| Commit message (Collapse) | Author | Age | Lines |
| | |
|
| | |
|
| |
|
|
|
| |
They were deprecated in 1.26.0, along with the 'colors' and 'colors2'
sections.
|
| |
|
|
|
| |
They were deprecated in 1.26.0, and replaced by colors-dark and
colors-light.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Currently, if config option 'selection-override-modifiers' is set
to 'none', the mouse is _always_ grabbed for selection, making
mouse tracking mode unusable.
A more reasonable behavior is to _never_ override mouse grabbing
instead.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, if a mouse binding contains _any_ modifier key set in
config 'selection-override-modifiers', it does not work since
these modifiers get unset in function 'match_mouse_binding()'.
Function 'resolve_key_binding_collisions()' is supposed to detect
this and display an error message, however, it does not work unless
the conflicting modifiers are identical.
For example, given the following config options:
selection-override-modifiers=Control+Shift
select-begin-block=Control+BTN_LEFT
Foot does not display an error message for this, but if you try to
select text with 'Control+BTN_LEFT', it behaves as if only 'BTN_LEFT'
is pressed.
This is due to function 'modifiers_disjoint()' being implemented as
"not equal" rather than "disjoint".
Fixed by properly implementing this function. Also clarify the
'selection-override-modifiers' config option in the man page.
As a side effect, this also fixes 'selection-override-modifiers=none'
conflicting with mouse bindings that do not contain modifier keys
(an empty set is disjoint with itself).
|
| |
|
|
|
| |
The trailing '+' is already removed in modifiers_to_str().
Don't do that twice.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before this, the col/row coordinates for the mouse pointer was only
updated when the mouse either entered our window, or was moved inside
it.
However, when a window is resized without using the mouse, and the
user then clicks (again, without moving the mouse), we can get a
button event outside the cell grid, that we still think is on top of
the grid.
Example: user moves mouse over the bottom row. The window is
shrunk. Enough that the mouse is now outside the cell grid, but still
inside our window. This means we get a button event outside the
grid. But since we haven't updated the col/row coordinates, we think
it's on a row, that no longer exists.
Closes #2435
|
| |
|
|
|
|
|
|
| |
It's a compositor bug to emit a keyboard_enter() event with a NULL
surface, but has been seen happening on Hyprland. Workaround by
ignoring the event.
Closes #2438
|
| |
|
|
|
|
|
|
|
|
|
| |
This will allow for foot to be registered as a MIME type handler for
directories, allowing a "open with foot" action via xdg-open.
Some applications pass a simple directory path (eg. qbittorrent) but
others pass it as a "file://" URL (lutris), which is also sometimes
percent-encoded (calibre) but sometimes not.
This patch should be able to handle all 3 of these variants.
|
| | |
|
| |
|
|
|
|
|
|
| |
Without keymap, never grab the mouse for selection when
mouse tracking is enabled, since we can't tell whether a
mod key is being held.
Fixes #2443
|
| |
|
|
| |
see https://github.com/mesonbuild/meson/issues/14948
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
wl_touch has no leave event, so unmapping the surfaces cannot drop touch
focus the way it drops keyboard and pointer focus. seat->touch.surface
kept pointing at the window being destroyed, and seat->touch.state stayed
HELD/DRAGGING, so the guard in wl_touch_up() and wl_touch_motion() did not
fire. A wl_touch.up dispatched from the roundtrips in wayl_win_destroy()
then dereferenced the freed window:
#0 wl_touch_up () at input.c:3587
#12 wayl_roundtrip () at wayland.c:2416
#13 wayl_win_destroy () at wayland.c:2297
Reproduced by touching a foot window and holding while the window is
destroyed: 3/3 crashes before, 0/7 after.
|
| |
|
|
|
|
|
|
|
|
|
| |
wayland 1.26 adds wl_fixes::ack_global_remove(). Clients should call
it when they've received a global_remove() event, after doing internal
cleanup.
Not tested yet:
* Waiting for arch to update wayland to 1.26
* None of the compositors I use (or test on) implement it yet.
|
| |
|
|
|
|
|
|
|
| |
When running e.g. "foot true", we may reach a state where the window
has received the configure event, but we've also already decided to
shutdown. In this case, we may not have a grid (render_resize()
doesn't do anything when we've started the shutdown routine).
Fixes a race that triggers a crash.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
DECSC/DECRC (ESC7/ESC8), CSI s/u and DECSET/DECRST 1048/1049 all
save/restore the SGR attributes through the same, global saved_attrs
slot.
An application emitting ESC7/ESC8 in the alternate screen to wrap
sixel rendering in a save/restore-cursor pair overwrites the slot
with its own SGR state. DECRST 1049 then restores those attributes,
instead of the ones saved by DECSET 1049. The shell prompt is left
with the application's attributes, even though the application reset
them (sgr0) before exiting.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
Selection coordinates may appear outside of the terminal grid
(e.g., during the "fast resize" stage of delayed resize),
resulting in bad width passed to pixman_region32_union_rect().
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Applications cannot infer whether their terminal view is observable from
keyboard focus alone, so they may continue expensive rendering while the
surface is hidden.
Implement private mode 2033 and the one-shot visibility query. Report
Wayland xdg_toplevel suspension as not visible and otherwise remain
conservative by treating the terminal as potentially visible.
Amp-Thread-ID: https://ampcode.com/threads/T-019fb2b7-f245-7124-a69a-1484812532bb
Co-authored-by: Amp <amp@ampcode.com>
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Use the smallest of the *implemented* version, and the *preferred*
version.
Before this, we used the minimum of our *required* version, and
the *preferred* version, which is both useless and wrong, since we'll
always pick the required version (since required <= preferred).
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Interactive resize of a terminal window breaks current selection,
if scrollback is not empty, and option 'resize-delay-ms' is not 0.
This behavior is introduced by commit 296e75f4f5, which:
- Subtracts selection row by view offset every time render_resize()
is called, resulting in bad coordinates.
- Does not properly restore the original selection coordinates
during the final text reflow stage.
|
| | |
|
| |
|
|
| |
Closes #2405
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
| |
* Allow single-quotes _inside_ URLs (but still not at the end)
* Require at least one character inside balanced parenthesis or
quotes.
Closes #2417
|
| |
|
|
| |
Closes #2401
|
| |
|
|
|
| |
* ruff 0.16.0
* python 3.14.6
|
| |
|
|
|
|
|
|
|
| |
The IME commit handler appended text to the search buffer and refreshed
the search box, but never called search_find_next(). CJK input into
scrollback search therefore showed no match until another key press
re-triggered the regular search input path. Call search_find_next() in
the commit branch, and apply the same fix to unicode-mode.c. To allow
this, search_find_next() is no longer static.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
See the added comment for reasoning.
The reallocarray(3) function in libc could have been used conditionally,
when available, but since it can be implemented simply and efficiently,
the portable version may just as well be used in all cases.
See also:
* https://pubs.opengroup.org/onlinepubs/9799919799/functions/reallocarray.html#:~:text=adding-,reallocarray
* https://android.googlesource.com/platform/bionic/+/HEAD/docs/status.md#:~:text=reallocarray
* https://github.com/termux/termux-packages/tree/26826aec834182c/x11-packages/foot
* https://github.com/termux/termux-packages/blob/26826aec834182c/x11-packages/foot/reallocarray.c#L6
* https://github.com/termux/termux-packages/blob/26826aec834182c/x11-packages/foot/xmalloc.c.patch#L7
* https://github.com/termux/termux-packages/pull/17818
* https://github.com/termux/termux-packages/issues/23833 (see expanded build log)
|
| | |
|
| |
|
|
|
|
|
|
| |
This patch fixes rightmost boundary check for a word wrapped to
another line, where a multi-column character appears as the last
character in the original line.
Closes #2411
|
| | |
|
| |
|
|
| |
Closes #2367
|
| |
|
|
| |
Closes #2362
|
| |
|
|
|
|
|
| |
This can be quite a lot, but hopefully less than 0x7fffffff, which is
the max value a VT parameter can have...
Closes #2366
|
| |
|
|
|
|
|
|
|
|
| |
It's circular, but there's no need to walk through it more than
once...
csi: kitty: limit pop count to the size of the flag stack
It's circular, but there's no need to walk through it more than
once...
|
| |
|
|
|
|
|
|
|
| |
These characters have special meaning, and when they appear in the
font *family name*, they need to be escaped in the font pattern.
Before this patch, even if the user did escape them properly, _we_
un-escaped them in FcNameUnparse(), causing the wrong font to be
loaded later.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In server mode, if a terminal instance triggered a tracked
notification (e.g. a bell notification, or a kitty notification), then
the foot server crashed if the terminal instance was closed before the
notification.
This happened because the notify_done() callback's user supplied
argument was the terminal instance, and this pointer was not verified
before it was de-referenced.
Fix by searching the wayland global's list of terminal instances, and
ignore the callback if the wanted terminal instance is gone.
Closes #2397
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is to help e.g. scripts that wants to launch terminal instances
in a terminal agnostic way.
Many terminals were originally for X11, where the '--class' choice was
obvious. When they were ported to Wayland, it made sense for them to
keep the '--class' command line option, but have it set the
application ID instead (since "class" is not a thing on Wayland).
Let's add the alias, but minimally documented. We don't want to
_encourage_ its use after all.
Closes #2368
|
| |
|
|
| |
Like we do with all other surfaces.
|
| |
|
|
| |
Closes #2363
|
| |
|
|
|
|
|
| |
This fixes a crash further down, where we try to lookup an existing
compose sequence.
Closes #2364
|
| |
|
|
|
|
|
|
|
| |
This fixes a copy-paste error where we read/modified/wrote the
private-palette instead of the shared palette.
Since the private palette is NULL in this case, that meant a crash.
Closes #2370
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Pad is always (in real life) 1, and no-one should be setting pan to
anything larger than 5 (the largest value possible to set with the
sixel init parameters). FWIW, no-one really uses anything but 1 here
either.
Clamping ensures we don't blow up allocations to something huge, or
worse, trigger an overflow that results in a pixel height of 0,
causing a division-by-zero.
Closes #2371
|
| | |
|
| |
|
|
|
| |
Some colors in dark theme have slightly different I assume more
correct values then previously
|
| |
|
|
|
|
|
|
|
|
| |
Eliminate the outer loop, so that when moving the cursor multiple
tab stops, we no longer iterate the tab stop list all over again.
This also fixes a DoS flaw when passing a very large value as
CHT/CBT argument, which may hang the terminal.
Closes #2360
|
| |
|
|
|
|
| |
This patch fixes a condition check in selection_update(), so that
during block selection, the selection area no longer gets incorrectly
updated when selecting back across the starting column.
|
| |
|
|
|
|
|
| |
That is, only do grapheme clustering when printing codepoints in
sequence, without any cursor movements in between.
Closes #2383
|
| |
|
|
|
|
| |
`dst_bottom > dst_top` holds for every multi-row destination (rows
increase downward), so the guard rejected all multi-row copies. Flip to
`dst_top > dst_bottom`, mirroring the sibling column check and DEC STD 070.
|
| |
|
|
|
|
|
|
| |
We use XKB_KEYSYM_MAX, which was added in 1.6.0. In other words, this
requirement isn't really new, but now we've formalized it in
meson.build.
Closes #2379
|
| |
|
|
|
|
|
|
|
|
| |
Copy-on-select (configured with 'selection-target') may accidentally
clear the clipboard, if the user drags the mouse a little bit when
clicking inside a terminal window.
Now we only copy if there is actual text being selected.
Closes #2327
|
| |
|
|
|
|
|
|
|
|
|
|
| |
dst_right was already being clamped, but not dst_left.
In addition to that, reject the sequence if dst_left > dst_right, or
dst_bottom > dst_top.
This is already being done for the source rectangle, in
params_to_rectangular_area().
Closes #2352
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If the input URI ends with a trailing '%' (or a trailing '%N'), we
read outside the provided buffer.
On NULL terminated input, this happened to work out since we'd
correctly detect an invalid %-sequence as soon as we read the NULL
terminator.
On input that is not NULL terminated, we're out of luck.
This patch fixes this by also checking we have enough input left to
even _try_ to read the %-digits.
Also add unit tests for this particular case.
Closes #2353
|
| | |
|
| |
|
|
|
|
|
|
|
| |
We make use of __COUNTER__, which recent versions of clang emit a
warning for (unless you compile with -std=c2y).
Our meson.build already handles this for the actual build, but if your
build is using gcc, -Wno-c2y-extensions isn't added to the build
rules (since gcc doesn't support the flag).
|
| | |
|
| |\ |
|
| | | |
|
| |/ |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When we initially implemented the kitty notification protocol, 128
bytes was enough.
Since then, the protocol has grown, and 128 bytes is no longer enough,
resulting in a crash when you try to query the protocol:
printf "\e]99;i=123:p=?;\e\\"
Results in
#0 0x7fe4fad3dcde in __sanitizer_print_stack_trace (/usr/lib/libasan.so.8+0x13dcde) (BuildId: 38f5643355445a477f07ffa9b6a83a55c87415a0)
#1 0x561ec61454e8 in print_stack_trace ../../debug.c:20
#2 0x561ec6145550 in fatal_error ../../debug.c:28
#3 0x561ec61462d0 in xvsnprintf ../../xsnprintf.c:42
#4 0x561ec614640a in xsnprintf ../../xsnprintf.c:52
#5 0x561ec6131fe6 in kitty_notification ../../osc.c:765
#6 0x561ec6139fec in osc_dispatch ../../osc.c:1511
#7 0x561ec60fcac8 in action_osc_end ../../vt.c:614
#8 0x561ec60fe6bb in state_osc_string_switch ../../vt.c:924
#9 0x561ec60ff437 in vt_from_slave ../../vt.c:1115
#10 0x561ec6078748 in fdm_ptmx ../../terminal.c:297
#11 0x561ec5ee6ed0 in fdm_poll ../../fdm.c:483
#12 0x561ec5f344d3 in main ../../main.c:676
#13 0x7fe4f9e2777d (/usr/lib/libc.so.6+0x2777d) (BuildId: 674f3ebbcb651fbc7d1189cb300acfd7d007849a)
#14 0x7fe4f9e278bb in __libc_start_main (/usr/lib/libc.so.6+0x278bb) (BuildId: 674f3ebbcb651fbc7d1189cb300acfd7d007849a)
#15 0x561ec5e77a14 in _start (/home/daniel/src/foot/bld/debug/foot+0x4b8a14) (BuildId: 9fca511350f3745edd3546ff800a763c1957e673)
Closes #2335
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We already detect if the final cursor colors (foreground/background)
are identical.
If they were, we replaced them with the default fg/bg.
Unfortunately, this can still result in an invisible cursor. For
example, if the current cell's default bg matches the current cell's
fg.
Instead, update the colors in two steps:
1. First, use the provided (cell's) fg/bg
2. If they too match, invert the cursor color
This should ensure the cursor is always visible
Closes #2323
|
| |
|
|
|
|
|
|
|
|
| |
When processing shortcuts, set seat->kbd.last_shortcut_sym. This
ensures the corresponding release event is ignored in the main mode,
if the shortcut triggered a mode exit.
For example, pressing Return in search mode would exit search mode,
and then immediately reset the search match, if the kitty keyboard
protocol mode with release events were enabled.
|
| | |
|
| |
|
|
| |
When enabled, compile with -fanalyzer
|
| |
|
|
|
|
|
|
|
|
| |
I'm guessing, but I'm not sure, that when I initially implemented
this, XTerm ignored shift in keypad escapes.
Testing again, it doesn't. And regardless of what XTerm does, I don't
really see any reason to exclude shift from keypad escapes.
Closes #2324
|
| |
|
|
|
|
|
|
|
|
|
| |
This option allows you to configure which underline style to use with
URLs.
The default is dotted. Since before this patch, URL underlines were
always plain underlines, this means the default URL underline style
has changed, from single to dotted.
Closes #2302
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In 3a2eb80d83d59d194a3d07da227431c634892ff5, we fixed an issue where
key *releases* triggered a selection reset, and viewport reset.
The same issue still exists in other input modes (unicode, search and
url mode); if the kitty keyboard protocol has been enabled, with
release events, any key press directed to e.g. search input handling,
will was not remembered, and the corresponding release event generated
a kitty keyboard event to the terminal application, and reset the
viewport and selection.
This means, for example, that scrollback search was unusable.
Fix by *never* doing any further (generic) key processing if an
alternate input mode is active.
Closes #2316
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When writing paste data to the terminal (either interactively, or as
an OSC-52 reply), we enqueue other data (key presses, for examples, or
query replies) while the paste is happening.
The idea is to send the key press _after_ all paste data has been
written, to ensure consistency.
Unfortunately, we only checked for an on-going paste. I.e. where the
paste itself hasn't yet finished. It is also possible the paste itself
has finished, but we haven't yet flushed all the paste buffers. That
is, if we were able to *receive* paste data faster than the terminal
client was able to *consume* it. In this case, we've queued up paste
data in the terminal. These are in separate queues, and when emitting
e.g. a key press, we didn't check if all _those_ queues had been
flushed yet.
Closes #2307
|
| | |
|
| |\ |
|
| | | |
|
| | | |
|
| | |\
| |/
|/| |
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The pre-apply damage thread may be running when we destroy a terminal
instance, and we need to wait for it to finish before destroying the
underlying buffer.
c291194a4e593bbbb91420e81fa0111508084448 did this, but failed to
realize the thread may get re-started in the roundtrips done later in
wayl_win_destroy(); after the wait added in
c291194a4e593bbbb91420e81fa0111508084448, we unmap all
surfaces (including the main grid), and roundtrip. This means the
compositor will release the currently active buffer, and that means
_we_ will trigger the pre-apply damage thread on it. This introduces a
race, where wayl_win_destroy() may reach its shm_purge() calls before
the pre-apply damage thread has finished. That typically causes foot
to crash.
Closes #2288
|
| | |
| |
| |
| |
| |
| |
| |
| | |
Recent clang versions warn on __COUNTER__, unless compiling with
-std=c2y (which breaks other things).
"Fixes" '__COUNTER__' is a C2y extension (__COUNTER__ is used by our
UNITTEST macro).
|
| | | |
|
| | |
| |
| |
| | |
Closes #2292
|
| | | |
|
| |\| |
|
| | | |
|
| |/ |
|
| |
|
|
|
| |
It hasn't been kept up to date for a long time, and no one seems to
care, so lets remove it.
|
| |
|
|
| |
Closes #2215
|