When going to alternate screen buffer on smcup, the terminal is
supposed to store the cursor position, rendition etc. Screen used to
store only the position (and that was just a little buggy anyway).
So change this behaviour to store all information about the cursor,
and restore everything on rmcup.
Apparently, the terminal is supposed to store the cursor for smcup
(CSI ? 1049 h) and DECSC (ESC 7, CSI s, CSI ? 1048 h) separately. So
be it.
Screen wouldn't do anything with 'hardcopy' if there was no
foreground window, except overwriting the specified file (when
hardcopy_append is off, of course). But the NEED_FORE flag wasn't
set. I am not sure why this was the case. So I have added the
NEED_FORE flag so that 'hardcopy' command fails if there is no
foreground window.
Make -h save scrollback with or without a file name given. Also,
preserve the unlikely case in which someone actually does want to
save display to '-h'; with
:hardcopy -- -h
or, to save both history and display to '-h'
:hardcopy -h -h
Also,
:hardcopy --
:hardcopy -h --
will, as they always have, save display or history and display to
'--'.
Currently, '@' or'-' prefixed commands suppress both error messages
and normal messages. It will be possible to differentiate between
the two, though, if there was need for it. These flags can be useful
for, for example, remote commands to not interrupt displays etc.
Expanding $PWD to expand to the correct current working directory
can be useful to figure out, for example, where log files are
created, and other purposes.
Showing the capabilities somewhere is useful for users, especially
when diagnosing a bug. The 'license' page seems as good a place as
any to stuff this info in.
When a layer displayed within a canvas is smaller than the available
area, we show '-' at the bottom, and now '|' on the right to mark
the actual size of the layer in the canvas.
Instead of killing all overlays when resizing, just kill the ones
that cannot be resized, and the resized the rest. This fixes a bug
where the window-list (or a group-window) is aborted when
layer-size changes (because window-size changed, or
caption/hardstatus etc. was toggled). This also makes the code
robust. So if you are looking at a window/display list, and the
window-size is changed, the list won't go away.
Double cell characters that end a line are not properly displayed
when there are split regions. Takeshi Banse detected the problem and
a fix for it in Savannag bug #29106. A better fix for this was
applied in the unicode++ branch, but that won't work in 'master'.
This is approximately the same fix, in a slightly different way.
Some distributions ( mandriva for example ) use
-Werror=format-security by default when build C software to try to
enhance code quality and prevent security bugs. As the policy is to
correct all occurences of the error ( so there is no false positive,
nor question about to fix it or not ), a patch that fix the
error on gnu screen 4.0.3 have been added to the package. I have
rediffed it against latest git snapshot, to be sent upstream.
It mainly add "%s" where applicable.
When there are no splits, pausing output on a layer isn't necessary,
since it doesn't hit the performance. But we do want to continue to
pause output when there are splits to improve scrolling performance.
Fixes savannah bug #29055. Thanks to Kazuo Teramoto for reporting
the bug and testing the fix.
Height of a window can be negative, when both hardstatus and caption
are turned on, and the terminal size is reduced to one-line height.
So handle those cases more gracefully.
Fixes savannah bug #29037.
It looks like 'ys' can be negative at times, which results in invalid
memory reads, and possibly writes. Valgrind log from Friedrich Delgado
Friedrichs in savannah bug #29050.
