Check out my first novel, midnight's simulacra!
Xcurses: Difference between revisions
(2 intermediate revisions by the same user not shown) | |||
Line 73: | Line 73: | ||
==See Also== | ==See Also== | ||
* [[Outcurses]], which actually went somewhere | |||
* [[Notcurses]], which modernizes NCURSES | |||
* [[TANGE]] | * [[TANGE]] | ||
[[CATEGORY: Projects]] | [[CATEGORY: Projects]] |
Latest revision as of 23:46, 19 February 2020
A curses variant implemented purely with X.org objects, perhaps even OpenGL objects or something. The result ought be significantly faster than a fullscreen Ncurses program atop a terminal emulator -- not to mention worlds more attractive -- all without so much as relinking existing applications. There exists one grim central theme behind Xcurses, and I call it UNTIE:
Whereas libaa and libcaca (beautifully) attenuate high-resolution graphics, reducing them to input suitable for low-resolution "character cell"-based displays, Xcurses (faithfully) renders programmatic character cell output as virtual vector primitives, applies optional transforms, and renders them to a high-resolution bitmap display. Xcurses will have knowledge of the input forms at the character set level, which ought allow for interesting effects...
- Scalable brackets
- True support of GTK/QT themes
- Palette initialized with the theme palette as opposed to a bunch of useless horseshit
- Replacement of box characters (when used in ACS_VLINE etc contexts) with beautiful, properly themed vector art
- Transparency at the
If non-automatic augmentation is considered, we can do fairly tremendous things, all at the level of ncurses complexity.
How might it work?
Take over libcurses.so
Currently...
[skynet](0) $ ls -l /usr/lib/x86_64-linux-gnu/libcurses.* lrwxrwxrwx 1 root root 12 Jun 27 12:28 /usr/lib/x86_64-linux-gnu/libcurses.a -> libncurses.a lrwxrwxrwx 1 root root 13 Jun 27 12:28 /usr/lib/x86_64-linux-gnu/libcurses.so -> libncurses.so [skynet](0) $ cat /usr/lib/x86_64-linux-gnu/libncurses.so INPUT(libncurses.so.5 -ltinfo) [skynet](0) $ file /usr/lib/x86_64-linux-gnu/libncurses.a /usr/lib/x86_64-linux-gnu/libncurses.a: current ar archive [skynet](0) $
instead, of course, we'd have...
[skynet](0) $ ls -l /usr/lib/x86_64-linux-gnu/libcurses.* lrwxrwxrwx 1 root root 12 Jul 29 04:03 /usr/lib/x86_64-linux-gnu/libcurses.a -> libXcurses.a lrwxrwxrwx 1 root root 13 Jul 29 04:03 /usr/lib/x86_64-linux-gnu/libcurses.so -> libXcurses.so [skynet](0) $ cat /usr/lib/x86_64-linux-gnu/libXcurses.so INPUT(libXcurses.so.1 -ltinfo) [skynet](0) $ file /usr/lib/x86_64-linux-gnu/libXcurses.a /usr/lib/x86_64-linux-gnu/libXcurses.a: current ar archive [skynet](0) $
Link in X.org libraries
- Good:
- Only one object implementing the curses API exists in the core system
- Bad:
- Conflicts with Ncurses etc, requiring packaging work
- Nonsense:
- All binaries linking curses will get bigger. No, learn how shared libraries work.
- Curses apps won't run without an X session. No; Xcurses contains Ncurses as a subset and fallback.
- There might be a problem with some lame X library aborting without DISPLAY or something.
- Need verify this is not the case, or better yet become generally robust to this problem.
Dynamically open X.org libraries
- Bad:
- Bad kind of in the same way burning one's testicles is bad, which is to say unpleasant.
Hybrid linking
- Link all the X crap into libXcurses-bigfatbagofshit.so
- Dynamically open it from libXcurses once we've verified it's safe to do so, meaning both that:
- all component objects' _init functions (see GCC documentation, "How Initialization Functions Are Handled") may be safely invoked, and
- resource consumption in the event of non-use is negligible
Take over libcurses.so
Interpositioning
We could link libcurses to libXcurses in a non-canonical directory, allowing LD_LIBRARY_PATH/LD_PRELOAD to cause this libcurses to be loaded at runtime. If desired, this could take place in X startup scripts.
- Good:
- Normal, assumed-working curses is still around, in case we exhibit unworkaroundable bugs
- Signed+verified binary+library chains continue to work (without us, of course)
- Bad:
- Possible source of confusion
- Should the variable be unset or absolutely set by another application, programs linked to libcurses and loaded within that environment will function differently.
- sudo (without -E) clears the environment by design, as do other apps.
- Regular use of these variables is bad practice
- Security hole should any containing directory be writable
- Is ignored by tools which walk linker cache or standard lib directories
- Adding the new directory is a maintenance hassle
Relinking
Complications
- Graceful degradation
- TERM: In order to support