R.pipe(R.tryCatch, ...) anomaly about ramda

[kedashoe]

ah ty @jlissner ! That explains why adding tap first fixes things as pipe uses the first function to determine the arity of the function it creates

[kedashoe]

So I think the problem is the bare JSON.parse? I'm not sure why tap seems to fix this! The simplest solution is probably just adding an arrow function

      R.pipe(
        R.tryCatch(x => JSON.parse(x), R.always({})),
        R.pathOr("", ["choices", 0, "delta"]),
        R.objOf("delta"),
        R.mergeRight({ isDone: false })
      )

We created R.bind for situations like this, but it doesn't actually seem to solve it here.. not sure exactly what is going on, maybe someone else can shed some more light

[jlissner]

JSON.parse accepts 3 arguments, so ramda is auto-currying the remaining arguments, you can do R.unary(JSON.parse) to fix the issue:

R.pipe(
  R.tryCatch(R.unary(JSON.parse), R.always({})),
  R.pathOr("", ["choices", 0, "delta"]),
  R.objOf("delta"),
  R.mergeRight({ isDone: false })
)

[orsenkucher]

Yep, I mean it's not the issue with Ramda, just was asking on how to make this github issue more indexable for other people if faced with an alike situation.

[kedashoe]

Hi @orsenkucher , could you post some sample input?

[orsenkucher]

@kedashoe Hi!
Yes, sure!
I use this functions with deltas(data.toString()).
Here are few samples of console.log(JSON.stringify(data.toString()))

const data = /* Just paste string from sample */;
const res = deltas(data);
console.log(res);

"data: {\"id\":\"chatcmpl-7TFKQZlJPXyd33RDvEQKAV1QZKzE8\",\"object\":\"chat.completion.chunk\",\"created\":1687204790,\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-7TFKQZlJPXyd33RDvEQKAV1QZKzE8\",\"object\":\"chat.completion.chunk\",\"created\":1687204790,\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hello\"},\"finish_reason\":null}]}\n\n"

"data: {\"id\":\"chatcmpl-7TFKQZlJPXyd33RDvEQKAV1QZKzE8\",\"object\":\"chat.completion.chunk\",\"created\":1687204790,\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"!\"},\"finish_reason\":null}]}\n\n"

"data: {\"id\":\"chatcmpl-7TFKQZlJPXyd33RDvEQKAV1QZKzE8\",\"object\":\"chat.completion.chunk\",\"created\":1687204790,\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\".js\"},\"finish_reason\":null}]}\n\n"

"data: {\"id\":\"chatcmpl-7TFKQZlJPXyd33RDvEQKAV1QZKzE8\",\"object\":\"chat.completion.chunk\",\"created\":1687204790,\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\" is\"},\"finish_reason\":null}]}\n\n"

[orsenkucher]

Also if you call R.ifElse(..., ..., (x) => R.pipe(...)(x)) it will work.

[orsenkucher]

@kedashoe Thanks for your input!

Yeah, JSON.parse seems just not to work with Ramda.
(x) => R.curry(JSON.parse)(x) breaks it again.
I'm curious too, and it'd be nice if someone else could provide further insight.

R.tap(() => {}) doesn't actually do anything, but it likely changes the context in which R.tryCatch calls JSON.parse, making the error go away. So it's better to fall to another solution.

R.bind(JSON.parse, null) isn't working as well, so using the x => JSON.parse(x) workaround is probably the best bet for now. Even though it's less elegant, at least it behaves as expected

[orsenkucher]

Thanks @jlissner!
Seems obvious now ahaha. Should I rename the issue to "Auto-currying in tryCatch with JSON.parse" or something?

[jlissner]

I don't think it would be considered an issue. Ramda has made a choice to auto curry arguments in certain settings.

You could make a post stating your case for why ramda should not auto curry in a tryCatch if you feel strongly about it.

[kedashoe]

I don't think it would be considered an issue. Ramda has made a choice to auto curry arguments in certain settings.

Yes ramda does this pretty much everywhere to determine arity of the functions it creates, it would be quite a change to do something different at this point. It is usually the builtins/functions which accept optional arguments which catch people out. I will close now but feel free to discuss further.

[kedashoe]

Yep, I mean it's not the issue with Ramda, just was asking on how to make this github issue more indexable for other people if faced with an alike situation.

Ah good thought.. probably fine to leave it, but ya something like what you suggested is good