@Dennis isitblockedinrussia.com/?host=randomstuffonline.com Uh oh. Tio.run isn't blocked anymore, but my website is

Since 2013 and for 59 different reasons...

@ngn I'd rather wait a bit for the official fix, in case they decide to go with something incompatible. But if it takes longer, sure.

@Dennis i don't expect it to take long

Not the first time TIO's non-interactivity has been an issue. PTYs are impossible inside the sandbox, but I might fool enough interpreters by preloading my own version of isatty.

@Dennis i guess most language authors are unaware of rlwrap

there's no good reason to do anything fancier than read from stdin and write to stdout, in my opinion

the only trouble is with the prompt - rlwrap has no way to know when the execution of a silent comand has finished

On compiled languages, are the "Compiler Arguments" appended or prepended?

(relative to the default args)

Depends. I try to put them after stuff that you should be able to override, but before stuff you should not.

@Dennis I just sent my GitHub student DO credit code :)

@Stephen Thank you. :) Unfortunately, the field I entered the last code in no longer exists. It looks like you're only allowed to enter one promo code per account.

@Dennis does this mean VLANS? :o

@Downgoat Sort of. You can configure the firewall to only accept connections from your own droplets.

(Making an exception for port 22 would be a good idea.)

@Dennis Can you block outward traffic? Might allow sockets

Outward traffic is already blocked for runner. (DO firewall can't do that.) Two processes from different requests shouldn't be able to interfere with each other though.

How does TIO contain files created in /tmp?

@ngn readline tho

For user requests, /tmp is mounted via seunshare, just like /home/runner.

(silly) Idea: fork socket library, implement it through fifos, ldpreload it

LD_PRELOAD is not a security measure. You can simply unset the variable.

Not that I'd re-implement sockets if it was. :P

@Dennis Well, if you unset it, the sockets would just stop working

.oO(what socket library is it anyway)

Only if I somehow uninstall the real version from /lib64.

Wouldn't selinux go o no real sockets

Wait, no, I think those are system calls.

Oh, that's what you mean.

Yeah, that could work. It's probably way too much effort, but that could work.

sys/socket.h isn't very big but there's probably other stuff on other places

I think everything else would be based on this.

@Dennis kona was fixed

@Pavel what do you mean? something wrong with using libreadline?

@ngn Try it online!

@Dennis yay! thanks :)

Thank you for the bug reports.

@Dennis: I changed Alchemist slightly, can you please pull the latest version?

@BMO Done.

That was fast, thank you kindly :)

@ngn No it's great, I was saying you should use libreadline instead of making your users use rlwrap

@Pavel i'm a minimalist. i use no libs :)

@Dennis - found and fixed a bug in C: statements in psPILOT. Pull, please?

@Pavel seriously though, this makes it harder to use the language non-interactively, like in tio

@JeffZeitlin Done.

благодарю вас, señor!

I made some experiments on dev.tio.run that should make resource distribution a bit fairer. Fair enough that a fork bombs should barely have a noticeable impact on server performance. It would be great if y'all could play a bit with the test server to check that everything's still working.

(Try not to use it in PPCG answers.)

While working on the backend, I was also questioning the need for a third file descriptor. If TIO warnings/errors go simply in the debug drawer, with a few modifications, we could get rid of that awful message area at the bottom of the page. Example output for requests that: time out, produce too much output, request a non-existing language

It's probably not noticeable enough if the page is too long.

In order to get "The output exceeded 128 KiB and was truncated." you have to scroll pretty far

OTOH, if you don't scroll to the bottom you don't care that it was truncated in the first place

Sure, but if you're not reading the entire output, you probably don't care.

ninja'd

I really like populating the search field on invalid language

I was wondering if this meant more concurrent Q# processes but it just made the error message more interesting: paste.fedoraproject.org/paste/97gEqpfImMxuSZemNh3duw

What were you trying to do?

Run more than one Q# at the same time

Which program? Two concurrent Hello Worlds seem to be working.

Oh no two seems to be fine

The third one chocked

The process limit shouldn't have that effect.

There are 64 processes per request, 512 per IP.

So if 64 is enough for one request, that should allow you to run 8.

Maybe it needs too much RAM.

That would make sense

No, RAM usage seems to be low.

¯\_(ツ)_/¯

> Exception of type 'System.OutOfMemoryException' was thrown.

mystery solved

That doesn't make sense. It doesn't even start to use swap.

It works without cgroups, so it's definitely not RAM.

@Pavel The culprit was some file in /etc I had been experimenting with. Successfully ran 8 concurrent Hello Worlds on a single arena.

They almost timed out, but they worked.

Maybe it just throws OutOfMemoryException when allocation fails for any reason

Maybe. I'll reboot the test server now to check if the settings stick.

Interesting how convenient Q# is for stress testing though :P

And Hello World is enough. :P

Ah, there we go.

Using 5 GiB of RAM on a 2 GiB machine.

Reminds me of my last position when we used AWS, we would sometimes find servers reporting 300-400 load average in htop

Well, load is about of much stuff there is to do, not about how much stuff is being done. :P

I wonder how much slower an actual swapfile would be.

For this, probably not by much, as the memory is written sequentially and never read...

A tiny bit faster, actually.

@Dennis Zram?

Yes.

Let's bench pseudo-random writes.

We've had zram for forever though. Is it different now?

No, just checking if cgroups are causing problems.

For pseudo-random writes, regular swap is 5x slower than zram.