Refs #1745 -- Harden nav/hooks regression guard

[BaderEddineBenhirt]

This PR hardens the nav/hooks regression guard.

Changes:

• installs the hydration MutationObserver before the initial navigation using add_init_script.
• replaces fixed asyncio.sleep waits with condition-based Playwright waits.
• adds a dedicated nav-hooks-guard CI job so this regression guard is blocking while the broader Playwright job remains optional.

Tested locally:

• .venv\Scripts\python tests\playwright\test_nav_hooks.py
• git diff --check

[johnrtipton]

Thanks @BaderEddineBenhirt for the contribution. I'll review run it through the review process.
What has your experience with djust been? Rough, smooth, problems?

[BaderEddineBenhirt]

Thanks @johnrtipton for the quick response. My experience was mostly smooth. The setup with uv and the demo project was straightforward, and the existing Playwright regression script made the issue easy to understand and reproduce locally. The only slightly rough part was understanding which Playwright checks are intended to be blocking versus optional in CI. That is why I kept the broader Playwright job optional and added a dedicated blocking guard only for the nav/hooks regression.

Overall, the repo was clear to work with.

[johnrtipton]

Code Review (pipeline-run Stage 7) — APPROVE with follow-ups

Solid, well-motivated hardening of the #1745 nav/hooks regression guard. All 14 CI checks green (incl. the new nav-hooks-guard, 2m9s). No 🔴 blockers.

Strengths

• Condition-based waits replace fixed asyncio.sleep — waits on real signals (__demoWidgetMountCount, the djustHookMounted DOM marker, location.pathname, mount-count delta) instead of wall-clock. Faster and far less flaky.
• Observer installed before goto via add_init_script correctly catches the initial hydration churn (#1737) a post-goto evaluate() could miss; the late-root handling (document observer + DOMContentLoaded re-check) is sound.
• Not greenwashing: the script return 1 on any collected failure → sys.exit(exit_code) — a real failing gate (verified).
• Made blocking correctly: added to test-summary needs AND the success AND-condition — it genuinely gates, not just echoed.

🟡 Follow-ups (non-blocking — tracked, not gating this merge)

1. nav-hooks-guard is the only Rust-dependent CI job without maturin develop --release. Every other ext-dependent job runs uv sync + maturin develop. This job relies on uv's .venv cache (key = hashFiles(pyproject.toml, uv.lock)), so a PR that changes only crates/*.rs (no Python dep change) would test against a stale cached extension — a blocking render guard that can silently pass on stale code. Recommend adding the maturin develop --release step. (Maintainer fast-follow.)
2. Blocking Playwright job = flake risk to every PR's gate. playwright-tests is intentionally optional because browser jobs flake; promoting a Playwright job to blocking reintroduces that gate-wide. The condition-based waits mitigate it (good); per repo canon a new runner-only job ideally soaks continue-on-error before flipping. Green once here, so low risk — flagged for awareness.
3. ~80% CI duplication with playwright-tests (server harness). Consider a composite action / reusable workflow to prevent drift.

Minor

• Run step uses .venv/bin/python while siblings use uv run (cosmetic).
• Confirm new action pins (checkout@v6, setup-uv@v7, cache@v5, upload-artifact@v7) match versions used elsewhere in this workflow.

Merging; 🟡#1–3 filed as a follow-up. Thanks for the contribution!

APPROVE

[johnrtipton]

Retrospective — PR #1748 (refs #1745)

Task: harden the nav/hooks Playwright regression guard (condition-based waits + observer-before-nav) and promote it to a blocking CI job. External contribution by @BaderEddineBenhirt.

Quality: 4/5

Well-motivated, correctly-implemented hardening; merged green with no 🔴. Held back from 5 only by the blocking-gate-without-maturin gap (#1752) that a maintainer fast-follow should close.

What went well

• Replacing fixed asyncio.sleep with condition-based wait_for_function is exactly the right fix for timing-flaky browser tests — and it waits on real post-conditions (hook-mount markers, path change, mount-count delta), not wall-clock. Same lesson as the async-callback-stub canon: assert the state, don't sleep-and-hope.
• Observer-before-navigation via add_init_script closes a real coverage hole: a post-goto evaluate() can miss the initial hydration churn (Initial SSR render skips the comment/whitespace normalization that render_with_diff applies → first-hydration morphChildren re-render (flash) #1737) entirely.
• Verified the guard actually fails (return 1 → sys.exit) before trusting it as blocking — avoids the greenwashing trap (tech-debt: Pre-commit Self-Review should grep for stubbed JSDOM API shapes (greenwashing-catcher) #1037) where a "blocking" gate always exits 0.

What didn't / to improve

• A new Rust-ext-dependent CI job shipped without maturin develop while every sibling job has it — so the blocking render guard can test a stale cached extension on Rust-only PRs (tech-debt: harden nav-hooks-guard CI job (maturin build, blocking-soak, dedup) — follow-up to #1748 #1752 item 1). The reusable-workflow extraction (tech-debt: harden nav-hooks-guard CI job (maturin build, blocking-soak, dedup) — follow-up to #1748 #1752 item 3) would have prevented this by construction (one server-harness definition, one build step). Lesson: when adding a CI job that mirrors an existing one, factor the shared harness rather than copy-paste — the copy drifts (here, the missing build step).
• Promoted straight to blocking without a continue-on-error soak (rc4 Free-threaded hardening / perf polish (follow-up to #1432) #1534). Low risk given green CI + the flake-reducing waits, but the canonical sequence is soak-then-flip.

Verified

• All 14 CI checks green (incl. nav-hooks-guard); 0 failing/pending at merge. Approving review posted (pipeline Stage 7). 🟡 follow-ups tracked in tech-debt: harden nav-hooks-guard CI job (maturin build, blocking-soak, dedup) — follow-up to #1748 #1752.

RETRO_COMPLETE

[johnrtipton]

@BaderEddineBenhirt thanks for the PR, this has shipped in 1.0.3.