https://www.boundaryml.com logo
Powered by
Either it's broken on Windows or I've
# general
j
Either it's broken on Windows or I've got something configured weirdly but the short 5 minutes I spent searching BAML source for an explanation didn't find the actual part that's causing the error just the error message dispatch so I have no clue what's breaking here. https://cdn.discordapp.com/attachments/1119375594984050779/1522775188649935038/WindowsTerminal_WcksvQnuLh.png?ex=6a49b268&is=6a4860e8&hm=43f43ab6b60bfa0406cb07a45800f2f0527ef204026b004f61a87353b96fc878&
v
cc @rossir.paulo (BAML) who was working on our packaging side!
r
As soon as I’m back from the airport I’m on it!
j
I'm happy to help with figuring it out 🙂
v
what you can do meanwhile as a workaroudn is download the VSIX directly from our releases and drag it into your IED or choice
j
atm I'm just using codex and viewing BAML without any fancy stuff which would be awful except I love your syntax so it's fine for now lol
v
if you share what version you're on i can link you to actual file
j
v
actually you can also just run 
baml playground
then go to 
editor mode
in it
j
😂
v
oh XD
can you use 0.13+?
baml toolchain status
j
I'm running toolchain update atm
Also this example shows how clean the syntax is for readability because even without highlighting it's super clear what's actually happening: https://imgur.com/ZT3lpCk
v
you're gonna love our new interpolated strings in that case 🙂
https://beps.boundaryml.com/beps/49 I beleive this is live in 13+
j
ide install still fails but playground now is workingish, it gives me the web page but no content loads and I get this error after pressing open editor https://cdn.discordapp.com/attachments/1522775188822032550/1522778603803115612/firefox_SH8PItnnp2.png?ex=6a49b597&is=6a486417&hm=c605fb7180c120dab612ec65dd25a917b0c0d35c561275c7c2423ab044cc833c&
v
are you running the playground command in the directory that contains the 
baml.toml
?
also, sorry for the back and forth 🥲 we sadly haven't tested as rigorously on windows, but should be able to stabalize this soon
j
Nah it's alg I understand what you guys are doing, it just takes time before this stuff works smoothly and even then the more you develop the language the more things will break anyway
I don't think I have a baml.toml anywhere? Weird
v
oh did you do 
baml init
?
can just run that real quick
j
I was using codex so it should have picked that up, but I'll just manually do it
v
and it'll add it in
j
Okay I have the baml.toml in my agent-host workspace which itself already had the baml_src folder with files in it. But running baml playground still doesn't populate anything.
v
hmmm
j
I'm very good at breaking things without trying lol
v
ok i'll test with windows again and likely just need to find a local repro before i take a bunch of your time
j
Nah it's alg I'm happy to work on this stuff, I want the language to be better for myself anyway.
v
j
Plus at some point I plan on applying for a job with your team anyway, you have avery and a really cool project lol
v
i think an easy thing to patch here may be instead of 
--code
or 
--cursor
we could add in a 
--download <dir>
option which just gets the vsix locally
happy to have you take a stab at this
it should be super easy, but if we can make the actual playground work on windows, that would be great. no idea what the problem there is
if you have questinos about how to set it up or anything just ping us on this thread!
j
Yeah it shouldn't be difficult to work out, is there any kind of debugger for BAML?
v
you basically just do:
Copy code
cargo build -p baml_cli
then just use 
baml-cli playground
to test
not yet!
but we are working on a tracing solution that kind of gives a hindsight debugger
j
Yeah that's really all you need, just some way of tracing history
v
yea but locals aren't yet captured there
so that would be the main thing we would want
j
Looks like a path construction bug xD
v
🥲
j
Also that does look like an exactly windows shaped bug too
v
we need a bit more testing on windows. assumed that ci/cd had caught that but apparently not
we'll add a few more tests (likely will go out next canary release)
j
Yeah avery told me you guys don't use windows much which makes sense, I wouldn't use windows either if this pc wasn't both for gaming and development
v
yea i used to use it back in msft, and never again
though we should buy a windows machine for the office
so we can test this stuff
j
100% you need to buy a windows machine 😂
There's a lot that makes it feel very much like no one has used windows overly much with this, which makes sense tbh it wasn't really worth your time/effort until recently to get windows support clean
Definitely need to slowly refine and test it going forward tho
Okay I figured this out, command_on_path had windows handling for code.cmd already but command_on_path wasn't being called when you passed explicit arguments and also the .cmd check needs to happen before the normal check because otherwise it finds code.exe instead of code.cmd which becomes code with no extension and fails (I think due to rust assumptions on windows but I didn't care enough to find out, it works now lol). Want me to do a pull request with the fix and if so anything I should keep in mind? Also I could add that --download option while I'm at it if you want.
And people still treat AI functions as crazy sketchy behaviour meanwhile path ordering bugs have existed since path has existed as a concept lmao
v
oh nice great find. yea that --download option woudl be great to do that!
great find btw!
j
tbf codex does the manual labour, all I do is point it at the thing and interogate it like it owes me money until I get an honest an elegant solution lol
v
tbf that is mostly software engineering haha
j
atm yeah it really is, and It's the part that I've always been the best at largely because my physical disability makes actually writing code hard so I got really good at taking my time to understand problems. It feels like I've finally been unleashed on the world now that I don't need to write code and I have wings that have finally opened up letting me soar through all the problems my heart desires to tackle.
This tech really has changed my life already
v
thats pretty incredible. i mean even for me, i feel like i can finally actually just focus purely on thinking instead of making any tradeoffs
j
Yeah I've been reading through past projects thinking of what I'll use as examples for applying to your team and a theme that was really obvious to me was that I created a bunch of really cool projects that solved difficult problems but all of them were fundamentally limited in scope and I clearly had to take a bunch of shortcuts instead of implementing things exactly how I wanted. It kinda makes me think all the people talking about how slop will make code worse haven't really been reading either their own code or what other people have been doing already. The real slop is from us humans not the AI lol. It's a give/take thing where it rewards effort but more heavily punishes laziness as well imo.
--download or --extract? The file is part of the toolchain assets already so it's not really downloading anything just copying it into a folder and --download smells to me like the type of misleading name that'd make future debugging annoying.
--copy is more accurate to the function but leaking implementation also feels wrong to me, an intent based name that's not crazy misleading is imo the right option
I hand wrote an explanation of what happened in the PR so hopefully agents can avoid regressing this in the future because it feels like the type of surface ripe for accidental regression
And here's the playground fix: https://github.com/BoundaryML/baml/pull/3915
v
the first pr has what looks like a decent suggestion by code rabbit!
j
The .exe one? I don't actually believe using code.exe for this would work on windows. At least it doesn't work at all on my machine for some reason, it's only the .cmd path that actually works. Not sure exactly why it fails. Even manually running code.exe doesn't work on windows. So I'm pretty confident the suggestion is just code rabbit hallucinating. Windows rejects the .exe path in both cmd and ps. https://cdn.discordapp.com/attachments/1522775188822032550/1523000321193017405/WindowsTerminal_8eQp7bXq8E.png?ex=6a4a8414&is=6a493294&hm=9833a84f9678df8d7fe7d62723834468aa20189b928f9efdccfedc6f29a31f65& https://cdn.discordapp.com/attachments/1522775188822032550/1523000321696071822/WindowsTerminal_eSDLfQt4dR.png?ex=6a4a8414&is=6a493294&hm=eb53f5cf600d86a8e3afed2e8dad5898b97a77ac84a1782ee609511cf62d501a&
v
oh thats odd! i would have expected code.exe to work
windows is a strange beast XD
j
Yep I'm pretty sure this is a core OS quirk itself combined with the specific way vs code is setup
If it wasn't for games I'd never touch this OS lol
Windows is like the javascript of OS
b
the 
ide install
fix looks good to me, i need to look at the windows path resolution one a little more closely to understand what's happening there
j
This screenshot shows how it was mangling the path on Windows.
The Open in browser button for playground has a similar path bug as well. If you're too busy and want me to help with this I could find some time to do a deep dive of various Windows path bugs with Codex and just go through and crush them all out. Seems to be a repeating pattern. ETA: fyi I'm not overly fussed if you don't want me doing this since it'd be a rather wide set of fixes to your codebase which can be annoying if someone unfamiliar misses important nuances that only surface later. Plus it's probably a decent bit of time/effort so I would be perfectly happy if you would rather spend the time on it instead of me. But a solution for if you don't want me making a ton of changes without communication/input could be breaking the problem up into 2 parts. I can first go through and deep dive understanding the problem then come back to you guys and discuss wider solutions for the various issues I find then go implement those solutions. https://cdn.discordapp.com/attachments/1522775188822032550/1524413826081095780/Code_VcdvWwm8zT.png?ex=6a4fa882&is=6a4e5702&hm=f6bc6c48a89b6ec2535db50cace9bc9c31e6d67d6b37734cfca1d0de21db21a0&
b
We'd love help with getting our windows deployments to be bulletproof!
Left a comment on 3915 with a different approach I think is worth trying, lmk what you think
@jubjub feel free to start new threads in #1253172368671772732 and tag me to discuss fixing Windows handling issues! 🙂
For longer form discussion I'd say that discussing here in Discord would be better, GH is hard to have synchronous convos
j
Yep sounds good.
2 Views
PreviousNext
Powered by