feat(hub): pin the app controls while a library scrolls

Files, Videos, Music and Photos are read by scrolling, and everything that
steers that reading left with the first screenful. Three bands now pin under
the navigation bar, in a group and on the Search page alike: the tab bar (the
search field, on Search), the application's own toolbar, and the file table's
column heads. The group's name and description still scroll — they say nothing
a reader needs while walking a directory, and the height they would cost is
height the list does not get.

A band's offset is the heights of the bands above it, and those are not
constants: the toolbar wraps to three rows on a phone, grows a field while a
folder is being named, and loses its filter on Search. So each band measures
itself and publishes `--chrome-h` / `--toolbar-h` (static/sticky.js) and the
stylesheet does the arithmetic in calc(), rather than a number written down
twice — the fault CLAUDE.md already records against this layout twice over.
A band publishes height *plus its own bottom margin*, and paints that margin
as a ring of page colour, so the pinned layout is pixel-identical to the flow
layout and nothing shifts at the moment a band pins.

Three overflow faults came out of it, all of the same class and all of them
what "the header does not stay" actually meant on Android — a document wider
than the screen leaves everything pinned attached to a viewport the reader can
no longer see, the navigation bar included:

  - a directory's name cell was a bare <td>, so an unbreakable folder name
    (`Rage_Against_The_Machine_Discography_1992-2000_FLAC`) set the column's
    minimum: a 527px table in a 390px window
  - Search's group column did the same at 442px with an underscored group
    name. It also goes entirely below 768px, where there is no room for it and
    the breadcrumb already names the group
  - the shared-directories table has four columns of controls with a combined
    minimum near 440px, none of it compressible. On a phone the row stops
    being a row: the name and its eject/remove pair on one line, the two
    switches — each carrying the column head's own string as a label — on the
    next
  - and, found by measuring at 360px, the tab bar itself was 19px too wide

`.file-table` moves to separated borders: a collapsed border belongs to the
table rather than to the cell, so the column heads lost their rule the moment
they pinned.

Measured, not read. tests/harness/sticky_header_probe.py drives the shipped
GroupPage and SearchPage against a stub node, walks to each application,
scrolls to the end and reports every rectangle — 11 views x 4 widths x 2
engines. Its fixture says what real data says: the first version used
`note-007.txt` and `un groupe`, which fit any screen, and found none of the
above. A fixture narrower than real data tests the fixture.

Also: `test_desktop_shell` no longer looks for the CSP after the first `-->`,
which made it fail on correct markup as soon as a comment was added above it,
and `search-page.js` joins test_hook_ordering's file list.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tx16FhyD2BUdpooGb5jcyN

1 files changed, 297 insertions, 0 deletions

diff --git a/packages/meshbay-hub/tests/test_sticky_header.py b/packages/meshbay-hub/tests/test_sticky_header.py
new file mode 100644
index 0000000..32b8124
--- /dev/null
+++ b/packages/meshbay-hub/tests/test_sticky_header.py
@@ -0,0 +1,297 @@
+"""
+Scrolling a library does not cost you the controls that steer it.
+
+Files, Videos, Music and Photos are read by scrolling, and everything that says
+*what* is being read — which application, where in the tree, which filter,
+which column sorts — used to leave with the first screenful. Those controls pin
+under the navigation bar now, in three bands: the group's tab bar (the search
+field, on the Search page), the application's own toolbar, and the file table's
+column heads.
+
+Two of the three offsets are `calc()` over a height nothing declares: the
+toolbar wraps to three rows at a phone width, grows a field while a folder is
+being named, and loses its filter on the Search page, so sticky.js measures it
+and publishes `--chrome-h` / `--toolbar-h`. That is exactly the arrangement
+CLAUDE.md has been bitten by twice — two subtractions in different files, each
+correct on its own, the page a few pixels wrong at every window size and
+nothing in either file to show it. A number in a stylesheet cannot be read to
+find that out. So this scrolls the shipped pages in a real browser and measures
+the rectangles: every band pinned, edge to edge, nothing overlapping, nothing
+showing between.
+
+Three widths, for three different reasons. At 1100 the toolbar is one row; at
+420 it is three, and a stack that only ever adds up on a desktop is the whole
+class of fault this is written against; 390 is an actual handset. And both
+engines: Firefox is half of MeshBay's readers, `position: sticky` is exactly
+the kind of thing engines disagree about, and the first report that a band did
+not pin at all came from a phone.
+"""
+
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+HARNESS = Path(__file__).parent / "harness" / "sticky_header_probe.py"
+STATIC = Path(__file__).resolve().parents[1] / "src" / "meshbay_hub" / "static"
+
+pytestmark = pytest.mark.skipif(
+    shutil.which("google-chrome") is None or not (STATIC / "style.css").exists(),
+    reason="Chrome or the SPA stylesheet is not available")
+
+ENGINES = ["chrome"] + (["firefox"] if shutil.which("firefox") else [])
+
+# A rectangle read back from the browser is rounded, and a border can land
+# either side of a half pixel. One pixel of slack, never more — the faults this
+# exists for are tens of pixels.
+SLACK = 1
+
+# Eleven views (seven in a group, four on the Search page) at four widths.
+EXPECTED_CASES = 44
+
+
+@pytest.fixture(scope="module", params=ENGINES)
+def measured(request):
+    proc = subprocess.run(["python3", str(HARNESS), "--engine", request.param],
+                          capture_output=True, text=True, timeout=900)
+    assert proc.returncode == 0, (
+        f"probe failed in {request.param}: {proc.stdout}{proc.stderr}")
+    cases = json.loads(proc.stdout)
+    assert cases, f"the probe measured nothing in {request.param}"
+    return cases
+
+
+def test_every_view_was_reached(measured):
+    """
+    A case that never got to its application reports an error rather than an
+    empty measurement — and an empty measurement would satisfy every assertion
+    below by having nothing to assert on.
+    """
+    broken = [f"{c.get('engine')} {c['name']}: {c['error']}"
+              f"\n  console: {c.get('logs')}"
+              f"\n  on screen: {c.get('text')}"
+              for c in measured if "error" in c]
+    assert not broken, "\n".join(broken)
+    assert len(measured) == EXPECTED_CASES, (
+        f"expected {EXPECTED_CASES} measurements, got {len(measured)}")
+
+
+def test_the_page_really_scrolled(measured):
+    """
+    Everything below compares a scrolled page with an unscrolled one. A view
+    that fits its window compares it with itself and proves nothing — so this
+    fails the fixture, not the layout, before a silent pass can happen.
+    """
+    flat = [c["name"] for c in measured if c["scrollY"] <= 0]
+    assert not flat, "nothing scrolled, so nothing was tested: " + ", ".join(flat)
+
+
+def test_the_bands_stack_under_the_navigation_bar(measured):
+    """
+    From the top: the navigation bar, then each band, each starting exactly one
+    gap below the one above it — the gap being the upper band's own
+    `margin-bottom`, which it keeps when it pins and paints as a ring. Land
+    short of it and two bands come into contact, which is what the first
+    version shipped and what was reported: two rounded panels flush against
+    each other in Files, and the row of controls glued to the tab bar in
+    Videos, Music and Photos. Land past it and a stripe of the list shows
+    through between them.
+    """
+    faults = []
+    for case in measured:
+        after = case["after"]
+        edge = after[".nav"]["bottom"]
+        for selector in case["bands"]:
+            band = after[selector]
+            if abs(band["top"] - edge) > SLACK:
+                faults.append(
+                    f"{case['engine']} {case['name']}: {selector} pinned at "
+                    f"y={band['top']}, expected y={edge}")
+            edge = band["bottom"] + band["gap"]
+    assert not faults, "\n".join(faults)
+
+
+def test_a_band_keeps_the_gap_it_has_in_the_flow(measured):
+    """
+    Pinning must not change the spacing. Measured against the same page before
+    it was scrolled: the distance from a band to the next one down is the same
+    whether it is pinned or sitting in the flow, so nothing shifts at the
+    moment it pins.
+    """
+    faults = []
+    for case in measured:
+        for upper, lower in zip(case["bands"], case["bands"][1:]):
+            flow = case["before"][lower]["top"] - case["before"][upper]["bottom"]
+            pinned = case["after"][lower]["top"] - case["after"][upper]["bottom"]
+            if abs(flow - pinned) > SLACK:
+                faults.append(
+                    f"{case['engine']} {case['name']}: {upper} → {lower} is "
+                    f"{flow}px apart in the flow and {pinned}px pinned")
+    assert not faults, "\n".join(faults)
+
+
+def test_bands_stay_pinned_at_the_bottom_of_the_page(measured):
+    """
+    A sticky element only sticks inside its own parent's box, so a band whose
+    containing block ends before the page does comes unstuck partway down —
+    invisible to any test that scrolls a fixed amount. Scrolled to the very
+    end, every band is still where it was.
+    """
+    faults = []
+    for case in measured:
+        for selector in case["bands"]:
+            if abs(case["bottom"][selector]["top"]
+                   - case["after"][selector]["top"]) > SLACK:
+                faults.append(
+                    f"{case['engine']} {case['name']}: {selector} is at "
+                    f"y={case['after'][selector]['top']} partway down and "
+                    f"y={case['bottom'][selector]['top']} at the end")
+    assert not faults, "\n".join(faults)
+
+
+def test_each_band_stays_whole(measured):
+    """
+    Pinned and *entire*: a band clipped by the navigation bar above it, or with
+    its own content spilling out of the rectangle it reserved, is not visible
+    just because its top edge is in the right place.
+    """
+    faults = []
+    for case in measured:
+        for selector in case["bands"]:
+            before, after = case["before"][selector], case["after"][selector]
+            if after["height"] <= 0:
+                faults.append(f"{case['name']}: {selector} has no height at all")
+            elif abs(after["height"] - before["height"]) > SLACK:
+                faults.append(
+                    f"{case['name']}: {selector} is {after['height']}px pinned "
+                    f"but {before['height']}px in the flow — pinning resized it")
+            if after["top"] < case["after"][".nav"]["bottom"] - SLACK:
+                faults.append(
+                    f"{case['name']}: {selector} runs up behind the navigation bar")
+    assert not faults, "\n".join(faults)
+
+
+def test_the_list_is_what_moves(measured):
+    """
+    The bands hold still and the content goes past them — not the other way
+    round, and not everything holding still because the page never moved. The
+    content is checked to have travelled by the full scroll: a band that
+    dragged its list along with it would show up here as a short journey.
+    """
+    faults = []
+    for case in measured:
+        selector = case["content"]
+        before, after = case["before"][selector], case["after"][selector]
+        travelled = before["top"] - after["top"]
+        if abs(travelled - case["scrollY"]) > SLACK:
+            faults.append(
+                f"{case['name']}: {selector} moved {travelled}px while the window "
+                f"scrolled {case['scrollY']}px")
+    assert not faults, "\n".join(faults)
+
+
+def test_the_offsets_come_from_the_measurement(measured):
+    """
+    The published heights are the bands' own, not a number that happens to
+    agree at one width. This is the assertion that fails if sticky.js stops
+    observing — a stale `--toolbar-h` still stacks perfectly at the width it
+    was measured at, and only at that one.
+    """
+    faults = []
+    for case in measured:
+        bands = case["bands"]
+        upper = case["after"][bands[0]]
+        chrome = upper["height"] + upper["gap"]
+        if case["chromeH"] != f"{chrome}px":
+            faults.append(f"{case['engine']} {case['name']}: --chrome-h is "
+                          f"{case['chromeH']!r}, {bands[0]} is {upper['height']}px "
+                          f"tall over a {upper['gap']}px gap")
+        # Only Files pins anything below its toolbar, so only Files publishes
+        # a toolbar height; the others leave the property withdrawn.
+        if ".file-table th" in bands:
+            mid = case["after"][bands[1]]
+            toolbar = mid["height"] + mid["gap"]
+            if case["toolbarH"] != f"{toolbar}px":
+                faults.append(f"{case['engine']} {case['name']}: --toolbar-h is "
+                              f"{case['toolbarH']!r}, {bands[1]} is "
+                              f"{mid['height']}px tall over a {mid['gap']}px gap")
+        elif case["toolbarH"] not in ("", "0px"):
+            faults.append(f"{case['engine']} {case['name']}: --toolbar-h left "
+                          f"behind as {case['toolbarH']!r} by a view with no table")
+    assert not faults, "\n".join(faults)
+
+
+def test_no_view_scrolls_sideways(measured):
+    """
+    Nothing here may be wider than the window.
+
+    This is the fault that was reported from a phone, and it presented as the
+    sticky header not working at all — including the navigation bar, which had
+    been `position: sticky` since long before any of this. That is the tell: on
+    Android a document wider than the screen leaves everything pinned attached
+    to a viewport the reader can no longer see, so a header that is doing
+    exactly what it was told looks like a header that was never pinned.
+
+    The cause was a table column, and the reason no measurement here found it
+    first is worth keeping: the fixture said `note-007.txt` and `un groupe`,
+    which fit any screen. A fixture narrower than real data tests the fixture.
+    It now carries the names a music library actually has, and walks into a
+    folder, which is where the rows are directories — and a directory's name
+    cell was the one that had no wrapping rule on it.
+    """
+    faults = []
+    for case in measured:
+        if case["overflowX"] > 0:
+            faults.append(
+                f"{case['engine']} {case['name']}: the page is "
+                f"{case['overflowX']}px wider than its window"
+                + (f" — widest: {'; '.join(case['widest'])}" if case["widest"] else ""))
+    assert not faults, "\n".join(faults)
+
+
+def test_the_group_column_is_dropped_at_phone_widths(measured):
+    """
+    Search's group column goes at phone widths, where there is no room for it
+    and the file name is what the screen is short of — the same treatment the
+    type and date columns already get.
+
+    Nothing is lost with it, which is why it was the column to drop: the Search
+    page re-roots every result under a folder named after its group, so a
+    reader is always inside exactly one group and the breadcrumb above the
+    table names it. The column repeated that on every row.
+
+    Both halves are asserted, and the wide half is the one that matters: a
+    column that stopped rendering altogether would satisfy "hidden on a phone"
+    perfectly.
+    """
+    present = {c["name"]: (c["width"], c["groupColumn"]) for c in measured
+               if c["groupColumn"] != "absent"}
+    assert present, "no case rendered a group column at all"
+    wide = {n: g for n, (w, g) in present.items() if w > 768}
+    narrow = {n: g for n, (w, g) in present.items() if w <= 768}
+    assert wide and narrow, f"need both sides of the breakpoint, got {present}"
+    assert all(g == "shown" for g in wide.values()), (
+        f"the group column is missing on a wide screen: {wide}")
+    assert all(g == "hidden" for g in narrow.values()), (
+        f"the group column still takes room on a phone: {narrow}")
+
+
+def test_the_toolbar_wraps_at_a_phone_width(measured):
+    """
+    The reason the heights are measured rather than written down. If the
+    toolbar were the same height at 420 as at 1100, the whole mechanism would
+    be arithmetic nobody needs — and this test would be the one to say so
+    before the next reader replaces it with a constant.
+    """
+    tall = {c["view"]: c["after"][".file-toolbar"]["height"]
+            for c in measured if c["width"] == 420 and ".file-toolbar" in c["bands"]}
+    wide = {c["view"]: c["after"][".file-toolbar"]["height"]
+            for c in measured if c["width"] == 1100 and ".file-toolbar" in c["bands"]}
+    assert tall and tall.keys() == wide.keys()
+    for view in tall:
+        assert tall[view] > wide[view], (
+            f"{view}: the file toolbar is {tall[view]}px at 420 and "
+            f"{wide[view]}px at 1100 — it no longer wraps, and the column heads "
+            f"could pin against a constant")