).
The problem was that the new namespaces.pl wants to call
AddModuleDescription which doesn't exist in oddmuse-2.2.6.pl -- but
that's the Oddmuse version that is used for the upgrade test.
When using days for RecentChanges, the More... links after the first
one reverted to $RcDefault instead of keeping the same time period.
This is now fixed.
Tests were added. As I wrote the tests, I realized that GitRun was
printing to STDOUT even though we had created a local *STDOUT and opened
it to write to a memory file (like PageHtml does). Apparently this is
not inherited to the child process. I now use the solution provided for
in the perldoc for open. Unfortunately it requires a temp file. Trying
to use a memory file results in an error. Temporary files are a mess if
I want to access the output: I need to close it and reopen it for
reading. That's why GitRun only does that if $GitDebug is set. This is
what most of the tests will use: set $GitDebug and examine $GitResult.
At the same time I realized that the error message was never being
triggered. The previous code was tricky because it only considered
showing an error message in a non-void context, ie. when called from
maintenance. I wasn't sure this was an actual benefit and decided to
remove it altogether. I we really want to, we can set $GitDebug and
examine $GitResult in GitCleanup.
Move the writing of the config file (and the resetting of the important
variables) into a separate file so that tests can reset the config file
without clearing all the pages.
Using stat to get the last edit timestamp can lead to a failing test.
To be absolutely certain, use the timestamp stored in the page file.
As it stands, $Now is used for the timestamp in the page file but if
this is a new file (revision 1), then the index file gets rewritten
and its timestamp changes.
