DRAFT: Staged scheduler #1026
DRAFT: Staged scheduler #1026
Conversation
Note that this commit is huge and it will be split before a review is requested.
|
|
||
| public function init(fd: i32) throws -> CollectedReader { | ||
| // NOTE: mut so codegen does not declare it as const u8 * | ||
| mut raw_buffer = libc::malloc(size: 4096) |
There was a problem hiding this comment.
this feels very awkward. does Jakt runtime not have AK::ByteBuffer?
There was a problem hiding this comment.
Probably it's there but I just didn't know of it. Anyway, this is nowhere near finished. The scheduler gets stuck after the first chunk of jobs finishes, panics sometimes (maybe it's the queue implementation, I don't know), fails to properly collect stdout/stderr over the pipes (thus marking tests as failed because it couldn't find the error). Plus, the fork()/dup2()/exec() has to be replaced with an API that implements the needed posix_spawn* functions, since the Windows approach goes through a 'builder pattern' in a similar way that posix_spawn functions are meant to be used.
There was a problem hiding this comment.
I'm going to start by throwing out the pipes and just create the .err/.out/.cpp files directly and redirect the output to them, saving just their handle through Jakt::File. This will just need a little add-on to File to get fileno functionality, and will remove the polling of buffers that is not even working properly.
This is an attempt to remove the
run-one.shscript by integrating thedifferent stages of compiling a test binary into the Jakttest scheduler.
EDIT: Managed to work around #1025. Now scheduler doesn't work correctly, but at least I can compile it.