[lfoletti]

Why

A real audit of a court database built with Angular CLI showed two faults in sap probe:

• The served HTML is an empty shell: <app-root></app-root>, module scripts, 1 visible word. Modern Angular builds do not write ng-version in the served HTML, and the inlined CSS made the shell 130 KB (other shells pass 200 KB). The probe reported SPA probable : False.
• The site has a small sitemap, so the probe chose tier 1 « Sitemap ». Every address in that sitemap serves the same empty shell. The usable path was the data API that the application calls (found by reading main-*.js and the chunks as text, as the skill says).

What

• visible_words(html): the words outside scripts, styles, noscript and template.
• Angular is also recognised by <app-root.
• SPA: (a known framework, or an empty mount element with module scripts) and fewer than 50 visible words. The old rule stays, with a bound on visible words.
• _decide_tier_doors: a SPA goes to tier 4 before the sitemap, with the hint « chercher l'API de données dans le code de l'application (scripts lus comme texte, jamais exécutés), sinon Chrome » and a note that the sitemap only lists addresses.

Failing first

Without the change in accesspath/probe.py, 3 of the 4 tests of tests/test_spa_coquille.py fail (the fourth checks that a page with text is not a SPA):

git stash push -- accesspath/probe.py && python -m pytest -q tests/test_spa_coquille.py; git stash pop

Full suite: 781 passed. Live probe on the audited site: tier 4, frameworks: ['angular'], SPA probable : True.