initialisms yak shave | Dagger | Page 1

hushed cipher Jun 10, 2025, 2:42 PM

#

time="2025-06-10T14:37:47Z" level=info msg="http: panic serving @: argument \"url\" not found\ngoroutine 448 [running]:\nnet/http.(*conn).serve.func1()\n\t/usr/lib/go/src/net/http/server.go:1947 +0xbe\npanic({0x2abb700?, 0xc0209dfab0?})\n\t/usr/lib/go/src/runtime/panic.go:792 +0x132\ngithub.com/dagger/dagger/dagql.Field[...].Args(0x0?, {0xc011214160, 0x8, 0xc00fd64630?})\n\t/app/dagql/objects.go:1230 +0x4f4\ngithub.com/dagger/dagger/core/schema.(*gitSchema).Install(0xc0227abe80)\n\t/app/core/schema/git.go:50 +0xa5e\ngithub.com/dagger/dagger/core/schema.(*CoreMod).Install(0x3989000?, {0xc0209de4a0?, 0xc003fc0840?}, 0xc00116cc40)\n\t/app/core/schema/coremod.go:61 +0x63e\ngithub.com/dagger/dagger/engine/server.(*Server).initializeDaggerClient(0xc000778608, {0x3996f78, 0xc003ac1680}, 0xc0001ffe00, 0xc001162fc0, 0xc0006627d0)\n\t/app/engine/server/session.go:583 +0x15d1\ngithub.com/dagger/dagger/engine/server.(*Server).getOrInitClient(0xc000778608, {0x3996f78, 0xc003ac1680}, 0xc0006627d0)\n\t/app/engine/server/session.go:824 +0xbd6\ngithub.com/dagger/dagger/engine/server.(*Server).serveHTTPToClient(0xc000778608, {0x3989230, 0xc001117960}, 0xc0001ff7c0, 0xc0006627d0)\n\t/app/engine/server/session.go:995 +0xef6\ngithub.com/dagger/dagger/engine/server.(*Server).ServeHTTP.httpHandlerFunc[...].func1(0xc0001ff7c0)\n\t/app/engine/server/session.go:1548 +0x31\nnet/http.HandlerFunc.ServeHTTP(...)\n\t/usr/lib/go/src/net/http/server.go:2294\ngithub.com/dagger/dagger/engine/server.(*Server).ServeHTTP(0xc000778608, {0x3989230, 0xc001117960}, 0xc0001ff7c0)\n\t/app/engine/server/session.go:904 +0x162\nmain.main.func2.2({0x3989230?, 0xc001117960?}, 0x0?)\n\t/app/cmd/engine/main.go:423 +0x9d\nnet/http.HandlerFunc.ServeHTTP(0xc003ac1050?, {0x3989230?, 0xc001117960?}, 0x419534?)\n\t/usr/lib/go/src/net/http/server.go:2294 +0x29\ngolang.org/x/net/http2/h2c.h2cHandler.ServeHTTP({{0x3964740?, 0xc026f58e10?}, 0xc001254f50?}, {0x3989230, 0xc001117960}, 0xc0001ff7c0)\n\t/go/pkg/mod/golang.org/x/net@v0.41.0/http2/h2c/h2c.go:125 +0x673\nnet/http.serverHandler.ServeHTTP({0xc003ac1020?}, {0x3989230?, 0xc001117960?}, 0x6?)\n\t/usr/lib/go/src/net/http/server.go:3301 +0x8e\nnet/http.(*conn).serve(0xc001128ea0, {0x3996f78, 0xc0034e68a0})\n\t/usr/lib/go/src/net/http/server.go:2102 +0x625\ncreated by net/http.(*Server).Serve in goroutine 407\n\t/usr/lib/go/src/net/http/server.go:3454 +0x485"

#

here's how it looks in the CLI:

❯ dagger-dev -m workspace shell
▼ connect 9.1s ERROR
! start session: connect buildkit session: read response: unexpected EOF
├─● starting engine 0.0s
├─▶ connecting to engine 9.1s
╰─✘ starting session 0.1s ERROR
  ! connect buildkit session: read response: unexpected EOF

#

gonna nuke the volume and start fresh in case it's somehow related

sleek fjord Jun 10, 2025, 2:45 PM

#

🤔 odd odd odd

#

um that's really strange

#

which version?

hushed cipher Jun 10, 2025, 2:50 PM

#

commit 94398f28962a76056d9d8e9383a876af59ec954d (upstream/main, upstream/HEAD, main)
Author: Justin Chadwell <me@jedevc.com>
Date:   Tue Jun 10 15:01:09 2025 +0100

    chore: prep for v0.18.10 (#10559)

first time was against 7a93970f5

#

huh i'm also seeing this inside dagger call test specific --run Interface/IfaceCall (test unrelated, it's just what i happened to run)

sleek fjord Jun 10, 2025, 2:52 PM

#

so the stack trace seems like there's an issue with setting up the dagql handlers

#

we're trying to attach docs for url - but somehow it's not being found

hushed cipher Jun 10, 2025, 2:53 PM

#

docker volume rm dagger-engine.dev finally finished, so trying a fresh dev engine now

#

very strange how CI wouldn't have caught this if it's a real thing

hushed cipher Jun 10, 2025, 2:54 PM

#

hushed cipher `docker volume rm dagger-engine.dev` finally finished, so trying a fresh dev eng...

nope, same error

time="2025-06-10T14:53:40Z" level=info msg="http: panic serving @: argument \"url\" not found\ngoroutine 357 [running]:\nnet/http.(*conn).serve.func1()\n\t/usr/lib/go/src/net/http/server.go:1947 +0xbe\npanic({0x2abc780?, 0xc000b48080?})\n\t/usr/lib/go/src/runtime/panic.go:792 +0x132\ngithub.com/dagger/dagger/dagql.Field[...].Args(0x0?, {0xc000ab0160, 0x8, 0xc0009b2360?})\n\t/app/dagql/objects.go:1230 +0x4f4\ngithub.com/dagger/dagger/core/schema.(*gitSchema).Install(0xc0006a8b50)\n\t/app/core/schema/git.go:50 +0xa5e\ngithub.com/dagger/dagger/core/schema.(*CoreMod).Install(0x398a440?, {0xc00066feb0?, 0xc00066b860?}, 0xc000882930)\n[... snipped ...]

sleek fjord Jun 10, 2025, 2:55 PM

#

okay super weird, i do not see the same issue 🤔

hushed cipher Jun 10, 2025, 2:55 PM

#

something something random iteration order, and my CPU is in a different timezone?

#

will poke through code and see what i can find

#

oh this could be related to local changes

#

i started shaving the initialisms/acronyms yak

sleek fjord Jun 10, 2025, 2:57 PM

#

AH

hushed cipher Jun 10, 2025, 2:57 PM

#

and "url" is a little sus

sleek fjord Jun 10, 2025, 2:57 PM

#

ngl that does sound very sus

hushed cipher Jun 10, 2025, 2:57 PM

#

i got here because LLM() was being turned into l-lm thinkies

#

okok, most likely false alarm, sorry for the immediate post-release scare 😂

sleek fjord Jun 10, 2025, 2:59 PM

#

😅

hushed cipher Jun 10, 2025, 3:06 PM

#

confirmed 🙇‍♂️

#

initialisms yak shave

#

alright im gluing the hair back on the yak, this seems like a nightmare. unless someone really wants me to keep going

sleek fjord Jun 10, 2025, 3:08 PM

#

😢 damn, yeah, it really is.
we're so dependent on magic casing in so many places

hushed cipher Jun 10, 2025, 3:08 PM

#

basically, we want "LLM" to convert to "llm" when used as an arg, but "EvalsLLM" instead of "EvalsLlm" for ToCamel

#

which doesn't seem possible with strcase?

sleek fjord Jun 10, 2025, 3:08 PM

#

i seem to rajat gave it a go a while back, and he converted a load of stuff

hushed cipher Jun 10, 2025, 3:09 PM

#

ah right

sleek fjord Jun 10, 2025, 3:09 PM

#

the problem at it's core is that doing this in a backwards compat way is not? super fun

#

this has been in the back of my head for an eventual v1.0, we would completely rip out strcase and centralize everything neatly in one place

hushed cipher Jun 10, 2025, 3:09 PM

#

yeah for sure

#

we've also got our custom implementation in the go sdk

#

er, codegen

sleek fjord Jun 10, 2025, 3:10 PM

#

yup, each sdk does it's own thing as well 😢

hushed cipher Jun 10, 2025, 3:10 PM

#

which i started moving to the go sdk so we can at least share it between CLI/codegen/engine, but yeah, that's the smallest part

sleek fjord Jun 10, 2025, 3:10 PM

#

i know @tawny zodiac is working on / interested on working on a thing to centralize some of the logic of codegen in one place

#

that feels like a step towards the right place?

hushed cipher Jun 10, 2025, 3:10 PM

#

this would be a surefire candidate for that yeah

#

i had a thought to just have a .json file of our casing rules or something, so all the SDKs can load it up

#

wonder if there's any prior art to this

sleek fjord Jun 10, 2025, 3:12 PM

#

🤔 that's a fun idea

#

i've thought about having the engine do all the splits? and then the sdks just send/recv nameParts []string or something

#

but also, yes, horrible

hushed cipher Jun 10, 2025, 3:16 PM

#

just about anything is better than the status quo haha. that could be an API or something thinkspin

#

an oddly specific one, but it kinda makes sense, we already have a notion of 'original name' vs. 'converted name' so it's within the fold to some extent

sleek fjord Jun 10, 2025, 3:17 PM

#

i mean if it's all in the typedef api, then we could just hook in a nameParts field to everything there

#

then each sdk could pick up on those... it would even be possible to have a easy backwards-compat check (if <version, use old name, otherwise, join together nameParts in some way)

hushed cipher Jun 10, 2025, 3:22 PM

#

i think you would still need knowledge of initialisms on top of that when joining, but that also becomes language specific. Ruby would want EvalsLlm, Go would want EvalsLLM

sleek fjord Jun 10, 2025, 3:24 PM

#

maybe it's not just []string? something like []struct{part string, isInitialism bool}

#

or similar

tawny zodiac Jun 10, 2025, 3:33 PM

#

Yeah, I started experimenting with that in the codegen module. I think it's a good solution, that can be added to TypeDefs. Basically you'd keep initialisms here, like:

LLM
evals LLM

So Go would preserve the casing before upper casing each start of each word:

LLM
EvalsLLM

Typescript would lower case everything and camel case it:

llm
evalsLlm

Python would lower case everything and replace the separator with underscore (similar for CLI):

llm
evals_llm

It could be up to each SDK to return the right word split so the engine, or at least have it has a possibility for override in case the engine does it wrongly. As for the other direction, could also be converted to target conventions by the engine rather than languages so SDK just has to pick the right result from convention rather than doing the conversion itself.

hushed cipher Jun 10, 2025, 3:35 PM

#

ah yeah that's a cute way to condense the info into just []string
if lowercase => not acronym
if uppercase => acronym, decide whether to preserve (language-specific)

tawny zodiac Jun 10, 2025, 3:37 PM

#

Yeah, that's been on my mind for a long time for centralizing codegen, which I want it to be based on typedefs rather than graphql schema, but this could be added to typedefs directly as a first step. Would make these conversions much simpler and consistent between languages.

#initialisms yak shave