Do not commit the surface with the new scale and the old buffer.
Leave it to the next rendering pass to commit the surface.
Fixes the following protocol error:
wl_surface@10: error 2: Buffer size (717x795) is not divisible by scale (2)
Closes: https://todo.sr.ht/~exec64/imv/20
imv is currently binding to the latest version of all the interfaces,
but should instead bind the latest version that it supports. imv is not
compatible with wl_output v4 and so was crashing when the latest wlroots
offered it.
This patch pins a maximum version for each wayland interface that is
bound.
https://todo.sr.ht/~exec64/imv/1
Before this commit, the background rectangle was drawn to cover only one line
of text (which was ok since multiline overlay text was not supported). Now
the code correctly calculates the dimensions on the rendered text and put the
rectangle of the right size under it.
This effectively splits the imv_canvas_printf function into two separate
functions, one to create a layout containing given text and the other one
to show it on the canvas. These functions can be useful for the future code
I will add to display text with a background rectangle behind it. Existing
imv_canvas_printf function does not allow this
This prevents wordexp function from splitting the output of shell expansion
into “words” and, as a consequence, removing the newlines from it. With this
commit applied, setting
overlay_text = $(echo -e 'hello\nworld!')
displays
hello
world!
in Imv window. Without this commit, the same overlay_text setting would display
hello world!
since wordexp splits the string "hello\nworld!" into two words and connects
them with a space.
This should not break any of the commands executed in $(), because the IFS
variable is *not* inherited by their shell. The commands don't see this change
and run with the default IFS value.
OpenGL is used for rendering because `cairo_rectangle` and
`cairo_pattern_t` had performance problems on zoomed and rotated images.
The size and contrast of the chequered pattern is also reduces to match
other image viewers and editors.
Closes#253
The current implementation of freeimage backend automatically rotates the
images based on EXIF info. Libjpeg backed doesn't do that (although libjpeg
itself may be able to, I didn't check). This commit makes Imv use freeimage for
jpeg files when both backends are available to enable automatic image rotation
when possible.
A cleaner solution could be to add automatic rotaiton functionality to libjpeg
backend as well, but this quick hack is better than nothing ☺
This seems to have a better behavior, this way, the fonts in Imv seem to have
the same size as the fonts in other programs (Sway, Waybar) when configured in
the same way.
Before this commit, the code in src/wl_window.c seemed to handle the Wayland
scaling [1] correctly, it was sending resize events to imv to update the buffer
size accordingly. One thing it didn't update is the scaling of fonts that Pango
renders on Cairo.
This commit simply forwards the scaling factor (computed as [1] requests)
together with updated buffer dimentions in resize event, and when the resize
event is handled it calls cairo_surface_set_device_scale to notify Pango/Cairo
of the scaling.
For X11, I simply assume the scaline factor is always 1. This seems to be what
the old code did: `grep scale src/x11_window.c` gives no matches. AFAIK, X11
does not have an established way of telling clients what scaling factor to use
(and never updates it at runtime).
[1]: https://wayland-book.com/surfaces-in-depth/hidpi.html
* inih is now widely available as a system library
* if not, a fallback from Wrap DB can be used
Signed-off-by: Stephan Lachnit <stephanlachnit@protonmail.com>
This issue was spotted on musl, where killing imv with the Q shortcut
while it was waiting for image paths in stdin led to the application
hanging until stdin received an EOF manually with ^D, since the stdin
stream was locked by the fgets() function.
Switching to pipes and a helper thread allows us to use read() for
reading from STDIN_FILENO while still taking advantage of the fgets()
logic for handling newlines and buffer size. read() is a thread
cancellation point (fgets() is only an optional one), which makes it
possible to cancel the helper thread cleanly and force the thread using
fgets() to exit when the write-end of the pipe it's reading from is
closed.
Since this isn't essential functionality, it's ok to simply not provide
it. In cases where XDG_RUNTIME_DIR was empty (but not unset) or set to a
directory where the user didn't have write permissions, socket creation
would fail and lead to segmentation faults in imv, due to the return
value of imv_ipc_create() not being checked.
Add newline to the prompt for reading paths from stdin.
Since string formatting isn't being used, it's possible to use fputs()
directly in the logging callback in imv.c.
split color_rgba into color_rgb and alpha
added configuration options to man pages
cleanup gitignore: the man pages output is now in the build directory,
so there is no to ignore man pages individually
* first steps for colored overlay
* overlay color wip
* add option for overlay position
if overlay_positiion_bottom is true overlay is at the bottom
* remove enum for overlay position
Previously, they would be sorted by however readdir returned them.
I suppose a proper configuration option to control sorting would be
nice, but for now, I think replacing forced random order with forced
alphabetical order is an improvement.
mmap with MAP_SHARED would fail if the compositor provides a read-only
descriptor with keymap. And at least weston applies F_SEAL_WRITE to the
memfd if supported by the platform.
Fixes#263
Calling `rsvg_handle_close` doesn't do anything here because the handle
is already constructed with data or file. `g_object_unref` is called
instead to free memory.