Sharing our experience building a Scala 3 Notebook & Kernel in SMILE Studio (and feedback on scala-cli repl for notebook hosts)

[haifengl]

Hi scala-cli team,

We wanted to share that we have built an interactive Scala 3 notebook and kernel system in SMILE Studio (an agentic data science IDE for the JVM supporting .scala, .sc, and .ipynb notebooks) and invite the team and community to try it out and share feedback.

Context & How it works

In SMILE Studio (ScalaKernel.java), we built our Scala notebook kernel to drive a Scala 3 REPL subprocess over stdin/stdout with SMILE's classpath, native FFM/BLAS flags (--enable-native-access=ALL-UNNAMED), and a predef initialization script (predef.sc) pre-loaded so users can immediately work with DataFrames, ML models, and plots in .scala / .sc notebooks.

We initially built the subprocess kernel around scala-cli repl (passing --java-opt, -cp, and --repl-init-script) and also support launching dotty.tools.repl.Main directly from the bundled classpath so the IDE works out-of-the-box without requiring a separate install step.

Key capabilities in SMILE Studio's Scala Notebooks

• Multi-cell .scala, .sc, and .ipynb notebooks with syntax highlighting, Markdown rendering, and AI code generation/completion.
• AST-aware multi-line cell execution using dotty.tools.repl.ParseResult.isIncomplete to chunk cells into complete top-level statements.
• Live Kernel Explorer that tracks session variables (DataFrames, matrices, trained ML models) for one-click visualization and model serving.
• Graceful cell interruption (Ctrl-C) and fast session reset (:reset) without restarting the child JVM.

Feedback for scala-cli repl in IDE / Notebook integrations

From wiring scala-cli into a desktop notebook UI, a few features would make scala-cli even more powerful as an embeddable notebook backend:

1. Machine-readable REPL framing / status markers: When driving the REPL over stdio, detecting end-of-evaluation and distinguishing compiler errors (-- [E006]) or uncaught exceptions from normal stdout requires prompt/regex parsing. An optional structured output flag or EOT marker per evaluated unit would simplify host integration.
2. In-cell dependency directives (//> using dep): Supporting dynamic dependency resolution mid-session in the REPL (similar to how notebook magic commands add Maven artifacts to a running kernel) is a huge ergonomic win for data science workflows.

Try it out

• Repository: https://github.com/haifengl/smile
• Scala Kernel: studio/src/main/java/smile/studio/kernel/ScalaKernel.java
• Scala API: scala/

We welcome the scala-cli team and community to try it out and share your thoughts and feedback!

[Gedochao]

• Machine-readable REPL framing / status markers: When driving the REPL over stdio, detecting end-of-evaluation and distinguishing compiler errors (-- [E006]) or uncaught exceptions from normal stdout requires prompt/regex parsing. An optional structured output flag or EOT marker per evaluated unit would simplify host integration.

@haifengl This is worth considering as a feature request (cc @SolalPirelli @warcholjakub), but would be worth raising separately in the compiler bug tracker: https://github.com/scala/scala3/issues
While the REPL is launched from Scala CLI, the REPL itself lives in the Scala 3 compiler repo.

• In-cell dependency directives (//> using dep): Supporting dynamic dependency resolution mid-session in the REPL (similar to how notebook magic commands add Maven artifacts to a running kernel) is a huge ergonomic win for data science workflows.

This was already delivered in Scala 3.10.
https://www.scala-lang.org/news/3.10/#repl
https://github.com/scala/scala3/releases/tag/3.10.0

[haifengl]

Really cool. Will try it with 3.10. Any plan to back port it to 3.9 LTS? Thanks!

[Gedochao]

I think we might have it in 3.9.1. Needs to wait its turn in the back port queue.

[Gedochao]

Nice project there!
Note that this is not a Scala CLI issue in any capacity, let me convert this into a discussion.