Public compatibility roadmap¶
Owner: fontdone maintainers
Open goals: 3
Deletion condition: all three goals below are complete, their evidence remains reachable in public guides, and this page and its index entry are removed. This roadmap describes product gaps, not instructions for an agent session.
Complete the C replacement contract¶
The target is the full pinned FreeType 2.14.3 contract: functions, constants, types, layouts, callbacks, ownership, state, errors, modules, headers, artifacts, and platform behavior. The current measured subset is in Evidence.
Completion requires all 12 categories, every required denominator, exact public
observations, and external C/C++ consumers against shipped artifacts on all
supported targets. A declaration or validation-only route is insufficient.
make c-abi-contract-complete is the gate; ordinary measurement keeps debt visible.
Complete source coverage¶
Reach 100% of lines, branches, functions, and regions in source-bound all-lane coverage. Preserve the full test/input matrix and production denominator. Add legitimate public parity inputs for missing behavior. Undefined C behavior is not an acceptable oracle. See Development.
Accept performance budgets¶
The benchmark measures operation latency, throughput, complete-process RSS, and exact unstripped artifact sizes. Record qualifying clean ten-sample runs on the same environment before reviewing regression thresholds.
The most-sampled current environment has 7 / 5 qualifying clean runs.
The policy remains collecting_baseline. Existing samples do not automatically
approve thresholds. make bench-regression fails closed until reviewed budgets
are active. Follow Benchmarking.
Evidence ledger¶
| Goal | State | Evidence | Verification |
|---|---|---|---|
| G01 | OPEN | Full C contract and platform scorecard | make c-abi-contract-complete |
| G02 | OPEN | Source-bound all-lane coverage | make test-coverage-all; all four metrics exactly 100% |
| G03 | OPEN | C/Rust samples, memory, sizes, and reviewed thresholds | make bench-regression |
An alpha package release does not close these goals. The ordinary release policy requires passing runnable comparisons and honest incomplete evidence.