Skip to content

runtime-support.md

Latest commit

 

History

History
110 lines (88 loc) · 5.76 KB

runtime-support.md

File metadata and controls

110 lines (88 loc) · 5.76 KB
Raw

Runtime support and repeatable checks

English · Deutsch

The full feature set targets Python 3.9 or later and Git 2.36 or later on macOS and Linux. Python 3.9 supplies the string operations used by the single-file program; Git 2.36 supplies NUL-delimited worktree registration output. See the Python string API and Git 2.36 worktree format. An older Git that cannot prove an inventory produces an unavailable state; there is no guessed fallback to newline-delimited paths.

The TUI needs standard-library curses, a POSIX terminal and a UTF-8 locale. No third-party Python modules are required. macOS app shortcuts use open -a; Linux reports that platform limit and can still change to the repo through the wrapper. The zsh wrapper, installer and build/release scripts need zsh. A Linux host without zsh can run the Python TUI and CLI; its zsh-specific tests are skipped. Windows and OS versions beyond the recorded checks have not been validated.

Reproduce a check

python3 tools/check-compatibility.py
python3 tools/check-compatibility.py --python python3.9 --python python3.14 --suite core
python3 tools/check-compatibility.py --python python3.15 --git /usr/bin/git --screens

Each combination records interpreter, Git and product versions, the source hash, exit codes, duration and test count. Full is the default and includes the existing safety, CLI and release fixtures. Core selects only the separately maintained feature-contract suites; it does not substitute for the full safety suite. --screens verifies the six generated images. No runtime is installed or updated. Temporary PATH wrappers select the requested Git without changing global settings. An unavailable or below-floor runtime is a failed check, not a skipped success.

Validation record, 2026-10-16

0.30.3

System Python Git Result
macOS 26.7.1, arm64 3.15.0 2.56.0 842 tests and six identical images
macOS 26.7.1, arm64 3.9.6 2.56.0 157 core checks
Linux, x86_64 3.12.3 2.43.0 788 executed, 54 zsh checks skipped

Four regressions were reproduced before correction. Tests cover HEAD changes including unborn/detached states, ref changes across graph reads, and overflowing JSON exponents. Real English/German terminal flows retain the original repo during an all-remote fetch when the work view changes selection. Real CLI runs report a raced scan with exit 1 and a JSON overflow with exit 2.

0.30.2

System Python Git Result
macOS 26.7.1, arm64 3.15.0 2.56.0 828 tests and six identical images
macOS 26.7.1, arm64 3.9.6 2.56.0 144 core checks including 13 integration cases
macOS 26.7.1, arm64 3.14.8 2.56.0 13 integration cases
Linux Mint 22.2, x86_64 3.12.3 2.43.0 774 executed, 54 zsh checks skipped; six identical images

The HEAD OIDs of all 27 demo repositories also match between macOS and Linux. Each regression builds two complete sandboxes to check independence from the temporary root path. The committed submodule URL is relative; local clone setup stays offline.

0.30.1

System Python Git Result
macOS 26.7.1, arm64 3.15.0 2.56.0 827 tests and six identical images
macOS 26.7.1, arm64 3.9.6 2.56.0 143 core checks including twelve integration cases
macOS 26.7.1, arm64 3.14.8 2.56.0 twelve new integration cases
Linux Mint 22.2, x86_64 3.12.3 2.43.0 773 executed, 54 zsh checks skipped; six identical images

Real terminal flows ran in English and German: a selected LF-path commit, an unchanged neighboring literal-backslash-n path, preserved foreign staging, an offline comparison with Git/SSH disabled, and a confirmed push to a temporary local bare remote. Commit suggestions retained disabled module contents. Only test repositories were changed.

0.30.0

System Python Git Result
macOS 26.7.1, arm64 3.9.6 2.56.0 815 tests; Python also pinned for shell subprocesses; six images identical
macOS 26.7.1, arm64 3.15.0 2.56.0 815 tests and build.sh --check pass
macOS 26.7.1, arm64 3.9.6 / 3.14.8 2.54.0 Apple / 2.56.0 131 core checks each, four combinations
Linux Mint 22.2, x86_64 3.12.3 2.43.0 761 executed, 54 skipped because zsh is absent; six images identical

The zsh checks ran on macOS. Matrix output records skipped tests explicitly. A version floor is a code/feature requirement, not evidence that every intermediate release has been run. The JSON reports distinguish full checks from core checks.

Scan measurement

python3 tools/benchmark-scan.py creates one temporary repo, three local bare remotes and either one or 25 branches on the same commit. It records five local scans and five fetch scans for each case; setup is excluded. It contacts no external host and removes the temporary repos. Results include individual times, medians and Git call counts. Use the same interpreter/Git and comparable host load for timing comparisons. Recorded before/after data shows 57 → 33 calls locally and 160 → 136 with fetch for 25 identical tips. The one-branch case stays at 33 / 136. Timing varied with host load; process counts are the stable result. Ref/HEAD rechecks and per-transfer configuration checks are retained. Counts are shared only inside one pinned ref inventory.

The recorded consistency_guard measurement for 0.30.3 adds four checks across the complete graph measurement. Both the one- and 25-branch cases use 37 local or 140 fetch-scan calls. These checks detect changes between remote counters, branch inventory and upstream reads; the scan reports a local measurement error.