|
| 1 | +--- |
| 2 | +category: release |
| 3 | +permalink: /news/3.8.3/ |
| 4 | +title: "Scala 3.8.3 is now available!" |
| 5 | +by: Wojciech Mazur, VirtusLab |
| 6 | +--- |
| 7 | +# Scala 3.8.3 is now available! |
| 8 | + |
| 9 | +## Release highlights |
| 10 | + |
| 11 | +### Local coverage exclusions with `// $COVERAGE-OFF$` blocks ([#24486](https://github.com/scala/scala3/pull/24486)) |
| 12 | + |
| 13 | +Coverage-instrumented builds can now disable coverage for a selected region of code, instead of excluding a whole file or class. This is useful for generated code, intentionally defensive branches, or support code that would otherwise distort coverage results. |
| 14 | + |
| 15 | +```scala |
| 16 | +//> using scala 3.8.3 |
| 17 | +//> using options --coverage-out coverage-data |
| 18 | + |
| 19 | +class Parser: |
| 20 | + def parse(input: String): Int = |
| 21 | + input.toInt |
| 22 | + |
| 23 | + // $COVERAGE-OFF$ |
| 24 | + def debugFallback(input: String): Int = |
| 25 | + if input == "zero" then 0 |
| 26 | + else -1 |
| 27 | + // $COVERAGE-ON$ |
| 28 | + |
| 29 | +@main def CoverageTest = { |
| 30 | + val parser = Parser() |
| 31 | + assert(parser.parse("42") == 42) |
| 32 | +} |
| 33 | +``` |
| 34 | + |
| 35 | +Only the code between the markers is skipped by coverage instrumentation. The rest of the file is still measured as usual. |
| 36 | + |
| 37 | +### Safe mode for capability-safe code ([#25307](https://github.com/scala/scala3/pull/25307)) |
| 38 | + |
| 39 | +Scala 3.8.3 introduces **safe mode**, a new experimental language subset that can be enabled with `import language.experimental.safe` or `-language:experimental.safe`. As described in the [safe mode reference](https://www.scala-lang.org/api/3.8.3/docs/experimental/capture-checking/safe.html), this is not just “stricter capture checking”: it is a capability-safe subset intended for agent-generated or otherwise untrusted code. |
| 40 | + |
| 41 | +The underlying model is also described in the research paper [Tracking Capabilities for Safer Agents](https://arxiv.org/abs/2603.00991), which proposes using Scala 3 with capture checking as a programming-language-based safety harness for AI agents. |
| 42 | + |
| 43 | +When safe mode is enabled, the compiler rejects unchecked casts and unchecked pattern matches, forbids escape hatches such as `caps.unsafe`, `@unchecked`, and runtime reflection, turns on capture checking with mutation tracking, and restricts access to global APIs unless they are known-safe or explicitly reviewed. |
| 44 | + |
| 45 | +That last point is what makes the feature practical. Safe code is meant to call a restricted set of APIs directly, while effectful or implementation-dependent behavior can still be exposed through wrappers marked `@assumeSafe`. The implementation in [#25307](https://github.com/scala/scala3/pull/25307) makes that boundary explicit: `@assumeSafe` declarations are themselves written outside safe mode, and safe code calls them from within the restricted subset. |
| 46 | + |
| 47 | + |
| 48 | +```scala |
| 49 | +// app.scala |
| 50 | + |
| 51 | +//> using scala 3.8.3 |
| 52 | +//> using file CheckedMailer.scala |
| 53 | +//> using options -experimental |
| 54 | + |
| 55 | +import language.experimental.safe |
| 56 | + |
| 57 | +object PotentiallyUnsafeApp: |
| 58 | + val address = EmailAddress("team@scala-lang.org") |
| 59 | + CheckedMailer.send(address) // ok |
| 60 | + println(address) // error: rejected in safe mode |
| 61 | + address.asInstanceOf[String] // error: rejected in safe mode |
| 62 | + address match |
| 63 | + case EmailAddress(rawAddress) => ??? // error: rejected in safe mode |
| 64 | +``` |
| 65 | + |
| 66 | + |
| 67 | +```scala |
| 68 | +// CheckedMailer.scala |
| 69 | + |
| 70 | +import scala.caps.assumeSafe |
| 71 | + |
| 72 | +@assumeSafe |
| 73 | +object CheckedMailer: |
| 74 | + def send(to: EmailAddress) = |
| 75 | + scala.Console.out.println(s"Sending message to $to") |
| 76 | + |
| 77 | +opaque type EmailAddress <: String = String |
| 78 | + |
| 79 | +object EmailAddress: |
| 80 | + @assumeSafe def apply(value: String): EmailAddress = value |
| 81 | + def unapply(value: EmailAddress): Option[String] = Some(value) |
| 82 | +``` |
| 83 | + |
| 84 | +In the example above, the safe code in `app.scala` can call `CheckedMailer.send`, but the effectful operation is isolated behind an `@assumeSafe` boundary. By contrast, direct calls to `println`, unchecked `asInstanceOf` casts, or `scala.caps.unsafe` helpers are rejected in safe mode. |
| 85 | + |
| 86 | +### Scala 2 JVM optimizer ported to Scala 3 ([#25165](https://github.com/scala/scala3/pull/25165)) |
| 87 | + |
| 88 | +Scala 3 now includes the port of the Scala 2 JVM backend optimizer. The optimizer is opt-in: compiler flag `-opt` enables local bytecode optimizations, while `-opt-inline:...` controls which classes and packages may be inlined across call sites. This brings Scala 3 to feature parity with the Scala 2 optimizer and opens the door to performance gains for JVM applications. |
| 89 | + |
| 90 | +Rather than enabling blanket inlining everywhere, it is usually better to start from explicit filters. The `-opt-inline` setting accepts a comma-separated list of patterns; `**` matches all classes, `a.**` matches a package and its subpackages, `<sources>` matches classes compiled in the current run, and a leading `!` excludes matches. The last matching pattern wins. |
| 91 | + |
| 92 | +```scala |
| 93 | +//> using scala 3.8.3 |
| 94 | +//> using options -opt |
| 95 | +//> using options "-opt-inline:<sources>,my.app.**,!java.**,!org.example.**" |
| 96 | +//> using options "-Wopt:at-inline-failed-summary,no-inline-missing-bytecode" |
| 97 | +``` |
| 98 | + |
| 99 | +In this configuration, the optimizer may inline code from the current compilation (`<sources>`) and from `my.app` subpackages defined in external dependencies, but not from the JDK or the `org.example` packages. This is often a good starting point for applications. For libraries, the conservative choice is usually to inline only from `<sources>` or from packages you fully control. |
| 100 | + |
| 101 | +The optimizer port also brings additional settings: |
| 102 | + |
| 103 | +- `-Wopt:...` enables optimizer warnings. Available choices are `all` or `at-inline-failed-summary`, `at-inline-failed`, `any-inline-failed`, `no-inline-mixed`, `no-inline-missing-bytecode`, and `no-inline-missing-attribute` |
| 104 | +- `-Yopt-specific:...` enables individual optimization passes such as `copy-propagation`, `box-unbox`, `nullness-tracking`, `closure-invocations`, or `redundant-casts` |
| 105 | +- `-Yopt-inline-heuristics:default|everything|at-inline-annotated` adjusts how aggressively the compiler chooses call sites for inlining |
| 106 | +- `-Yopt-log-inline:<prefix>` logs inliner activity for matching methods |
| 107 | +- `-Yopt-trace:<prefix>` traces optimizer progress for matching methods |
| 108 | + |
| 109 | +The `-Wopt` options let you choose between a one-line summary for failed `@inline` calls, detailed per-callsite diagnostics, reporting for heuristic inlining failures, and warnings for cases where inlining could not even be decided because bytecode or Scala inline metadata was unavailable. |
| 110 | + |
| 111 | +> The `-Y...` optimizer flags are primarily intended for debugging and internal use. As with other `-Y` settings, they are not stable user-facing interfaces, and their exact behavior may change between releases. |
| 112 | +
|
| 113 | +As with the Scala 2 optimizer, inlining external code comes with a compatibility trade-off: if you compile against one version of a dependency and later run against a different one, any inlined bytecode will not pick up the dependency's runtime bug fixes or behavior changes. In practice, that means aggressive cross-library inlining is best reserved for applications with tightly controlled runtime classpaths. Read more about binary compatibility of optimized code in the [Scala 2 optimizer documentation](https://docs.scala-lang.org/overviews/compiler-options/optimizer.html#binary-compatibility) |
| 114 | + |
| 115 | +The long-term plan is to build on this work in Scala 3.9 by enabling optimizations for the Scala standard library and the compiler itself. |
| 116 | + |
| 117 | +### `-print-lines` is deprecated for removal, but remains accepted as a no-op ([#25330](https://github.com/scala/scala3/pull/25330)) |
| 118 | + |
| 119 | +Scala 3.8.3 restores the `-print-lines` flag for compatibility, but only as a deprecated no-op. This avoids breaking existing builds in a patch release while giving users time to remove the setting from their build definitions. |
| 120 | + |
| 121 | +The flag no longer has any effect and is scheduled for removal in Scala 3.9.0. |
| 122 | + |
| 123 | +--- |
| 124 | + |
| 125 | +For the complete list of changes and contributor credits, see the [release notes on GitHub](https://github.com/scala/scala3/releases/tag/3.8.3). |
0 commit comments