Installer reports failure on macOS when WeasyPrint deps missing (false positive masks successful install)

[localevolution]

Environment

• macOS 15.5 (Darwin 25.5.0, arm64)
• Python 3.14.5
• Installer ref: v1.9.6-codex.5 (default)
• System libs glib/pango/cairo not present (typical fresh Mac)

Symptom

bash install.sh exits non-zero with a NoneType traceback from the diagnostic helper:

[INFO] Bootstrapping Python runtime...
[ERROR] Codex SEO runtime bootstrap failed.
Traceback (most recent call last):
  File "<string>", line 10, in <module>
    notes = payload.get("verification", {}).get("notes", [])
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'get'

The install actually completed: venv built, core/visual/google/ocr requirements installed, Playwright Chromium installed. Only WeasyPrint (optional PDF reports) failed because libgobject is missing. The installer just can't tell.

Root cause (three stacked bugs)

1. scripts/verify_environment.py:49 — WeasyPrint pollutes stdout at import time

DEPENDENCIES includes (\"weasyprint\", \"weasyprint\"). check_dependency calls importlib.import_module(\"weasyprint\"). On macOS without GLib, WeasyPrint writes a multi-line warning to stdout during its import side-effects, e.g.:

-----

WeasyPrint could not import some external libraries. Please carefully follow the installation steps before reporting an issue:
https://doc.courtbouillon.org/weasyprint/stable/first_steps.html#installation
https://doc.courtbouillon.org/weasyprint/stable/first_steps.html#troubleshooting 

-----

That text is emitted before the script prints its JSON, so the captured stdout is <weasyprint warning> + <json>.

2. scripts/bootstrap_environment.py:91-98 — parse_json_stdout can't handle mixed output

def parse_json_stdout(step: dict[str, Any]) -> dict[str, Any] | None:
    if not step[\"ok\"] or not step[\"stdout\"].strip():
        return None
    try:
        return json.loads(step[\"stdout\"])
    except json.JSONDecodeError:
        return None

json.loads on the polluted stdout raises JSONDecodeError and returns None. bootstrap_environment then sets verification = None, computes core_ready = bool(None and ...) = False, and reports ok: False even though every required step actually succeeded.

3. install.sh:52 — payload.get(\"verification\", {}) returns None, not {}

notes = payload.get(\"verification\", {}).get(\"notes\", [])

The default in dict.get(key, default) only fires when the key is missing. The bootstrap payload sets \"verification\": None explicitly, so payload.get(\"verification\", {}) returns None and the next .get crashes — masking the real diagnostics the helper was supposed to surface.

Suggested fixes

Fix 1 — Stop importing WeasyPrint inside the verifier (or silence its stdout)

Option A: Move WeasyPrint out of the import-loop check and probe it with a subprocess, redirecting stdout/stderr.

Option B: Wrap the WeasyPrint import in a contextlib.redirect_stdout(io.StringIO()) block inside check_dependency for that package.

Option C: Defer the WeasyPrint dependency check to a dedicated function that captures both streams.

Minimal Option B patch:

import contextlib, io

def check_dependency(module_name: str, package_name: str) -> dict[str, Any]:
    try:
        with contextlib.redirect_stdout(io.StringIO()), contextlib.redirect_stderr(io.StringIO()):
            module = importlib.import_module(module_name)
        version = getattr(module, \"__version__\", None)
        return {\"package\": package_name, \"module\": module_name, \"ok\": True, \"version\": version}
    except Exception as exc:
        return {\"package\": package_name, \"module\": module_name, \"ok\": False, \"error\": str(exc)}

Fix 2 — Make parse_json_stdout tolerant of leading noise

def parse_json_stdout(step: dict[str, Any]) -> dict[str, Any] | None:
    if not step[\"ok\"] or not step[\"stdout\"].strip():
        return None
    text = step[\"stdout\"]
    try:
        return json.loads(text)
    except json.JSONDecodeError:
        pass
    # Fall back: extract the last balanced top-level JSON object.
    start = text.find(\"{\")
    while start != -1:
        try:
            return json.loads(text[start:])
        except json.JSONDecodeError:
            start = text.find(\"{\", start + 1)
    return None

Fix 3 — Defensive .get chain in install.sh

dict.get(k, default) returns default only when k is absent. The bootstrap payload sets verification: None explicitly, so guard with or {}:

# install.sh — print_bootstrap_diagnostics
notes = (payload.get(\"verification\") or {}).get(\"notes\", [])

Same pattern applies to any other .get(..., {}).get(...) chains where the value could be None.

Verification

After applying the three fixes, bash install.sh on a fresh Mac without GLib should:

1. Complete bootstrap and report [OK] Codex SEO installed successfully!
2. Emit [WARN] Optional bootstrap groups failed: report (or similar) to flag the missing PDF capability
3. Leave a working venv that passes verify_environment.py --json with core_ready: true, full_ready: false

Workaround for users hitting this today

The install actually works. Run:

~/.codex/skills/seo/.venv/bin/python ~/.codex/skills/seo/scripts/verify_environment.py

If core_ready: YES, the skill is usable. For PDF reports: brew install pango glib cairo libffi gdk-pixbuf then reinstall weasyprint in the venv.