Radiation Hardening

Aiden Chow

00:49

ugh i just test my bot against burrowing wheel and it only survive 5 deletions

Number Basher

02:12

Wem

@mousetail update reminder

Aiden Chow

05:27

posted an ascii piet bot :D

probably like 10th place is my estimation

@mousetail u already have python and c installed right, i dont have to specify the download instruction for those?

mousetail

06:03

@AidenChow Yes

Aiden Chow

@mousetail k good to know

mousetail

As long as you have a compatible version of both

If you need a older python version or clang or something you need to specify instructions

Aiden Chow

@mousetail i think latest version of python is fine

not sure about c tho

npiet is from 5 years ago

maybee u can do a quick test to see if it works?

mousetail

C doesn't change that fast

Aiden Chow

@mousetail lol idk, i dont use c

mousetail

06:15

C is over 50 years old

Aiden Chow

@mousetail oh really? i didnt know that lol

does my new bot work btw?

Number Basher

hi

Aiden Chow

@NumberBasher hi!

@NumberBasher i have no idea what that is

mousetail

06:44

Unlikely

Aiden Chow

^

it needs to be voted on

also @mousetail did u get my piet bot to work or no

Number Basher

Any suggestions on the lang?

mousetail

@AidenChow Not yet

Aiden Chow

@NumberBasher probably ask in tnb or smth, not in this chatroom

@mousetail oh rip, whats the error?

mousetail

I'm trying to find it

It dumps all errors in a 10 MB giant log file

Number Basher

06:47

@AidenChow seems like nobody is in the chatroom https://chat.stackexchange.com/rooms/240/the-nineteenth-byte

@mousetail good luck finding it

mousetail

@NumberBasher Doesn't mean you can use unrelated chatrooms

Aiden Chow

^

we tryna talk about the koth rn

Number Basher

~~True, then everybody go to TNB to see my lang~~

@AidenChow ok

mousetail

@AidenChow This seems to be the error:

cannot read from `src.png'; reason: unknown PPM format

Aiden Chow

bruh it requires PPM ???

mousetail

06:48

IDK what that means

Aiden Chow

bruuuuh the specs literally says it can accept png files

so confused

mousetail

Maybe it needs more dependencies to support png

For example if you just ./configure python in a blank container it won't have SSL support

Aiden Chow

have no idea what ssl is but ok

mousetail

@Bubbler

Traceback (most recent call last):
  File "//golfish/src/golfish.py", line 21, in <module>
    from fractions import gcd
ImportError: cannot import name 'gcd' from 'fractions' (/usr/lib/python3.10/fractions.py)
Traceback (most recent call last):
  File "//golfish/src/golfish.py", line 21, in <module>
    from fractions import gcd
ImportError: cannot import name 'gcd' from 'fractions' (/usr/lib/python3.10/fractions.py)

Aiden Chow

bro i need to ask bubbler about this error, he basically gave me the install and run instructions and i just copy pasted

idk how to convert it to smth npiet can take

some flag maybe

mousetail

06:53

I guess both new submissions require input from bubbler then. Thief works properly at least

Number Basher

Bubbler, the actual bot

mousetail

Bubbles implies being under water which is typically bad for bots

Number Basher

~~Bubbler (someone who bubbles) implies having to bubble, which implied being under water, which implies a bad bot. Therefore, Bubbler is bad. Good logic.~~

Aiden Chow

seems like the important part of the npiet code is this:

#ifdef HAVE_GD_H
fprintf (stderr, " ");
#else
fprintf (stderr, "out ");
#endif
fprintf (stderr, "GD support, with");
#ifdef HAVE_GIF_LIB_H
fprintf (stderr, " ");
#else
fprintf (stderr, "out ");
#endif
fprintf (stderr, "GIF support, with");
#ifdef HAVE_PNG_H
fprintf (stderr, " ");
#else
fprintf (stderr, "out ");
#endif
fprintf (stderr, "PNG support)\n\n");

it seems like it has some if statements to check for png and gif support

problem is i cant read c code that well so idk the conditions that are needed

mousetail

@AidenChow You probably need to check the makefile rather than the code itself

It will decide which defines are enabled

Or check the configure file, which will generate the makefile

Aiden Chow

07:08

@mousetail ummmm its like hella long

i cant make sense of any of it

mousetail

Important line from the readme:

if compiled with gd-lib and png-lib support (if avail), graphical
trace output can be created by npiet - great fun ;-)

Aiden Chow

@mousetail ya but we dont want graphical trace output

mousetail

True, but it implies png-lib is a seperate thing

Aiden Chow

ok well bruh wtf is png-lib, gotta go searching for that now

mousetail

Yea, I've used libpng (github.com/libpng/libpng) but not sure if it's the same or different from png-lib

Aiden Chow

07:15

lol try it out, it might be

Number Basher

probably not the same, considering there's too many libraries

Aiden Chow

Libpng

libpng is the official Portable Network Graphics (PNG) reference library (originally called pnglib). It is a platform-independent library that contains C functions for handling PNG images. It supports almost all of PNG's features, is extensible, and has been widely used and tested for over 23 years. libpng is dependent on zlib for data compression and decompression routines. libpng is released under the libpng license, a permissive free software licence, and is free software. It is frequently used in both free and proprietary software, either directly or through the use of a higher level image...

> (originally called pnglib)

so im thinking its probably the same thing

Number Basher

hmm...

Bubbler

@mousetail For this error you can just edit from fractions import gcd into from math import gcd

Aiden Chow

07:31

@NumberBasher why tf u posting this link here, this is the chatroom for the koth

Aiden Chow

07:44

@mousetail did it work with libpng?

Number Basher

I guess mousetail is not here

mousetail

08:29

@AidenChow How do I install it

@Bubbler Can you please fork the repo if you want to use a modified version

Aiden Chow

08:46

@mousetail idk man :(

never heard of libpng until now

probably smth like git clone https://github.com/libpng/libpng.git libpng ?

and u need to run github.com/libpng/libpng/blob/main/autogen.sh after install it seem like

oh wait libpng requires some other thing called zlib

damn

mousetail

Can't you install it with apt?

Aiden Chow

09:11

@mousetail maybe, i searching on ask ubuntu rn

ok well i found smth

sudo apt-get install libpng-dev

and for the dependency, zlib:

sudo apt-get install zlib1g-dev

mousetail

Um now the entire controller broke

Traceback (most recent call last):
  File "/home/mousetail/SideProjects/Radiation-Hardened-Koth/src/main.py", line 202, in <module>
    asyncio.run(combine_all())
  File "/usr/lib/python3.10/asyncio/runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "/usr/lib/python3.10/asyncio/base_events.py", line 646, in run_until_complete
    return future.result()
  File "/home/mousetail/SideProjects/Radiation-Hardened-Koth/src/main.py", line 194, in combine_all
    results = await asyncio.gather(*(run_all(i) for i in range(0, 6)))

Aiden Chow

bro idek

u can read those errors better than i can

mousetail

I don't know how to recover from this other than restarting the container

Aiden Chow

wait so u ran the sudo thingy and now the whole controller is busted?

mousetail

I don't think it's specifically related to your program, one bot is generating a very large amount of output on one line

which exceeds the line length limit of pythons readline function

a very long limit

Aiden Chow

09:23

i could see that being my piet program going haywire

specifically, if enough chars get deleted, the code wont halt and keep outputting

or maybe not, idk

mousetail

I added | head -c1000 to the end

should maybe fix it

Aiden Chow

so my piet bot is working now?

@mousetail wat that do

mousetail

Limit to 1000 bytes

Aiden Chow

if the output is more than like 5 chars then its guaranteed not valid

mousetail

some bots output multiple extra newlines

I decided to allow it

Aiden Chow

09:26

oh ok

mousetail

I guess you could abuse this by adding 1000 line breaks after your answer instead of exiting

but I doubt it would benefit you

Aiden Chow

i could see myself utilizing that to "halt" my bot instead of actually halting it normally. halting a program in piet is always such a pain

mousetail

I'd consider that exploiting the controller

Your bot does quite ok now

Aiden Chow

ay top 10 nice

also wow thief just stole 1st from distributed dmg

amazing

mousetail

Exteremly impressive to beat a esolang with a practlang

their difference is extremly small, I'd consider them equal

Aiden Chow

09:34

rookie drop down to 4th rip

mousetail

So many good bots now playing that top4 is impressive

Aiden Chow

i feel like my rookie bot and comment skipper are pretty equal, sometimes my bot is above but other times comment skipper is above, its just kinda by chance i feel like

mousetail

Thief would be in trouble if aes fixes the unicode issues for burrows wheel and almost rare

but right now they can't really handle invalid unicode so it goes thief a advantage

Aiden Chow

aes? whos that

oh u mean ais lol

mousetail

oops yes sorry

I am way out of my depth writing a controller for something like this. I'm very concerned the ranking is influenced by bugs

Aiden Chow

09:49

@mousetail nah u hella good at coding, dw

also i updated my piet bot becuz i noticed a vulnerability, if u have could u run it again

Number Basher

Happy Birthday

Bubbler

10:58

mousetail

@Bubbler Yes that's your bot

Number Basher

I think because any remove of the ` characters destroy your bot

Bubbler

Err... I was pretty sure it'd survive backtick deletion but idk

Number Basher

idk, i dont know the lang

did you test your bot

maybe the bot had problems, not the deletion

the deletion just happened to make the problem visible

mousetail

@Bubbler No it dies immediatly regardless of characters

since the controller can't start up

Number Basher

11:09

erm

mousetail

What is happening is it's attacked there, then it dies. But it was already dead. So the highlight is a bit useless

Number Basher

11:31

This room is more active than TNB lol

mousetail

11:44

@NumberBasher TBF we where the majority of the talking in that room before they split

Number Basher

very true

mousetail

13:49

** (FunctionClauseError) no function clause matching in String.trim_trailing/2

    The following arguments were given to String.trim_trailing/2:

        # 1
        :eof

        # 2
        "\n"

    (elixir 1.12.2) lib/string.ex:1215: String.trim_trailing/2
    (osabie 1.0.1) lib/reading/input.ex:8: Reading.InputHandler.parse_multiline_string/1
    (osabie 1.0.1) lib/reading/input.ex:8: Reading.InputHandler.parse_multiline_string/1
    (osabie 1.0.1) lib/reading/input.ex:29: Reading.InputHandler.read_input/0

Bubbler

I deleted my golfish bot, maybe the interpreter itself has something wrong with nonascii i/o

mousetail

@Bubbler It's just the GCD error

you can fork it to fix the issue

ais523

14:05

@mousetail in order to clear LC_ALL, the variable has to be unset entirely; easiest way to do that is probably to use the env command as a prefix to the Jelly command

mousetail

@ais523 Won't that clear LANG too though?

ais523

so something like env -i LANG=C.ISO-8859-1 jelly f programfile

env can set variables as well as clearing them

(this should also make it possible to have environment variables set inside timeout

mousetail

Ok nice

I'll probably replace all my environment variable settings with this then

To not polute the global scope

ais523

$ dpkg -S /usr/bin/env
coreutils: /usr/bin/env

(in case it isn't installed)

but coreutils seems pretty important

mousetail

Yea I needed coreutils for base64

ais523

14:09

IIRC there are some Linux distros with only two programs installed in the "usual" places (/usr/bin and /bin) and env is one of them, I forget what the other one is

the reason is that env is always in /usr/bin so that you can use it to run programs via a hardcoded path, in cases where the calling program doesn't know how to do a path lookup

so if you want to run, say, Perl no matter where it happens to be on this OS, you can write /usr/bin/env perl – env is always in /usr/bin and it knows how to look for perl

mousetail

Yea I've seen programs use #!/usr/bin/env python but never really thought about what it meant

ais523

it will only look on $PATH though, so if the program is somewhere weird you still need to give the full path

also, wow, some mod must have put in a lot of effort working out which messages to move

when this chat room was created

Number Basher

true, @emanresuA did a lot of work

ais523

@mousetail reason we're using ISO-8859-1 (or "Latin-1" for short) is that its codepoints are in order in Unicode – in Latin-1, the byte with value x maps onto the Unicode codepoint with value x, e.g. an 0xA0 byte is interpreted as the Unicode character with codepoint 0xA0, so all 256 possible bytes map on to 256 different Unicode characters in a way that makes codepoint math work

Kevin Cruijssen

@mousetail What does the input look like when you get that error? If I look at the 05AB1E source code, it seems to check for the trailing """ for multiline inputs. Do you also have a leading """? And which of the programs from the other answers was used as input?

mousetail

14:22

@KevinCruijssen Give me a moment I can check. The logfile is huge so it's a bit of searching

mousetail

14:43

Ok running the controller now consistently BSODs my PC

F not sure how to run it now

Sometimes it just freezes my cursor for 10 minutes

ais523

try reverting changes you made recently?

mousetail

Only change I made was adding env to all commands that use environment variables

ais523

are you using the Windows or Linux version of Docker? Probably running the Linux version through WSL would be less likely to cause BSODs

BSODs are so hard to debug on Windows, unfortunately

(the last time I got the equivalent on Linux I did manage to debug it to the extent of finding a vlable workaround, although it's a pain – with no working kernel, you can't save the debug information anywhere nor take a screenshot, so I had to copy it all down by hand using pencil and paper)

mousetail

15:22

I was on linux. By BSOD I meant my window manager crashed

This was my work computer, but now I'm at home on my windows laptop which was typically must worse at running the challenge

Aiden Chow

@mousetail i dont think u updated my piet bot yet in the leaderboard

also my bot is dropping places, sad

might need to do another revision for the newer bots

mousetail

Yea as expected the unicode fix made all ais's bots OP

Aiden Chow

15:45

@mousetail ya ais just too good

also happy bday!!!!

Oskar Skog

16:12

@mousetail I noticed Thief was being mostly killed by simple first character deleters, so I hardened it a bit more.

Well, maybe not simple. Some of them are pretty hard to kill.

Aiden Chow

@mousetail you accidentally linked my revised piet bot with kevins bot protect and shoot randomly

ais523

16:49

@OskarSkog fwiw, the greatest number of deletions I've been able to harden a first-character-deletion bot against is 27

(i.e. the program can withstand any 27 characters being deleted)

but that would die in 30 rounds to a first-character-deleter itself, so I'm not convinced it would do very well

I think that part of the reason Distributed Damage works so well is that it's 192-hardened against deletions from the start, so if it ends up in front of a first-character-deleter, it has time to kill pretty much everything

mousetail

I'm interested to see how well a last character deleter would work

ais523

in general, the way the hill works is pretty weird when there's a mix of defensive and offensive programs – if an offensive program ends up being attacked by a defensive program, it gets a huge advantage, whereas if it ends up attacking a defensive program, it gets a huge disadvantage

mousetail

Still need to find a suitable language to write it in

ais523

for my programs: it would pretty much kill Comment Skipper in 5 (by causing it to always output the index as 0, killing it unless it happens to want to delete the first character anyway)

Distributed Damage survives 192 last character deletions as long as the first character is intact

The Burrowing Wheel can survive 3 last character deletions (its chaff is intentionally front-loaded due to the hill composition)

and Almost Rare can survive 27 I think – I chaffed that area because it was an early bot and I wasn't sure what the hill would look like

last character deleters are hard to write because you need a language with EOF handling, and that can express arbitrary numbers in decimal

mousetail

I don't think such a bot would be top tier against but be strong in some specific situations so maybe be top 50%

If hardened enough

ais523

17:00

A Pear Tree might be a good language for that

its implicit input handles byte counting and EOF the way you'd want, and has a "length" builtin that lets you get at the index of the last character easily

so unhardened it'd be something like #r3TQJ\n/.$/;$_=length."$&"

and that gives you a lot of room for redundant copies and chaff

(the #r3TQJ start is a good tip for testing A Pear Tree programs – it's an easily typable checksum that checksums itself rather than the rest of the program, so it gets you around the a partridge error without limiting what you can write later on)

oh, needs to be /.$/s I think, so that it matches newlines

and " $&" not just "$&"

mousetail

That would probably live quite long at least

ais523

it's a pity that length has a fairly long name

oh, and need to subtract 1 from the length because we need to return 0-indexed positions

#r3TQJ\n/.$/s;$_=(length-1)." $&"

ugh, needs to be (length)-1

otherwise it takes the length of the number -1

not a great time to hit a syntax ambiguity :-(

mousetail

Thanks, but I'd actually like to find some less specialized language that happens to ignore invalid lines

Oskar Skog

What about Perl? Are the memes true enough?

mousetail

A pear tree is basically pearl with checksums

what memes?

Oskar Skog

17:15

@mousetail: That anything is a valid Perl program

ais523

it isn't true, keyboard mashing is almost always invalid in Perl

unless it consists pretty much entirely of letters/digits/semicolons

mousetail

The entire point would be that invalid code would do nothing

ais523

the "83% of paint splatters are valid Perl programs" results are generally obtained by tweaking the OCR to not output spaces or punctuation

mousetail

Not do something else than what I want

ais523

you might want to check the polyglot for ideas? it's full of languages which ignore anything they don't understand

mousetail

17:17

Yep I'll check it out

ais523

but I don't know how many of them can read to EOF

mousetail

You don't want to know how many whitespace interpreters I had to try before I found one that could read EOF

ais523

I had a similar issue with brainfuck once, trying to find an interpreter that could distinguish EOF from all 256 valid bytes

I got there in the end but I had to go through so many interpreters to get it to work

mousetail

It's one thing language designers never think about

just have to find a interpreter bugged in a way that's usable

ais523

I think about it

Oskar Skog

17:20

I happen to have an unreleased tailstrike program, but it's yet another Python and I don't think it'll live very long.

ais523

but in many cases it has nothing to do with the goals of the language

e.g. TC-proving languages generally don't need I/O at all because you don't need I/O for Turing-completeness

and with low-level languages intended for translating between languages, I often intentionally don't have working EOF handling because the languages I'm translating into also don't have working EOF handling, so if I mandated a working EOF, the translation wouldn't be possible

mousetail

Makes sense, it's rarely useful but when you need it there is no alternative

ais523

EOF is also hard to define from the point of view of inputting into a program, too

UNIX/Linux doesn't have an actual EOF condition, just a "no more data available right now" condition; it's possible to read a file, get "no more data", read it again, and get actual data again

(this can happen if the file is appended to between the reads, or if it's a terminal and the user sent an EOF manually with Ctrl-D)

mousetail

That's good enough though right? You use EOF when getting input from a file, other things can send a EOF only when mimicking a file

Yes you can type after cntrl-D but I wouldn't expect a program to deal with that in any sensical way

ais523

it's generally good enough because few programs read after EOF to see if the file has grown

(less does, incidentally; if you try to scroll beyond EOF, it'll send another read system call at the file to see if it's become any longer)

Oskar Skog

17:37

Well, if you reach the end of the content of the file as it currently is, did you or did you not reach the end.

ais523

I'm the sort of person who thinks that there's a conceptual problem with the concept of "end of a file"

e.g. /var/log/syslog I would not consider to conceptually have an end, just a point it's reached so far

in many older OSes, files always had to be a multiple of, e.g., 512 bytes long – the "conceptual end of file" was signalled in-band

Oskar Skog

Huh, I didn't think it'd be that strong. tio.run/##1ZJBS8MwFMfv71OElkGLTpa2YwzxC0zmSYogooUVUUo7Vi/…

thejonymyster

19:37

Assuming there was a program that would cause any program to fail but itself would always fail, what would happen to it scoring wise?

"cause any program to fail" meaning that if a program tried to remove a byte from it, itd fail, and "itself would always fail" meaning that it never outputs validly

why are words so hard

like i know the empty string is automatically disqualified, but imagining it weren't, what would it score?

challenge body says it'd be "unbeatable" but that's only true if it goes last :P

(right?)

mousetail

19:54

@thejonymyster it would earn 0 points, since none of the failers would be atributed to itself

See this comment chain:

If a bot causes its attacker to die by including null bytes or invalid encoding, does that count as a kill? –
Oskar Skog
8 hours ago
1
@OskarSkog No, since there is no way to tell. The kill is always awarded to the last attacker, not the victim. There would be no way to tell if the previous attacked or the victim caused the crash except maybe for the very first but I don't want to special case that.

thejonymyster

20:26

ohh right, thanks :-)

Oskar Skog

Why does Python 3 have to be so bloody annoying, it won't accept ISO-8859-1 no matter how try.

Aiden Chow

20:50

@mousetail before u update the leaderboard, u gotta fix the switch up with my bot and kevins bot. u accidentally put my bots code into the other bot

mousetail

Yes it's been fixed in the repo

still need to figure out how to run the controller without crashing my PC

Aiden Chow

@mousetail why whats happening

mousetail

Burrows Wheel is making me run out of RAM

Aiden Chow

@mousetail oh damn riiipp

mousetail

When it was bugged it died pretty fast but now that it's fixed it's a problem

ais523

20:54

@mousetail it creates a multidimensional 256×256×256 array, and I'm not quite sure how much overhead Python has for those things

mousetail

I don't know what's happening exactly btw, this is just my theory

ais523

this is what happens when writing algorithms tersely (to shrink the core) is more important than writing them efficiently

mousetail

256x256x256 is a good amount

ais523

it should be well within the capabilities of modern computers, though

2^24, so 16 MiB of data + however much overhead Python has

Perl has a ridiculous amount of overhead for this sort of thing (factor of 10-100, I think); I don't know whether Python is the same

mousetail

How did you get 16MB, I got 134

Wait MiB

Yea in python even integers are stored by reference

ais523

20:57

256 is 2^8, so 256^3 is 2^24, = 2^4 MiB

mousetail

so you can count about 16 bytes each

ais523

ah, so this is getting up into the 256MiB – 1GiB range

mousetail

It should be able to be handled but I'm running multiple processes at once

ais523

depending on how much is needed to specify the data type

mousetail

Time to implement a memory optimized Jelly interpreter

ais523

20:58

although, hmm, I'm not actually sure that the entire array is created at once

the way Jelly implements €, I would have thought that only one 256×256 plane of the array is being processed at a time, then deallocated before moving onto the next one

mousetail

Python never (or extremly rarely) frees any memory. It will reuse memory from garbage collected objects but never release it back to the OS in normal circumstances.

Anyway, I'll look into it tomorrow. Probably just need to reduce the number of concurrent processes

ais523

ooh, theory: Burrowing Wheel itself has a reasonable memory usage, but the memory usage explodes when it gets damaged

I just tested: memory usage is about 3 MiB in an undamaged run

but damage could easily make it go much larger

it might make sense to stick a ulimit on there in order to prevent excessive memory usage?

ulimit -m 1048576 will limit programs to 1 GiB of memory, which is low enough to prevent runaway swapping but high enough to allow any likely program

(note: this form of ulimit is not reversible within the current shell, you have to exit the shell and restart it to remove the limit)

Oskar Skog

I've found out that using LC_ALL with Python 3 neither helps or is necessary. I've updated only the description of Bastarf (no code change), made major changes to Tailstrike, and updated Thief to not quickly die from a tailstrike.

ais523

I did realise that there was a cleverer/more interesting way to write Distributed Damage, but it would be some effort

the idea is to write the same program four different ways, rather than literally the same program four times

to make it harder for things like Almost Rare and Thief to shut it down

I'm also wondering whether I should harden The Burrowing Wheel against the closing ” getting deleted, because it's a very vulnerable point and some programs can identify it

Oskar Skog

I think LC_ALL was messing with the Jelly answers, and that only one of my answers was using it without even needing it. Searching for "LC_ALL" on the page only turns up two hits.

@ais523 I think that would be a good idea, Thief hits it immediately.

ais523

21:17

I'm not sure how much it makes sense to update programs once they're posted

I did put lots of copies of the other vulnerable characters in the chaff to specifically guard against this sort of thing, but forgot that ” was also vulnerable

err, ”

it is so hard to tell those quote marks apart and I may have just written the same one twice

Transcript for