mosaic-headless 1.26.0 → 1.27.0

package/references/failure-modes.md

@@ -344,3 +344,58 @@ Two rules come out of it:
   that lesson: for anything with a visible and an invisible state, a suite that
   only exercises the visible one is testing half a component, and the half it
   skips is the one the visitor sees most of the time.
+
+## A fragment link in a SHARED header is dead on every page but one
+
+`#services` points at an id in the CURRENT document. Put it in a header that every
+page shares, and on the one page that has that section it works, while on the other
+seven the browser looks for the id, does not find it, and does nothing — no
+navigation, no error, no feedback of any kind. The visitor clicks and the page sits
+there. It is worse than a 404 for exactly that reason.
+
+This shipped. The nav verifier had a check for it and the check could not see it:
+it fetched every menu href and asserted HTTP 200, and `#services` resolves to the
+*current* URL, so it returned 200 on all eight pages while being dead on seven.
+"Every link answers 200" and "every link does something" are different claims.
+
+Two things come out of it:
+
+- **Write shared-shell links absolutely** — `/#services`, not `#services`. A link
+  in a component that appears on more than one page has to mean the same thing
+  from all of them.
+- **Assert the fragment resolves on the page that carries it.**
+  `verify_navigation.py` now opens every page and fails on any `[href^="#"]`
+  whose target id is absent from that document. It found the four in the header
+  immediately — and then a fifth and sixth in the masthead, which `/intro/` shares
+  with the homepage and which I had not thought to look at.
+
+The general form: a check that confirms a resource EXISTS is not a check that the
+control WORKS, and anything living in a shared layout needs verifying on every
+page that includes it, not on the page it was designed for.
+
+## Three ways a form comes back empty, none of them an error
+
+A form page built in one pass rendered two fieldsets, one stray label and **zero
+controls**. The commit succeeded, the response carried no exception, and the HTML
+was valid. Every field had been dropped on the way in, for three different reasons:
+
+- **`label` and the inputs need a `field` above them.** They declare
+  `canBeNestedChildFor` and walk the parent chain looking for one; committed under
+  a plain div they are refused and discarded. `placement-rules.csv` now carries a
+  `nested_under` column naming exactly what each type is looking for — 54 types
+  have one, and the names rarely share a prefix (`radio-input` wants `choice`,
+  `label` wants `field`, `loop-pagination` wants `loop`).
+- **`fieldset` does not take fields.** It renders `fieldset-legend` +
+  `fieldset-content`, and direct children are dropped. Neither type was in
+  `node-types.csv` at all, because they are declared by an
+  `*ElementMResourceFactory.php` and every extractor here walked
+  `*ElementTypeFactory.php` only. Two types were missing from the catalog; the
+  walk now covers both and the count went 126 → 128.
+- **`select-option` is a leaf and its label is the `text` PROPERTY.** Give it a
+  `wysiwyg-text` child and the option renders with an empty label: a dropdown that
+  opens onto five blank lines, which looks like a styling problem and is not.
+
+The through-line is the one this file keeps recording: Mosaic accepts a write it
+cannot use and says nothing. The defence is to read the delivered page back and
+count what arrived — `verify_components.py` now asserts the control count, the option
+labels and that an empty required form is actually refused.

package/references/placement.md

@@ -115,3 +115,41 @@ Reach for the measured table when you are deciding what to build with, and this
 when you are deciding what may go inside what. `tools/check_placement_predicts.py`
 re-scores the numbers above after any re-sweep, so the claim stays honest if the
 plugin changes.
+
+## `nested_under`: what a type is looking for above it
+
+`rule` is the `canBeParentFor()` answer — "will the editor let me drop this here".
+Many types also implement `canBeNestedChildFor()`, which walks the parent chain at
+runtime looking for ONE ancestor class, and until now this table only recorded
+that such a rule existed. The `nested_under` column now names it. 54 of 128 types
+have one, and the names usually share no prefix with the type, so there is no way
+to infer it:
+
+| type | needs above it |
+|---|---|
+| `radio-input`, `checkbox-input`, `choice-label` | `choice` |
+| `choice`, | `choice-field` |
+| `label`, `fileupload-input` | `field` |
+| `choice-field`, `fieldset`, `field`, `hidden-input` | `choice-field` \| `field` \| `form` |
+| `loop-item` | `loop-items` |
+| `loop-pagination` | `loop` |
+| `list-item` | `list` \| `list-loop-items` \| `wp-menu` |
+| `accordion-item` | `accordion` \| `accordion-loop-items` |
+
+A node committed without its required ancestor is **accepted and discarded**. The
+commit returns no exception, the page renders, and the element is simply not
+there — which is how a form came back with two fieldsets and no fields.
+`build_page.py` reads this column and now carries the ancestor chain down the walk,
+so a correct placement is allowed and an incorrect one is refused before the write
+rather than silently lost after it.
+
+Two other things the table was short of, both fixed the same way — by reading the
+source the column was derived from rather than the one it was convenient to read:
+
+- **`takes_children`**, from the factory's `getDefaultData($data, $children)`
+  signature. `dropdown-wrapper` answers `none` to `canBeParentFor` and holds the
+  menu's links regardless.
+- **two missing types.** `fieldset-legend` and `fieldset-content` are declared by
+  an `*ElementMResourceFactory.php`; every extractor walked
+  `*ElementTypeFactory.php` only, so they were absent from the catalog and a
+  fieldset appeared to have no inner structure.

package/scripts/check-release.mjs

@@ -101,9 +101,10 @@ const counts = {
   "data/dialog-verification.csv": 19,
   "data/slider-verification.csv": 30,
   "data/token-verification.csv": 15,
-  "data/mobile-audit.csv": 58,
-  "data/component-wall-verification.csv": 11,
-  "data/navigation-verification.csv": 24,
+  "data/mobile-audit.csv": 60,
+  "data/component-wall-verification.csv": 13,
+  "data/form-surface-verification.csv": 8,
+  "data/navigation-verification.csv": 36,
   "data/page-modal-verification.csv": 19,
   "data/interaction-rules-verification.csv": 17,
   "data/dialog-form-verification.csv": 49,

package/sites/_moksa.py

@@ -281,8 +281,12 @@ def mask_line(attr, text, tag="h1", **st):
 # studio's own site; everything that demonstrates what Mosaic can DO is one
 # labelled group, which is also the only honest shape - those five pages are a
 # set, not five more sections of the homepage.
-NAV = [("服務", "#services"), ("作品", "#works"),
-       ("產品", "#products"), ("聯絡", "#contact")]
+# These are sections of the HOMEPAGE, and the header is shared by every page. A
+# bare `#services` is therefore a dead link on all seven of the others: the
+# browser looks for that id in the current document, does not find it, and does
+# nothing at all. Absolute, so they mean the same thing from anywhere.
+NAV = [("服務", "/#services"), ("作品", "/#works"),
+       ("產品", "/#products"), ("聯絡", "/#contact")]
 
 # Read from the shipped tables rather than typed, so the menu cannot claim a
 # number the evidence no longer supports.
@@ -300,6 +304,8 @@ CONVERTED_COUNT = _count("conversion-batch.csv",
 
 # label, href, one line of what it demonstrates
 CAPABILITIES = [
+    ("表單", "/forms/",
+     "下拉、單選、複選、上傳、必填規則與錯誤訊息"),
     ("元件牆", "/components/",
      "分頁、手風琴、表單、地圖——都能直接操作"),
     ("彈窗與觸發", "/dialogs/",
@@ -1460,7 +1466,7 @@ HEADER = box("mk-shell-top", {}, [
              "children": [
                  # the wordmark is the only route back to the top on a one-pager, so
                  # it has to be a link. `menu-link` with a url renders a real <a>.
-                 {"type": "menu-link", "data": {"attrID": "mk-logo", "url": "#mk-home"},
+                 {"type": "menu-link", "data": {"attrID": "mk-logo", "url": "/"},
                   "style": {"&": {"_": {"display": "flex", "alignItems": "baseline",
                                         "columnGap": "9px", "cursor": "pointer"}}},
                   "children":
@@ -1514,7 +1520,7 @@ HEADER = box("mk-shell-top", {}, [
                   ] + [capability_dropdown()]},
                  # not a pill: a bracketed link, the way a document cross-references
                  {"type": "button", "data": {"attrID": "mk-header-cta",
-                                             "url": "#contact"},
+                                             "url": "/#contact"},
                   "style": {"&": {"_m": {"display": "none"},
                                   "_t": {"fontSize": "12px"},
                                   # 13px of the bright orange on the glass header is
@@ -1696,8 +1702,8 @@ MASTHEAD = section("mk-mast", [wrap("mk-mast-in", [
                                     if n == 1
                                     else {"color": {"token": "--mk-accent"}})}},
           "text": text}
-         for n, (text, href) in enumerate([("服務項目 →", "#services"),
-                                           ("精選作品 →", "#works")], start=1)
+         for n, (text, href) in enumerate([("服務項目 →", "/#services"),
+                                           ("精選作品 →", "/#works")], start=1)
      ]},
           ]),
           # the right-hand column: what a spec sheet puts in its header block
@@ -3224,6 +3230,10 @@ if __name__ == "__main__":
     # helpers it reads are pure functions - two copies produce identical dicts.
     import _dialogs
     import _sliders
+    import _forms
+    SITE["pages"].append({"slug": "forms", "post_id": 282,
+                          "title": "表單",
+                          "tree": apply_type(_forms.TREE, DISPLAY, CJK, "600")})
     import _components
     SITE["pages"].append({"slug": "components", "post_id": 281,
                           "title": "元件",