Age | Commit message (Collapse) | Author |
|
to be performed from the task manager.
|
|
differently.
|
|
|
|
|
|
class, but, meh.
|
|
stick on another one's thread, if needed. (and it is)
|
|
TaskScheduler.
The idea is that we need to support multiple task managers from multiple threads.
So that revamp means we now should be able to support that, except the TaskScheduler needs to implement a round robin system, to distribute tasks across multiple task managers. But at least, the fundamental redesign to permit this is here.
|
|
|
|
from Handle...
|
|
|
|
|
|
|
|
|
|
a file descriptor can do less than being asked.
|
|
exception from a .h file...
|
|
|
|
fleshed up, and writing the actual server should now be sort of straightforward.
|
|
|
|
contains a potential task switch.
|
|
|
|
come in handy.
|
|
still needs coding, and won't work as it. And it needs a unit test.
|
|
|
|
|
|
|
|
|
|
unit test.
|
|
|
|
|
|
|
|
at least the Input class.
|
|
method to Handles. Also preparing the inclusion of Lua by adding LuaJIT.
|
|
turn that off.
|
|
functions.
Switching out libcoro for the native Fibers interface for Win32.
Switching out the asm version for the ucontext version for Linux.
|
|
instead of htons - which should be fundamentally the same, but, *shrug*)
|
|
-) Using clang instead of gcc for the native target (*sigh*)
-) Fixing libeio (*sigh*)
-) "Fixing" the connect() call in Sockets because Darwin is stupid (*SIIIIIIIGH*)
Although the Threads unit test isn't working properly.
|
|
Doesn't work though; I'm guessing IPv6 isn't really mingw32 thing. At least, it's not WinXP's. Will try later to make something that switches between IPv4 and IPv6 intelligently.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
-) The IO class needed an assignment operator, as well as the ability to NOT contain a Handle.
-) The Task Manager wasn't properly starting up pending tasks that got added during startup.
|
|
say it's okay to get a EAgain exception, if the operation is interruptible.
|
|
-) setPreemptible now returns the previous state
-) Events may now not have an owner
-) Events may now have a callback (even though it still needs an owner to get into the ev_loop, for these kind of watchers...)
TODO:
-) think about a way to start a watcher without requireing a Task
-) add a few more things in the Tasks unit tests
|
|
Hence removing a few now unnecessary asserts.
|
|
|
|
mechanism to re-arm an event.
|
|
|