2 Todo

Kevin Lange edited this page 2015-08-26 19:25:36 -07:00

Table of Contents

• Compositor functionality
• Japanese / i18n updates
• Multiarch / Ports
• Cleanup and POSIX expansion
• New EXT2 drivers
• TODO as of November, 2012
• TODO for 0.4.0 Distribution Release
• TODO for 0.5.0
• Doc Revamp
• TODO for Microkernel Launch (0.5.0?)
• Service Modules (aka "Services")
• Future Servers
• Things that are not services
• TODO as of Septemember 2012
• C++
• Terminal Fixes
• Windowing System
• Harddisk Drive Extras
• Toolchain
• Microkernal Readiness
• Services to Implement
• Old I/O goals
• I/O

Hold up! This page is very outdated and is being considered for removal. A more up-to-date view of our "TODO" list is available through our issue tracking.

---

Compositor functionality

• Alt+Tab keyboard window switching (with previews)
• Advertise window titles + icons
• New panel with window list
• Rewrite client-server protocol
  • New resize handshake
  • Only sane event path; fix window creation (fix "wait for reply" commands)
  • Better packet structs for events (window.left? ew!)

Japanese / i18n updates

• IME
• Support for multibyte/wide characters in rline and terminal linebuffer
• JIS keyboard mapping in the compositor (config file???)
  • Applications shouldn't care.
  • What to do with with special characters?

Multiarch / Ports

• Split out x86 / PC-specific bits into new arch/ directories
• Fix up uses of explicitly sized types where possible
• x86-64 port
  • Good for verification of multi-arch readiness
  • Much easier to develop for than ARM port
• ARMv6 port is underway

Cleanup and POSIX expansion

• Rewrite underlying process model
  • task/process split should match architecture dependency split
  • Better allocation management, process tracking
  • Cleanup is a pain right now - we need to stop eating processes that finish with parents that might want their statuses.
  • Sleeping, wait queues, and IO polling/select depend on some rewriting

New EXT2 drivers

• The EXT2 drivers are poorly written, they need a complete rewrite
• Allow for multiple EXT2 instances
• Operate on block devices rather than directly with ATA drives
  • Allows for effortless ramdisk implementation with same code
  • Will make MMC port a lot easier for RPi

TODO as of November, 2012

• force a build run

• Integrate Cairo into build toolchain

  • autogen means extra effort needed
  • required to ship with new compositor
  • this has been done for a while

• Fix static initializers in C++

  • The best method for this is probably going to be writing a dynamic loader, so...
  • this has been done for a while

• Write a dynamic loader

  • as of March 2014, finally starting to work on this

• Pretty much everything below

TODO for 0.4.0 Distribution Release

• User Interface
  • Graphical Login
  • More applications
• Working boot on real hardware
  • Confirmed working on Mini 9 with hard disk boot

TODO for 0.5.0

• CD support
  • Boot from "CD"
  • Distribute with Grub
  • CD image generator in-repo
• Stable Harddisk writes
  • Screenshot functionality
  • Attempt an installer?
  • ATA writes in general seem to be stable, work on EXT2 FS write support

Doc Revamp

• Get rid of old, outdated TeX/PDF manual
• Build a modern doxygen-powered documentation system for kernel functions for use by kernel developers.
  • Tried to do this, doxygen got very angry - might look at it again later?
• Also include doxygen documentation for included libraries (lib/graphics, etc.)

TODO for Microkernel Launch (0.5.0?)

• Replace ramdisks with ELF service executables
  • Boot with multiple modules = boot with multiple services.
  • vfs.srv, for example
  • replace "services" with "modules" and this is done...
• VFS as a service.
  • It would be super awesome to write this in a language that is more flexible.
  • Actual file system drivers as separate modules, or what?
• Service bindings
  • Essentially, a system call interface to discovering available services.
  • require_service(...) system call for usable errors when a service is missing?
    • Do module dependencies count? That's in...
• Deprecate old graphics applications
  • And rename the windowed versions.
• Environment variables
  • Support them in general
  • Push things like graphics parameters to environment variables
• Integrate service-based VFS into C library
  • Which probably means integrating shmem services into the C library
• Services in a separate ring
  • Compositor as a service
  • Compositor shmem names integrated with service discovery
• For VFS, need better IPC for cross-process read/write/info/readdir/etc. calls

Service Modules (aka "Services")

• vfs.srv The virtual file system server. (required to provide file system endpoints)
  • Keeping as core for now, possibly a module in the new revision? VFS is kinda important...
• ext2.srv Ext2 file system server. (provides /)
  • module
• ata.srv ATA disk access server. (provides /dev/hd*)
  • module
• compositor.srv The window compositing server. (provides shmem regions)
  • staying in userspace
• ps2_hid.srv The keyboard/mouse server. (provides /dev/input/ps2/*)
  • two modules: ps2kbd and ps2mouse, and they provide /dev/kbd and /dev/mouse respectively.
  • I'll look into the other dev nodes later...
• serial.srv UART serial communication server (provides /dev/ttyS*)
  • module: serial

Future Servers

• usb.srv Generic USB device server (provides /dev/input/usb/*)
• proc.srv Process information server (provides /dev/proc; uses lots of kernel bindings)
• net.srv Networking server (provides /dev/net)
• gfx.srv Block-access graphics server (provides /dev/fb*)
  • This is a module at the moment; might be abandoning a lot of these service ideas...

Things that are not services

• ELF support is not a service

TODO as of Septemember 2012

C++

• Build with C++ support

Terminal Fixes

• Mouse features; mouse support in windowed mode
• Tab completion in shell (this is mostly a shell-specific thing)

Windowing System

• Graphical Login Manager
• Finish GUI toolkit
• File manager app

Harddisk Drive Extras

• VFS support is still almost entirely non-existent
  • VFS works pretty nicely.
• Write support for EXT2 is still sketchy
  • This is still true as of March 2014 - actually, it's worse, write support is temporarily gone
• Still lacking fast read/write for IDE - needs more DMA!
  • Still true in 2014

Toolchain

• Finish GCC port
  • Finished in April/May, 2013
• Port ncurses/vim/etc.
  • This was compleated circa March 2013
  • Native development requires good tools.
  • Also port genext2fs.
• Directory support needs to be better integrated into the C library still
  • readdir and family have been integrated

Microkernal Readiness

• Deprecate ramdisks replaced with new ramdisk module, works better
  • Haven't used them in development in over a year
  • Not useful anywhere else due to their limiting sizes
  • As of March 2014, ramdisks have been reimplemented (again) in an actual ramdisk module
• implement module execution
  • Instead of loading a ramdisk, modules should be standard binaries
    • Actually, both are supported now, so that's done.
  • The binaries will be executed in a new "service mode"
    • This was isn't going to happen, at least not for kernel modules.
• Implement "servicespace"
  • Userspace, but at a different ring
  • Special access features, like extended port access
  • Higher priority scheduling

Services to Implement

• PCI Service
• Graphics Management Service
• Compositor as a service?
  • Technically, it already is.
• Virtual File System Service

Old I/O goals

I/O

• /dev file system
• /dev/fbN and /dev/ttyN for virtual framebuffer terminals and graphics
  • Framebuffers still need work to actually support mapping, but can provide size/address information via ioctl.
• /dev/ttyS0 for serial I/O
• SATA read/write drivers (/dev/sdaN)
• /dev/ramdisk (read-only)
  • Actually in /dev/ramN now
• EXT2 drivers should operate on a /dev/* file
  • Working on this with new ext2 drivers
    • The new driver is in the current strawberry head
  • Need better block device handling; *64() operations
    • Handled in the block device by doing extra reads, read-before-write operations
• Mounting of /dev/* files using a filesystem handler
  • A workaround is provided for /dev so that mounted nodes can be listed
  • A general solution is being worked on in the new VFS
• VFS tree