### Problem description
At least on Windows (I have tested), it fails to save a layout on the
non-portable version of ImHex (unless we have an administrator
privilege).
The log (after an attempt to save a layout as "sample") will look like:
| Component | Message |
| --------- | ------- |
| `libimhex` | `Failed to save layout 'sample'. No writable path found`
|
But the underlying problem is platform-agnostic. It can be also a
problem on other platforms in other ways.
### Implementation description
The layout manager incorrectly queried whether the empty path
(effectively the current working directory) is writable before saving
the layout (not each "layouts" directories it queried earlier).
This is the snippet of the root cause.
```cxx
std::fs::path layoutPath;
for (const auto &path : hex::fs::getDefaultPaths(fs::ImHexPath::Layouts)) {
if (!hex::fs::isPathWritable(layoutPath))
continue;
layoutPath = path / fileName;
}
```
Look at the argument we are passing to `isPathWritable`. `layoutPath` is
a default (empty) `std::fs::path` object and will not be updated until
the directory describing itself is confirmed to be writable.
That caused a problem on non-portable version of Windows because:
1. The current working directory is usually the one of the executable
(`imhex-gui.exe`) and
2. That directory (`C:\Program Files\ImHex` by default) is usually not
writable unless ImHex is executed with an Administrator privilege.
The argument to `isPathWritable` should be `path` (containing one of the
`layouts` directories) and this PR fixes so that.
### Screenshots
### Additional things
This issue is hard to notice when developing because, to reproduce this
bug, the current working directory MUST NOT BE writable (usually
writable when we develop, even when we are working on the non-portable
Windows builds).
This PR compress the debug info in the ELF files built.
This has no impact on the packages (e.g. .deb files) because they themselves have compression, but once installed in the filesystem, they this compression will be beneficial
The compression is opportunistic, happens automatically when possible
For some reason, the web version doesn't work with this (most compiler tests after this seem to fail ?) so it is disabled there
More information: https://github.com/WerWolv/ImHex/issues/1714#issuecomment-2131373826
### Problem description
This PR implements some rudimentary Xcode support for building and
editing ImHex.
### Implementation description
#### Problem 1: Xcode is a multi-configuration buildsystem
The project is already rather CMake generator independent, thus it did
not need to change much to support Xcode's multi-configuration paradigm:
By default, CMake generates a `.xcodeproj` in which targets build their
artifacts into the specified `<>_OUTPUT_DIRECTORY`, postfixed by the
currently active configuration. To better fit the existing paradigm, I
instead opted ot introduce `IMHEX_MAIN_OUTPUT_DIRECTORY`. This variable
is equal to the previously used `RUNTIME_OUTPUT_DIRECTORY` when using
other generators, and is changed to include a configuration specific
_prefix_ when used with Xcode.
The result is different output directories when using Xcode, and no
changes when using any other generator.
#### Problem 2: ImHex does not support AppleClang
To allow building the codebase with Xcode, I have introduced
`IMHEX_IDE_HELPERS_OVERRIDE_XCODE_COMPILER`. Specifying this option to
`ON` will force CMake to honor the user specified compiler settings,
even when using the Xcode generator.
In practice this can be used together with the new "xcode" CMakePreset
to build the project with mainline clang using `xcodebuild`, or Xcode
itself by generating a buildsystem like so:
```
cmake --preset xcode -DCMAKE_PREFIX_PATH=/opt/homebrew/opt/llvm@17
```
This solution is of course not without flaws. The inner workings are a
particularly ugly hack, and mainline clang does not implement the
necessary extensions to allow Xcode to index the code. Regardless this
option is useful to enable future work in terms of bundling/signing
macOS applications in the "intended" way using Xcode without additional
source modifications.
#### Problem 3: Vanilla CMake + Xcode = Bad developer UX
By default, the CMake generated `.xcodeproj` is a mess. Tons of targets
are scattered about, and source files are not organized beyond grouping
them into a "Source Files" and "Header Files" group.
Even "Header Files" is missing, because the ImHex build system does not
regard private header files of libraries as sources of a target, and
Xcode does not try to guess this information.
The solution is twofold:
* Additional code has been added which organizes the targets into a neat
folder structure
* Additional code was added behind a configuration flag
`IMHEX_IDE_HELPERS_INTRUSIVE_IDE_TWEAKS` which automatically creates
source file trees in Xcode targets, and discovers the non-declared
header files via the folder convention.
### Screenshots
N/A
### Additional things
As a bonus: `IMHEX_OFFLINE_BUILD` assumes that ImHex-Patterns is cloned
into the source tree. I have added an additional fallback that tries to
locate it as a sibling folder of `${CMAKE_SOURCE_DIR}`, as this meshes
better with my filesystem setup.
The setup was tested with `CMake 3.29.2`, `Xcode 15.2`, and `llvm@17`
from homebrew.
### Problem description
#### Problem 1
In borderless mode ImHex disables the standard macOS titlebar rendering
and input processing. As a result double clicking the titlebar does not
trigger the native macOS behavior set in `System Settings -> Desktop &
Dock -> Double-click a window's title bar to [Zoom/Minimize/Do
nothing]`.
#### Problem 2
The ImHex window shows up as blank/transparent when de-minimizing it
from the dock.
#### Problem 3
Widgets experience ghost hover inputs from the past position of the
cursor during live resizing.
### Implementation description
ImGui elements consume input events in the order they are drawn. As a
result by "drawing" an `InvisibleButton` over the content area of the
titlebar we can catch unprocessed clicks in the titlebar area.
Connecting this button's double clicks to the native window is then a
trivial endeavour.
The blank windows was caused by the rendering stack clearing the GL
buffer, but proceeding to draw nothing in it. I have short circuited
this path.
Ghost hover inputs were squelched by consistently moving the ImGui
cursor to `0, 0` during a live resize. The OS will dispatch a cursor
positioning event once the resizing ends, restoring normal behavior.
### Screenshots
N/A
### Additional things
N/A
---------
Co-authored-by: Nik <werwolv98@gmail.com>