from the commit(s) from issue 633. The diff parser assumed a properly
formatted diff that denotes empty context lines with a single space in
the first column. This single space however was missing, because the
hg and git backends got the diff through PHP's exec() function and
this returns already line-splitted output, but - and this is the actual
problem - removes trailing whitespace at the end of each line, essentially
making " \n" only "\n". When splitting this string now again with
PREG_SPLIT_NO_EMPTY the empty line was completely lost in the diff output.
To make it clear that an empty line does not mark a context line now, but
should stop the diff parsing, the Diff parser now also defaults to 'false'
as line type.
This commit fixes issue 688.
The diff parser code was rewritten for clarity and speed and now handles
a couple of ugly cornercases, like SVN's property change output and single
change chunks, much better. Since the path parsing was unified as well,
the SCM interface gained a new method `getPathStripLevel()` which determines
how many path components need to be shoven off a file name for the SCM
to form a valid path in the workspace (similar to patch(1)'s --strip option).
Fixes issue 627. Automated tests follow.
* Scm.php: new SCM method "getChanges" which returns all available
change information grouped by type
* Monotone.php: implement getChanges via get_revision
* <other scms>: rename "changes" member for getCommit to "diff" which
matches better
* Source.php: query the commit's changes and set them in the template
* commit.html: render the changes, type-by-type. Link to the tree or
the individual diff if applicable
* styles.css: some initial style sheet work
Instead of returning a command which gets executed and which should
pass through / stream its output data to the client, we're just
returning an instance of Pluf_HTTP_Response. This is needed, because
some SCMs, most noticable monotone, have no locally executable command
to provide a snapshot archive (and probably never will for our kind
of setup).
We therefor added a little BSD-licensed class "ZipArchive" which allows
the creation of pkzip-compatible archives on the fly by letting it eat
the file contents directly feed from the (remote) stdio instance.
Download performance is ok and lies between 15K/s and 110K/s, but at
least we do no longer block the browser while we pre-generate the zip
file server-side.
Thanks to Patrick Georgi for all his work!
indentifiers in IDF - the SCM function isValidRevision has been replaced
by a validateRevision() method which returns one of three states,
valid, invalid or ambiguous.
The source view can then act accordingly and display disambiguate view
for the latter, so the user can select for which revision he actually
wants to execute the requested action. Also, invalid revisions now lead
to another separate view, telling the user that it is invalid / does
not exist and pointing him optionally to the help page where he can read
further how to access his repository to push the first changes into.
(partially resolves issue 525)
We now have a limited support of the code review. Still some work to be
done to allow the submission of new patches on a given review and update
the status. For the moment, only pre-commit review is supported.
This is because git will return an error code when testing if a commit
exists and in fact does not exist. So it means that we get an exception
on all the string "commit foo" when foo is not a commit in the issue
text.
