| Commit message (Collapse) | Author | Lines |
|---|
|
The 'spawn_sync' function has been deprecated for a very long time now,
but we could never use its intended replacement as its Vala bindings
were broken [1]. The underlying problem had been identified in [2] a
while ago, but has only been recently fixed in [3] and is now available
in Vala's 0.56 branch.
Since we rely on the child PID as a fallback for opinionated users or
shells that are unaware of OSC 7, we have to pass a callback function to
the 'spawn_async' call so we can capture the PID which is no longer
supplied by the method itself.
[1] https://gitlab.gnome.org/GNOME/vte/issues/227
[2] https://gitlab.gnome.org/GNOME/vala/-/issues/721
[3] https://gitlab.gnome.org/GNOME/vala/-/commit/f058e9e828f494ebf8d60cabff7999f219048623
|
|
|
|
Currently weltschmerz prefers working directory information obtained
from OSC 7 to that from procfs. If a user has not configured their shell
to emit OSC 7 escapes, the OSC 7 path may be out of date. Additionally
some users want only their shell to change the terminal's conception of
working directory, which is a behaviour better matched by the procfs
based working directory detection.
This change allows OSC 7 based working directory detection to be turned
off. The default remains to check OSC 7 first and then fall back to
procfs if there is not valid local path set with OSC 7.
The reason for turning OSC 7 off entirely instead of inverting the order
procfs and OSC 7 are checked in is that procfs based detection should
never fail under normal usage on systems that support it.
|
|
Currently the configuration object is located in the function level
scope of load_config() and deleted once the function exits. Any flags
that control the program behaviour must therefore be copied into
separate variables, which pollutes the the upper level namespace.
This change keeps the configuration object around after load_config()
exits and makes it accessible by other functions from the shared
namespace. An upcoming change needs this access.
|
|
Opening a terminal or file manager window requires weltschmerz to know
the current working directory. At the moment weltschmerz relies on the
application being run in the terminal to generate OSC 7 codes that
specify the path. However, OSC 7 is not yet widely supported by default,
and the VTE terminal emulation layer, which manages the OSC 7 path state
for weltschmerz, will overwrite local paths with ones that point to a
remote computer if an OSC 7 enabled application is run under ssh.
This change adds a fallback that uses the symlink to current working
directory located at /proc/<child pid>/cwd. On Linux this method should
always work and was how these features were implemented before adoption
of OSC 7. The procfs method is used as a fallback instead of the primary
method since it can only see the working directory changes of the direct
child process.
|
|
We currently consider periods, commas, colons, semicolons, exclamation
points and question marks following a URL as part of it. While ending a
URL with these characters is technically speaking valid, more commonly
they are a result of people ending a clause with a URL. Let's ignore
such characters if they are present at the very end of a URL.
Make sure to still match them, however, if any non-punctuation
characters follow. This way, almost all actual usage of punctuation
within URLs is unaffected.
|
|
This commit adds a new section, open-with, to the configuration file.
The options specified there will be added as "Open with …" menu items in
the URI context menu. This is to make it easier to perform other actions
on the URI than opening it in the system default browser; a user might
for example add another browser, media player, or a script that
preprocesses the URI before opening it.
|
|
The upcoming "Open with …" feature will need to spawn subprocesses in
very similar way as open_terminal(). Moving process creation to its own
function allows the same logic to be reused.
|
|
Currently weltschmerz has separate context menus for the two types of
URIs one can interact with, even though they contain the same option
with only a little different phrasing. This makes both use the same
context menu, and standardizes terminology to "URI".
This unification is in preparation for an upcoming "Open with …" feature
which will programmatically add items to context menu for URIs. Without
this, we would need to duplicate logic for each of the separate menus.
|
|
|
|
Currently the weltschmerz URL regex does not match URLs with quotes or
parentheses, considering the URL to end when one is encountered. On the
other hand, if a URL is surrounded by angle brackets it includes the
closing '>' in the URL match. Additionally, the regex allows URL to
contain a space if and only if it is the second character of the host
component of the URL.
Most of this appears to be down to bugs in the regex as it is currently
written. This rewrites the regex to be cleaner and easier to read, while
maintaining the intended logic of the original.
|
|
The proposed OSC 7 specification mandates [1] that the file:// URI
includes the hostname, so that the terminal emulator can ignore remotely
used programs' attempts to set the current working directory. However,
VTE does not check the hostname component, and neither does
launch_default_for_uri. Thus currently if one tries to open directory
after running a remote program that used OSC 7, weltschmerz tries to
open a file manager pointing at corresponding path on the local
computer.
weltschmerz already contains a function that checks the URI kept by VTE
and returns the path only if it is on the local computer. This commit
changes open_directory() to first use that, and then create a new local
file:// URI based on it, unifying the behaviour of 'Open directory' and
'Open terminal'.
[1] https://gitlab.freedesktop.org/terminal-wg/specifications/-/issues/20#note_294252
|
|
With the translation infrastructure now in place, make sure to mark
every user-bound message for translation. All but one of the strings in
terminal.ui are already marked. This commit marks all remaining messages
in the Vala code as well as the missing one from terminal.ui.
One of the messages in terminal.ui is a sample error message for the
infobar label. Since this message will never be visible to the user (the
code overrides the label's contents with any relevant warning messages),
we do not need to keep it marked.
Finally, give translators ample context for what they will have to
translate and fix a small inconsistency with the capitalization of
warning messages in configreader.vala.
|
|
Add ability to open a new weltschmerz terminal, either from context menu or
with a key combination. The new terminal is opened in the directory
indicated with OSC 7, allowing quickly opening additional terminals while
working.
Preferably this would launch the user's preferred terminal, as defined
in per-user settings, but this is not possible with glib[1]. For this
reason the option always launches another weltschmerz.
weltschmerz is launched using bare Process.spawn_async() on the name of the
program as defined in weltschmerz.vala. We considered using the AppInfo
database to retrieve the name of executable instead, but as there is no way
to query the database for terminal emulators we would have to go through
every single program installed on the computer and try to find one called
weltschmerz. As the change would merely replace the requirement of having
weltschmerz in PATH with the requirement of having a .desktop file with the
name in one of the standardized locations, this would not be worth it.
[1] https://gitlab.gnome.org/GNOME/glib/-/issues/338
|
|
Currently, weltschmerz only copies URLs to GTK's default clipboard. This
is defined to be CLIPBOARD, leaving the PRIMARY selection unchanged.
Since all other normal copy/paste operations start with making a
selection and therefore set PRIMARY, make sure to also set the PRIMARY
selection when copying a URL. This brings weltschmerz in line with how
other GTK apps behave.
|
|
No functional change, since vte_copy() defaults to false here. This
seems to be a remnant of when vte_copy() required specifying the 'html'
argument.
|
|
Previously, in commit c49fbfb (Set minimum width and height in window
geometry, 2021-06-17), we set the minimum size of the window to one
cell. We do not yet include our fixes from 89f8571 (Calculate base_width
dynamically, 2021-06-18) and so do not account for the scrollbar when
setting the minimum size hints.
A minimum size of one cell can lead to issues like [1] when resizing
aggressively. A fix for this has been committed in [2], but not yet
merged to any release branch. Setting the minimum size to larger values
seems to work around this problem.
Therefore, define a minimum window size of 28 x 3 cells - just big
enough to still fit the search overlay. Make sure to include base_height
and base_width so that the right size will be reported to the window
manager.
[1] https://gitlab.gnome.org/GNOME/vte/-/issues/340
[2] https://gitlab.gnome.org/GNOME/vte/-/commit/b333d66879963637099dc0bc5a702f50f34da67e
|
|
|
|
weltschmerz sets window geometry hints to indicate how the terminal
window is to be sized and resized. New sizes are always kept at a
multiple of the character width and height. Effectively, this ensures
that terminal windows never have "negative space" that is not used for
any terminal output.
In order to account for padding and non-terminal widgets, one may
specify base_height and base_width. These values indicate how much space
to add in a final step, after the character width and heights have been
multiplied with the amount of columns and rows respectively. For
example, allowed window widths are always:
(char_width * N) + base_width
Currently, we always use a base_width of 2 * 2. This is to account for
the padding of 2 pixels as specified in terminal.css.
weltschmerz uses GtkScrolledWindow and overlay scrolling by default.
With overlay_scrolling enabled, the scrollbar does not take up any extra
space. The previous commit works around a problem with GtkScrolledWindow
and sets PolicyType.ALWAYS when overlay scrolling is turned off. This
means that we now also have to account for the size of that scrollbar.
Since we cannot simply query the size of the scrollbar, we have to be a
bit more creative. Use the full window width and subtract the width of
the terminal at startup (we know it is always 80 columns) to get the
amount of space taken up by "anything else". This includes VTE padding
and the scrollbar size.
Importantly, this needs only be done once and is completely agnostic to
any further changes to the terminal or font size (which affects the cell
width and height). We only assume that both the size of the scrollbar
and the amount of padding is constant, which we feel to be a reasonable
trade-off.
|
|
weltschmerz uses GtkScrolledWindow to provide a vertical scrollbar to
the user. By default, this widget enables overlay scrolling. This means
that the scrollbar is painted "inside" the terminal, as if overlaid.
There's no configuration switch to turn this behaviour off, since we are
convinced that overlay scrolling is the right thing to do for terminals.
However, GTK allows users to turn off overlay scrolling in a global
setting [1] or by using the 'GTK_OVERLAY_SCROLLING' environment
variable. Some distributions turn overlay scrolling off by default.
When overlay scrolling is disabled, GtkScrolledWindow falls back to
rendering the scrollbar normally. Since our scrollbar policy is set to
AUTOMATIC, it is only drawn on demand. This means that it is initially
not be visible but can suddenly take up effective window space once it
appears.
When this happens, depending on what is going on in the terminal, one
can be left with a terminal that is scrolling down infinitely as the
reported column width flaps between two values: one accounting for the
size of the scrollbar and the other one ignoring it. Each time the
column width flaps, VTE sends SIGWINCH, possibly getting caught in an
infinite loop of internal resizes.
Since upstream declares GtkScrolledWindow incompatible with VTE [2] [3],
we have not attempted further investigation on this issue, but may
report an issue in the future.
We definitely want to keep using GtkScrolledWindow even with this
incompatibility - it is the most elegant and visually pleasing solution.
The alternative would be to use a GtkBox and supply our own scrollbar.
This scrollbar would either always be visible (even if there is nothing
to scroll) or entirely hidden. An implementation of this may be found in
the box-scroll branch. Note, however, that we do not guarantee that
branch to be up to date with the latest developments.
To work around this problem, detect when overlay scrolling is turned off
and fall back to PolicyType.ALWAYS. This ensures that the scrollbar is
always visible, mitigating the infinite resize loop. Crucially, compared
to the alternative solution, this will still enable users to enjoy
overlay scrolling if they (or their distribution) do not turn it off.
[1] https://developer.gnome.org/gtk3/stable/GtkSettings.html#GtkSettings--gtk-overlay-scrolling
[2] https://gitlab.gnome.org/GNOME/vte/-/issues/11#note_264921
[3] https://bugzilla.gnome.org/show_bug.cgi?id=733210#c2
|
|
Even though PolicyType.AUTOMATIC should always keep it hidden, be
explicit about our wish to show it under no circumstances.
|
|
Commit c49fbfb (Set minimum width and height in window geometry,
2021-06-17) introduced a trailing comma to make subsequent changes to
the initializer less of a hassle.
However, Vala only accepts trailing commas starting with 0.52.3 [1]. It
would be petty to require this version, especially since older versions
of Vala can still compile weltschmerz without problems.
[1] https://gitlab.gnome.org/GNOME/vala/-/commit/20fcf9ce42dca52c707b96ddf7457931d6ee96f5
|
|
weltschmerz currently does not indicate any minimum size for its window
geometry. If the minimum size is not set, GtkWindow uses its requisition
as the minimum size [1]. Make sure, instead, that we control this value
and set it to one cell.
[1] https://valadoc.org/gdk-3.0/Gdk.Geometry.html
|
|
Currently weltschmerz always uses the global GTK setting for cursor
blink, which means that the cursor blink setting is stored separately
from the rest of the configuration. Add a setting for overriding it in
the config file, so that all the relevant settings can be adjusted in
one standardized place.
Reviewed-by: Wolfgang Müller <wolf@oriole.systems>
|
|
See [1] for background. This was motivated by a warning from valac-0.50.
[1] https://gitlab.gnome.org/GNOME/vala/-/issues/1121
|
|
There was no VTE-related functionality in this method, so we can find a
better place for it now that a dedicated utility class exists.
|
|
VTE can leverage the child program's support of OSC 7 to keep track of
the current directory. This commit adds support for opening said
directory in the default file manager, either by using a shortcut or by
activating a new entry in the context menu.
|
|
This commit adds a "Copy to HTML" entry to the context menu. It is
active when VTE registers an active selection. Lines copied like this
will be put into the clipboard as formatted HTML, retaining nearly all
styling information.
For this, vte_copy now takes a boolean argument - whether or not to
format the lines copied as HTML. As two menu items now access vte_copy
directly, and there is no clean way of passing arguments to signal
handlers through terminal.ui, signals for both menu items are now
connected in the Vala code instead.
|
|
This commit adds support for the OSC 8 hyperlink escape sequence [1]. As
this is not a mature feature and there seem to be outstanding security
concerns [2], the setting that controls whether or not OSC 8 is
interpreted is disabled by default.
Just like gnome-terminal, weltschmerz will display a tooltip with the
canonicalized URI when hovering over a hyperlink.
[1] https://gist.github.com/egmontkob/eb114294efbcd5adb1944c9f3cb5feda
[2] https://gist.github.com/egmontkob/eb114294efbcd5adb1944c9f3cb5feda#security
|
|
This commit improves and simplifies weltschmerz's configuration
handling. Obtaining and parsing the KeyFile is split out into
ConfigReader, which is fully agnostic of the specific configuration
values.
Config now contains all configuration values in the form of primitive
types or class instances, and no longer delegates access to the KeyFile
structure directly. This means that the configuration file is read once,
and then kept in the Config instance.
Indirectly this commit also fixes a bug where weltschmerz would segfault
if one of the palette entries contained an invalid value.
|
|
|
|
|
