krastanov/julia-jet

JET.jl

Use this skill for all JET work. Keep the overview here and load the narrow reference that matches the task.

Warning

Use JET only on the latest stable Julia release. Nightly and pre-release Julia versions often break JET because it depends on compiler internals.

Environment Rule

• Prefer package-integrated JET runs via Pkg.test(...; test_args=["jet"]) or the package's documented JET arguments.
• Keep JET as a test-only dependency in test/Project.toml with root [workspace] projects = ["test"]; avoid new [extras]/[targets] entries.
• Avoid direct test/runtests.jl execution unless you are debugging the test router or worker behavior.
• Set JULIA_PKG_SERVER_REGISTRY_PREFERENCE=eager or ENV["JULIA_PKG_SERVER_REGISTRY_PREFERENCE"] = "eager" before Pkg operations in this workspace.
• Never read or edit Manifest.toml directly. If a JET failure appears after dependency churn, run Pkg.update() and Pkg.resolve() in the relevant package and JET environments first. If recurrent issues remain, delete the relevant Manifest.toml file and regenerate it with Pkg.instantiate() before blaming the analyzed code.

Choose a Mode

• Start with @report_opt to find runtime dispatch and captured variables.
• Then use @report_call or report_package to find type-level errors.

Quick Start

using JET

@report_opt sum(Any[1, 2, 3])
@report_call sum("julia")

using MyPackage
report_package(MyPackage; target_modules=[MyPackage])
report_file("my_script.jl")

Filtering

@report_opt target_modules=(MyPackage,) my_function(args...)

Read JET stack traces from the bottom up: the bottom frame is the actual problem site.

References

• references/config.md
• references/testing.md
• references/error-kinds.md
• references/fixing-dispatch.md

Related Skills

• julia-perf - performance work after dispatch cleanup
• julia-tests - test-suite integration patterns