Project URL: Glitch :･ﾟ✧
the code for this has some spaghettiness to it, but here it is for fun anyway, a bot that watches you type in the Glitch editor and edits with you.
and I don’t mean that it programs with you, it’s only an experiment to show how to interact with the live collaboration system, “OT.” here it’s set up so that when you type a comment starting with > , it’ll repeat the comment, starting with < , in uppercase.
help me think of wacky things to do with this.
This works, and is so cool! I’ll try using that to implement ChatGPT in Glitch.
Too late, I’m doing it myself. 
I actually started building it, but had some problems with the fetching URLs stuff so I gave up.
As you are working on your bots, let me know what roadblocks you hit, what API endpoints would help you succeed, what kind of bots you’d like to build for the Glitch editor, etc. I think this is super interesting work and love the idea of community-built bots on here!
lol done (send gpt-reset to reset your conversation, it also wipes when you stop the bot.js while it’s running)
do I fix conversations wiping when you stop bot.js? and also probably make a README.md on how to use it 
this is an awesome project, it’ll make MY work easier that’s for sure.
EDIT: There’s some documentation on how to use it now and I’m making it easier to use by the minute
For those trying to work with this: Your user token is in your cachedUser object inside localStorage, as your persistentToken value. This is neccesary to create the connection to your project. The rest I handle- including getting the project ID, you just have to give the username. Should work with private ones too since I pass in your user token at that point.
Heyyyy! I spent a good few hours working on this and I want to die.
It’s done now though! Go check it out, it’s somewhat easy to integrate now (and once you’ve done it, you can just keep the tab open to integrate ChatGPT with your project)
It also now respects whitespace fully 
yo @wh0 you know more than me about this, can you help? I tried to add whitespace support but it sometimes replaces in the wrong spot (I do put in the whitespace length to avoid that but it seems to not be working)
it also has ChatGPT support if you want to test that too
regex first capture group doesn’t look like a pattern for matching spaces. (. ) matches any character followed by one space. could try (\s*)(\S.*)?
Worked like a charm! A few changes and now it actually respects whitespace. You probably saw the dirty patch I made to fix an indentation issue near the bottom 
new issue: offsets in .html 
offsets in your indentation in .html / .svelte cause issues with the bot. trying to figure out the offset generation as we speak but it’s painful
So far I’ve figured out the offsets to make the line appear right BUT now we have the issue of broken bits after, I’m removing part of my code to see if it helps
EDIT: for some reason Glitch injects a SINGLE SPACE whenever the .html offset issue happens. no more auto new lines to mitigate. please tell me if this is the right way to do this
oh it’s better not to use start or end like that after sending the first transaction. every time you await there could have been other edits made that shift things around. you can rely on the transactions to be up to date
confused coder sounds
what should I change? I only use await to grab the message from OpenAI’s API and also in whatever code you had
const whitespaceLength = spacesRemoved.length;
...
unsubmittedTx.tx[docId][0][2] = spacesRemoved + "// < Contacting ChatGPT...";
...
unsubmittedTx2.tx[docId][0][0] += whitespaceLength + 5; // relative offset
unsubmittedTx2.tx[docId][0][1] = 22; // splice length
the relative offset in [0] looks right, for bringing you to just before the “C” in “Contacting”
the splice length in [1] looks one too many, as Contacting ChatGPT... is 21 characters. you have no guarantees what comes outside the transaction’s splices, so it’s not safe to do 22 there.
edit:
and that if thing that comes later shouldn’t trigger, as long as there are no concurrent edits going on. does it?
hey wait a minute, I could’ve sworn I put 21 there. >:/
anyway fixed, testing patch
EDIT: apparently a single space is injected after sending the first message which is why I set it to 22 >:?
also the extra space comes back.
it does, the .html offset issue causes it to flip (end no longer equals the value returned in unsubmittedTx2 and that new value breaks things, hence the offset fix
how much does the [0] change by?
it depends on how offset it is
one HTML object = 2
two HTML objects = 4
you get the point
it’s an annoying bug and I fucking hate it
my initial patch for it only fully accounts for removing the whitespace from one HTML object
weeeeird let me try in a remix
go ahead. I’m going to try to squash it and will edit if I get it down
EDIT: squashed, my code now has a fix
the offset is unsubmittedTx2.tx[0][0] - end + whitespaceLength + 5
subtract the offset from relative offset
add the offset to splice length
(Glitch, what the FU-)
okay after tons of printf debugging, here’s how things were going haywire. don’t blame Glitch, it was a bug in my code.
so as a result,
I’ve made the following fixes in the original maple-separated-seal:
Move the version increment to the end of the oplist apply-er:
function opListApply(opList) {
- version = opList.version + 1;
for (const op of opList.ops) {
docsVisitOp(op);
txsVisitOp(op);
readerVisitOp(op);
}
+ version = opList.version + 1;
}
this is a safeguard to make any transactions sent while in the middle of processing an incoming oplist fail on the server. okay and I just realized that I had not settled whether oplist is one word or two words together, as opList 
.
wait for a microtask at the top of readerVisitLine
function readerVisitLine(docId, unsubmittedTx) {
+ (async () => {
+ try {
+ // wait for a microtask, after we finish processing the whole oplist that
+ // this line appeared in.
+ await Promise.resolve();
+
if (!unsubmittedTx.tx) throw unsubmittedTx.reason;
...
const submitTxPromise = submitUnsubmittedTx(unsubmittedTx);
- (async () => {
- try {
const unsubmittedTx2 = await submitTxPromise;
this prevents us from submitting a transaction until after the whole incoming oplist is processed. note that the first if (!unsubmittedTx) throw is now required, as an oplist may create a line and then disturb it in the same version, before we finish awaiting that Promise.resolve().
This is only a partial fix sadly
Magic spaces were part of the problem. Using your new code:
I’ll update my bot to the new code but now I need to add even more code to handle the dumb spaces.
EDIT: New code makes it even harder to work around the dumb space issue >:/
I’m going to stay on the old codebase in order to keep the workaround for now
oh is this for when you don’t have a newline at the end of the string that the bot adds?
No the newline just moves the spaces AFTER the newline and makes mitigation worse
but isn’t it nice to have the extra newline and indent?
(indent) // > (your message)(your newline)
(bot indent) // < (bot message)(bot newline)
(added indent)(cursor)
that way you can start typing another // > ... right away
vs without the extra newline:
(indent) // > (your message)(your newline)
(bot indent) // < (bot message)(added indent)(cursor)It would be nice if the indent worked right, currently it only spawns the indent the editor expects (Which often doesn’t work with extra / removed indents)
do you have an example of that, when it’s wrong?
edge case but still a thing
it’s inserting responses at the same indent level as the request
the glitch editor has its opinion on how much to indent a new line, and it does so unrelated to the bot
there’s the problem. I want it to be on the same line as the first request, and ignore glitch’s annoying indent. I’m assuming this is impossible
also if you remove the newline the spaces just appear after the response anyway
I suppose you could scan forward in the document and detect spaces and delete them from the transaction
here’s an idea on how I could do this
grab the next ~200 characters from our // < ... starting point
make a regex statement matching that starting part and then all spaces until a non-space is hit
collect those spaces into a regex group and do some stuff with them- namely, removing or adding more spaces to compensate for the extra ones
I have no idea how to do that but it would work
wait did I read this right? did you want it like this?
(indent) // > (your message) < (bot message)(your newline)
(added indent)(cursor)
yes. the negative indent too
there’s no negative indent there.
but there’s a part where it sets the relative offset to end, meaning to insert the text right after the newline. you would instead use end - 1, so that it gets inserted before the newline. and with that, don’t insert any newlines, because that would make it look like the same line has been created again
This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.