row's height to fit the up-popping horizontal scrollbar, but all rows
just a little, so the heights did not match. I've reworked this to not
used the ill-advised rowspan any longer, but two separate tables whose
heights match each other now in both browsers.
Also I fixed a bug in the whitespace detection code - utf8 characters
where broken into single bytes, so apparently the [:print:] character
class does not accout for them, even in //u mode, so we're selecting
the characters that we want to make visible on our own (basically
control characters lower than space, I might add more).
into a separate container that can overflow and side-scroll for long lines.
This effectively removes the need for all kinds of line-breaking hacks
that have been applied before and only worked when the browser was
actually able to break a word group apart somewhere.
Lines are now always rendered as-is; as a nice side effect the line numbers
are always visible, independently how far one scrolled into one direction,
so the context is always clear. If the rendering area is made smaller, the
table rendering also degrades gracefully and provides horizontal scrolling
for views that did not need them before.
The size that is occupied by the number display is now also automatically
determined by the size that is needed to render the biggest line number
in a column. Empty columns are rendered with a zero size.
Currently all this works nicely with a recent version of Chrome, Firefox
still needs some fine tuning for the vertical positioning. Other browsers
are untested as of now.
Line endings are now preserved during the diff parsing. When the diff
is then rendered later on, we replace non-printable characters by their
ordinal counterparts, so a user can easily grasp changes when a hunk
comes with edits that might be invisible at first.
The expected format for the diff test suite has changed from serialized
PHP to var_export, which is easier readable, editable and understandable,
while still keeping parsable as well.
Support for old Macintosh line endings could not be added, mainly because
modern SCMs do not support single \r in their unified diff output either
and working around and parsing these "lines" would have been a major
headache with not much outcome (given the fact that all Macs that have been
sold since 2001 or 2002 have been BSD-based and as such used Unix line
endings by default).
This commit fixes issue 636.
It's allow to display this view for other members.
In the issue summary, we can now follow make links for each user display in the part "Unresolved: By Assignee".
Since IDF's text search component does not allow further restrictions
on the result set, we make a second, filtered query to restrict to
the item state ('open' or 'closed') and optionally a label. All in
all this is all harder than it could be, especially the tag cloud
is very monolithic and should be replaced by a data-driven component
that is less dependent on a single data / query and link usage, but
this would for now require too many changes.
Similar questionable is the code duplication for the index, listStatus
and listLabel view implementations that all do more or less the same.
The search implementation now only uses one implementation for a very
similar use case. It also removes the artificial restriction to 100
results we had previously there and does not query a record for each
single result (as was done with Pluf_Search_ResultSet previously).
On my way through this I tried to generalize a couple of i18n texts
and removed smaller issues like the "trailing comma" in label lists.
This partially fixes issue 548.
- Rename the test repo to match the test function so we can create
more test repositories for other tests at a later stage.
- Remove useless hooks and configs from the repo (they are not used
for our specific test and just need memory).
- Note the fix for issue 721 in NEWS.mdtext.
The SVN backend failed when trying to access historical information on deleted files.
There's also an initial test case for the SVN backend, testing this issue
and issue 364, which is about a similar problem for renamed files.
Reverting any of these fixes breaks the test.
Content-MD5 is a HTTP header to provide end-to-end integrity checks
(see RFC2616, 14.15). This doesn't protect against malicious
modifications, but against transmissions errors and storage errors
on the server.
The change also removes one redirect when downloading files.
