the "thread" update
**Threads**
OpenOS 1.6.4 brings the new thread library api. Documentation in our ocdoc wiki soon to come. An openos thread is an autonomous non-blocking detachable child process
* Autonomous: Threads asynchronously begin execution immediately after creation without needing to call resume. The thread proc may call coroutine.yield, but will resume on its own
* Non-Blocking: Threads can call computer.pullSignal (or any higher level wrapper such as event.pull, io.pull, etc) without blocking the main kernel process nor any other thread
* Detachable: By default, threads are scoped to the process in which they are created, i.e. their parent process. Any thread will block the parent process from closing unless:
A. The thread detaches from the parent process. In which case it does not block any process and runs independently, e.g. `t:detach()`
or
B. The parent process throws an exception or calls os.exit in which case all attached threads are killed, e.g. `os.exit()`
or
C. The thread is manually suspended, e.g. `t:suspend()`
**Command Redirection**
The other major improvement in this update is highly improved shell parsing for command substitution and io redirection. Some highlights include
* Can place before the command now, e.g. `2>/dev/null ./run_my_scripts.lua`
* Can properly use globbing or env vars as redirect targets, e.g. `./run_my_scripts.lua >$my_log_file`
* Fixed various bugs related to redirect and argument evaluation
**Memory**
50k free! As I love to do, this update reduces memory allocation needed to reach the shell prompt. The majority of the recent memory improvements are not just delaying allocation, but actual code cleanup and optimizations. Many of the changes are minor but they are numerous. With 1 stick of tier 1 RAM, openos reaches shell with 50k bytes free.
Changelog:
Bump version from 1.6.3 to 1.6.4
Cause grep to yield when taking too long
Remove additional gpu.bind calls during boot, reduces the number of screen resize calls
Move SHELL env var creation to /boot/94_shell.lua - This is an important user workflow fix to allow users to specify a custom SHELL without openos ever needing to load /bin/sh and its libraries
Fix event dispatch to not double call event timers in some scenarios
new /lib/thread.lua
fixed /lib/process from hiding exceptions in some scenarios
significant memory optimizations and code cleanup for /bin/sh
refactoring of command redirection, variable evaluation, glob expansion, and argument lists in /bin/sh
fixed term.setCursorBlink so that it properly waits for inf time for the next event when not blinking, rather than pulling every .5 seconds
fixed /bin/sh shell from losing exit_code when using || and && with multiple commands
fix /bin/lua error message when using os.exit
cherry picking serialization fix from 62471f7d320758bebb280666ed98388ea61cb4c8
Why a version change?
> This minor version change should not introduce breaking changes, but older code doing non standard things with process environments, `require` and the package library, shebang redirections, or direct calls to /bin/lua, may see some breakages. Also, `shell.resolve` has been fully reworked. This is a crucial and heavily used api. Thus, in case of any mistakes or bugs a version change will help pinpoint regressions. I've heavily tested the resolve code, but it is new code and deserves some bake time in the wild.
Changelog:
/bin/lua removes shebang line and calls load directly. Improves workflows that define custom environments, and loadfile no longer removes shebang lines
load: loaded code chunks now inherit the parent _ENV naturally, having a real-lua behavior
require: heavily optimized and made errors more natural and helpful, exposing more information about failure to require a library
/lib/pipes: commented out pipes.create which is a future feature to create event-boxed threads
process: more process crash text is dumped to the shell to help identify "out of memory" issues
shell.resolve: reworked and optimized! possibly breaking change: specifying an extension (ext) to `shell.resolve(name, ext)` will no longer return results to directories, but only files (if they exist). However, it was never the intent of this method to return results to directories when specifying an extension
various memory optimizations throughout. openos now allocates ~153k to boot to shell.
hostname in PS1 cannot be set by the init signal as that occurs before /etc/profile, which overrides PS1
Also, clean up some code to not try to change PS1, but just change HOSTNAME
Added `hostname --update` to update $HOSTNAME by reading /etc/hostname again
notable changes
### TTY
New /lib/tty library that provides the core readKeyboard and drawText functionality. /lib/term should still be used by user code and the tty api should be considered alpha. The intent in this separation of the code is to allow term api to remain unchanged and allow tty to adapt as improvements are made.
### delayloaded and now splitting
The delayloaded packaging system has been removed. The delayload system allowed for library object to appear loaded when in fact they were not. But this significantly increased boot time. The code now imploys "library splitting", saves tremendous memory, shortens boot time, and all libraries are fully usable even when partially loaded. In other words, it is all transparent to the user.
### boot memory cost
Numerous micro optimizations have been made. "boot to shell prompt" now idles at 156k allocated, or 40k free, up from ~26k free before this change. Yes, that's 14k less in this update.
