Milestones

List view

• M12: 1.0

  Gate: every number in the target table either met or publicly restated with the reason. Compatibility published in GraalPy's format, including the embarrassing numbers.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M11: tooling

  Gate: a real Django or FastAPI application runs, is testable with pytest, measurable with a profiler, and debuggable.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M10: the C-API layer

  Gate: numpy builds against kohebi and passes its own test suite.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M9: sealing and the profile handoff

  Gate: 10x CPython on the AOT-favourable set and 10x memory on data-dominated workloads. If this fails, restate the headline numbers.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M8: AOT, first version

  Gate: build times meet the M0.1 numbers on real programs. Output at least as fast as JIT steady state with no warmup. run and build behave identically across the test suite.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M7: the standard library

  Gate: at least 90% of CPython's test suite passing with every exclusion enumerated and justified. Top 100 PyPI packages install and import.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M6: the optimizing JIT

  Gate: 4x CPython geomean. Clean deopt and OSR stress runs. A comprehensible --deopt-stats report.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M5: Rust interop

  Gate: pydantic-core and polars rebuild and pass their own tests. Direct typed call under 15 ns. Miri clean over the boundary crates.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M4: CIR and the baseline JIT

  Gate: 2x CPython geomean on pyperformance. The CIR-to-stub path and the CIR interpreter agree on every operation.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M3: memory management and threads

  Gate: no leaks and no crashes collecting at every safepoint over the whole M1 corpus. At least 6x scaling on 8 cores. Single-thread no-GIL cost under 10%.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M2: the object model

  Gate: at least 2x smaller than CPython on data-dominated workloads, and never worse on any benchmark. Speed will be mediocre here and that is expected.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M1: the interpreter

  Gate: runs a non-trivial pure-Python program correctly, including classes, generators, exceptions, closures, and comprehensions. Passes a defined subset of CPython's test suite covering core language semantics.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• M0: de-risking

  Four experiments, no product code. Gate: all four answered in writing, including the two that can end the project.

  No due date

  •3/5 issues closed
  60% complete2 open3 closed