weird I am also now getting a lot of broken pipe issues

oh god a heisenbug has appeared

@Mego Seems that thiis Broken Pipe error is a celery bug that is still open. people say to consider: github.com/closeio/tasktiger

yeah, not using celery leads to no problem at all

It looks like this has to do when celery runs the nodejs subprocess on another thread

it must not be able to connect to the pipe leading to the error

checking the logs it seems to be the case

not sure if we can have celery use 1 instance per thread

@Mego yes. That would mean minimum of 200ms to render a string and wouldn’t scale well

@Mego doesn’t exist. you would have to rewrite syntax highlighter, katex, gfm markdown

Also would mean sacrificing client side previews.

Aha found solution

:D it works

@Mego does celery automatically cache or do we need to set it up that way

and how does it cache?

oh ok

is there a way to manually expire a cached call?

@Mego syntax highlighting client-side would be pretty big dependency as we'd have to link highlighting grammars for each language, parse DOM, etc.

but I fix rendering issue :D

github.com/Mego/PPCG-v2/pull/35

wait crap have some debugging garbage still in there

@Mego ¯\_(ツ)_/¯ seems to fix it for me

no errors on different threads

@Mego It's not impossible but it'll mean reinventing a lot of existing code and no live code preview

assuming we manage to get same things with python. we would either have to make sure we have a JS implementation that's exact same or we send a new rendering request to server every second to make a preview

@Mego rendering output like syntax highlighting and math should be the same

@Mego yes but MathJax, highlightjs, prismjs, KaTeX, etc are all JS so somewhere you’ll be spawning proc if you use a binding

Either way it seems to be working now so I’d rather not spend too much time fixing what’s not broken

@Downgoat I'd argue it makes sense to use a syntax highlighter that's much less buggy than Stack Exchange's is

especially for a competitive programming site, where people are often doing weird things with language syntax

the best syntax highlighter I've seen so far has been in the editor Kate, it's capable of coping with some very complex programs that many highlighters fail on; there are probably equally good ones though

there's emscripten → asm.js

although that's much harder to get working

(as opposed to webassembly, which many older browsers don't support, asm.js should work pretty much anywhere because it's a subset of JS)

I can see an argument for sticking to Stack Exchange's markdown syntax, although it's not a very good dialect (it's very hard to escape special characters, its syntax highlighting syntax is very verbose, and some of its features interact badly with other features, e.g. >! is a nightmare to use for anything but plaintext)

@ais523 exactly

@ais523 hopefully we are compatible with SE MD so we can easily translate posts.

been reading trough this, and I'd say that no automatic testing is one of the things that I like about PPCG. Some languages can be problematic - Processing, unless explicitly told to, after running code, will leave a window open (timing out and auto-scorers), and theoretical brute-force answers can't compete. I'd suggest that automatic testing should be optional, and if not enabled on the answer, it should allow people to vote that they've verified it working (or not working).

@dzaima I agree. And submission can be in many forms - functions, lambdas, programs, etc. Each can output by returning, exit code, or printing. This makes it problematic to perform automatic testing. And, with flexible I/O, it would be hard to be able to know what to supply to the program, and what format the output will be in.

@FlipTack in this message ais523 suggested that there could be a list of I/O methods (that the answer could choose from I guess), so that's not a big problem.

@dzaima I’ve been thinking about this but I’m not sure how we can have both lenient IO and auto verify it

We could allow challenge authors to write test scripts

@Downgoat I think the trick is to have each language specify how it handles all common types of IO in advance

e.g. C might want arrays specified as a length followed by each element

then the challenge specifies what inputs there are and what types they have, ditto outputs

and the automated tester generates an appropriate wrapper for each language

@ais523 what about things like "challenge: calculate size of pool. output length, width, height in whatever order you want consistent order."

this would prevent cheating via choosing an input format per-program to make it easier

@Downgoat I'm not convinced that that sort of flexibility makes a golfing site better (in fact, it's influenced the behaviour of many PPCG-user-created golfing languages to be bad at reordering outputs)

hm ok

I guess we can see how it goes and we can change the validation system if it doesn't work out

likewise, for truthy/falsey questions, I think having a consistent wrapper per language that knows what true and false are in that language makes more sense than the workarounds we normally use

Also I'm not sure on how to do voting

that said, I can certainly see that someone might disagree with me on this!

you've said that it should be based on tags IIRC

I don't like having "one score to rule them all" on posts because some posts are really hard to compare

So would that be like Question Categories or like generic 'voting categories' e.g. "Funny", "Interesting"

this challenge is my go-to one for looking at quirks with SE's voting system

the Haskell and Unary answers (the top two) just aren't comparable at all

the Haskell answer is presumably voted that high because it's very clever, scores a good score for the type of language it is, and is well explained

the Unary answer could have been submitted by anyone who knows the language with a few seconds of thought, and scores well because the language's existence is amusing and particularly appropriate for the scoring scheme

so ideally they'd want votes in different categories

ok I see

so Unary could highly-voted in like "Funny" but Haskell would get points in a "Clever" category

yep

if we do points, would we want to weigh these categories differently?

my plan was just to total all the points in each category and give people a score for each (this would work if there were only a few)

that way, people can compete in whichever categories they think are most interesting

so would you be able to vote an answer in multiple categories or just one

I think just adding them all together, equally weighted, would work if you wanted the equivalent of an SE rep score, though

in Slashdot you can only pick one and that seems to work fairly well, I'm not sure about a golfing site

I can see some cases where you'd want to vote in multiple categories

maybe the amount of 'points' you get is the amount of unique people who have voted for your answer

would we want to allow anyone to vote or do like an SE-like system where you need X amount of inital points

X for Stack Exchange is two upvotes, on PPCG that's trivial for any serious user

last time I posted an answer (as a guest) I got 40 rep without even logging in

so it's basically just an anti-spam screen

I think the ideal CAPTCHA would be to require people to post a competitive answer (in a way that prevents just copying or copying/modifying existing answers), that's going to be very hard for most spammers

and that's basically what SE's rep limit for voting does

you need a question that scores 3 or an answer that scores 2 (or some combination) to be able to upvote

that should be trivial for any legitimate user