Apparently on X11 the maximum icon size is 128x128. 256x256 is too
large for the X11 protocol because the X server unserializes the icons
using "unsigned long" which is 64 bits on Linux. So we have to use
64bits per pixel instead of 32, with 32bits padded to 0.
While there I also got rid of the kitty.rgba file replacing it with a
128x128 PNG file.
Fixes#3260
While developing the previous PR (https://github.com/kovidgoyal/kitty/pull/3043), I was using more compiler command line arguments, one of which was `-MMD`. This caused the .d file to be created. Without this argument, there is no need to create a temporary directory.
For some reason it can happen that the `UserNotifications` framework exists but the `UserNotifications/UserNotifications.h` header doesn't.
Closes https://github.com/kovidgoyal/kitty/issues/3042.
This allows compiling Objective-C files.
The temporary directory is needed because clang would try to write to `/dev/null.d`, which would obviously fail.
The new User Notifications Framework is only available on macOS 10.14 and above, while the old NSUserNotification API is deprecated in macOS 11 (Big Sur) and will probably be removed in the future.
This commit compiles a simple test program to see if the Framework is available and then uses either the new or the old API.
When piping the output of `setup.py` to another program, that program cannot usually deal with escape sequences well. To fix this, output the compilation progress on new lines instead of overwriting the current line.
Disabling Link Time Optimization is useful for Nix on Darwin because LTO is broken there and has been for a long time, see https://github.com/NixOS/nixpkgs/pull/19312.
This is currently worked around in the Nix package with a patch that removes the lines that add -flto to the compiler flags.
The current notifications framework has been deprecated in Big Sur. The new
framework only allows notifications from signed and notarized applications,
so people using kitty from HomeBrew/source are out of luck. And
notifications can only be displayed once the user grants permission. A
completely brain-dead design. Complain to Apple.
This required the kitty package to be moved from Frameworks to
Resources, because Apple cant codesign .pyc files and its notarization
requires everything in Frameworks to be signed. Hopefully that does not
break anything. Also removed the kitty-deref-symlink since it is not
needed anyway and was only present for backwards compatibility.
Fixes#2040
png2icns is used when building kitty with nix because iconutil seems to be closed-source.
libicns also has an iconutil clone called icnsutil in the source tree but the last release is from 2012, which does not include this utility yet.
According to the text just above https://docs.python.org/3/library/exceptions.html#EnvironmentError, `EnvironmentError` has been an alias of `OSError` since Python 3.3. Replacing it makes the code more consistent since `OSError` is used in other places in the code too.
When a linker process fails, `failed` will be set to a value other than `None`. When a second linker process fails, the `else` case will be taken because `failed is None`, which executes `compile_cmd.on_success()`. This function tries to rename or move the file generated by the linker but since the linker process failed, the file will most likely not exist. This will throw an error, which will prevent printing the actual error message from the linker.
Files are built in order by size anyway and mtime sort makes link order
mtime dependent which breaks reproducible builds on openSUSE as they
modify mtimes randomnly. See #1804
`file` and `arguments` are accessed using the square bracket syntax, while `output` is accessed using `.get()`. This commit uses square brackets for all three. This will throw a `KeyError` if `compilation_database` does not have the key, which is the correct behaviour IMO.
The macOS specific `package()` code changes the directory. This didn't cause any problems before 9135387cfa since `compile_commands.json` was written to the filesystem before `package()` was called. This is not the case anymore, which results in `compile_commands.json` being written into the `kitty.app` bundle. I fixed the problem by using a context manager to change the direcctory back after `kitty.app` was created.
Now there is only one launcher. Which means it can be used to start
kitty with profiling and ASAN in the natural way. The recommended
way to run kitty from source is now:
./kitty/launcher/kitty
The launcher also automatically re-execs to resolve symlinks on macOS.
I see a couple problems with using git to determine the files to delete:
- git needs to be installed
- The manual cleaning code is supposed to function correctly, so why not always use it? That way bugs in the manual cleaning code are also more likely to get discovered
- If some files ignored by git should not be deleted, they need to be added to a sort of blacklist. If some files should be deleted by the manual cleanup code, they need to be added to a sort of whitelist. If someone forgets to add files to one of these lists, then not deleting files that should be deleted is better than deleting files that should not be deleted.
Calling `Env()` without the `ldpaths` parameter after a call with the `ldpaths` parameter, keeps the list from the first call because the argument is mutable. This is probably not what was intended. This commit fixes this issue and also reverts a change to two lines introduced in 091e74d618, which was probably a workaround for this issue.
Using
```Python
with suppress(OSError):
os.remove('somefile.tmp')
```
instead of
```Python
try:
os.remove('somefile.tmp')
except OSError:
pass
```
makes the code more compact and more readable IMO.
This pattern was recommended by Raymond Hettinger, a Python Core
Developer in his talk "Transforming Code into Beautiful, Idiomatic Python" at https://www.youtube.com/watch?v=OSGv2VnC0go. The transcript is available at https://github.com/JeffPaine/beautiful_idiomatic_python
This should make tracking down the root cause of the
event loop pauses on X11 easier. And the infrastructure
should come in handy in the future as well.
This is needed for harfbuzz 2.0 on macOS. -std=c11 was supported as of
gcc 4.7 released in 2014. So hopefully this wont break compilation on any
systems otherwise capable of running kitty. Fixes#1196
