Bug Fix: filtering a model table to zero matches emptied it with no message¶

Summary: Adds the filter-empty state to model_table.html and makes filterTable() evaluate all active column filters at once, so filtering to nothing says so — and so two filters combine instead of overwriting each other.

🐛 Problem Description¶

Issue Summary¶

Issue #96 reported that the per-column filter inputs did nothing. Most of that was already fixed by #161 (merged 2026-08-20), which repaired the model_table.html ⇄ table_utils.js selector contract. Two of the issue's three acceptance criteria passed on main @ 23ad0ad.

The third did not. Typing a string that matched no rows hid every data row and left the user looking at a header above an empty body, with nothing to say whether the filter was too narrow or the data was missing.

Investigating it surfaced a second, unreported defect in the same function: filterTable only ever considered one column. It took (table, columnIndex, query) and set row.style.display for every row from that one column's value, so typing in a second column silently un-hid everything the first had filtered out. The two are fixed together because the "no matches" message cannot be correct while the match count is computed from one filter.

Affected Components¶

src/parodynews/templates/includes/model_table.html — the shared partial behind every model list, so this affected all of them at once.
src/assets/js/table_utils.js — filterTable().
User experience: rows vanished with no explanation; multiple filters did not combine.

Reproduction Steps¶

Open any model list page with at least one row.
Type zzzzzz into any column filter.
Observed: all rows disappear, no message.
Expected: a visible "No matching records" row, which clears when the filter is relaxed.

For the multi-filter half: filter column A to narrow the list, then type in column B. Observed: rows column A had excluded reappear.

🔍 Root Cause Analysis¶

Root Cause¶

filterTable preserved the server-side {% empty %} row (a single <td> with colspan > 1) but never created an empty state of its own, and no such row existed in the template for it to show. There was no code path that could produce the message.

The multi-column defect had the same shape: the function's signature only admitted one column, so per-column state could not be combined.

Contributing Factors¶

The {% empty %} row made the table look like it had an empty state; the case it covers (server returned nothing) is not the case users hit.
Nothing pinned "filtering to zero matches" in either test suite.

✅ Solution Implementation¶

Fix Description¶

Template — model_table.html now renders a second empty-state row, always present and shipping hidden:

<tr data-filter-empty style="display: none;">
    <td colspan="{{ display_fields|length }}" class="text-center text-muted"
        role="status" aria-live="polite">
        No matching records
    </td>
</tr>

It lives in the template, not in JavaScript, so the message stays translatable and the colspan tracks display_fields. It replaces content that otherwise disappears unannounced, hence role="status".

JavaScript — filterTable(table) now takes only the table and re-reads every input.filter on each keystroke. A row survives if it matches all active filters. The function counts data rows and survivors, then shows the message only when filters.length > 0 && dataRows > 0 && matches === 0 — so it never stacks on top of the server's own "No items found".

Re-reading every filter (rather than only the one that changed) is what makes clearing one filter re-apply the rest instead of revealing rows they still exclude.

style.display is the toggle for both, matching how the rest of the function works and keeping the row out of "visible row" queries while hidden.

🧪 Testing and Validation¶

Test Cases Added¶

src/parodynews/tests/test_model_table.py — FilterEmptyStateMarkupTests, six tests asserting the row renders, starts hidden, spans the displayed columns, carries role="status" / aria-live="polite", appears exactly once, and is present whether or not the server returned rows. These run in the default suite (-m "not e2e"), so the markup half is covered where no browser is available.

src/parodynews/tests/e2e/test_model_table_e2e.py — six Playwright tests for the behaviour: the message appears on zero matches, clears when the filter is relaxed or cleared, appears exactly once with two filters active, is announced to assistive technology, does not stack on the server empty state, and — the multi-filter regression — clearing one of two filters re-evaluates rather than revealing rows the other still excludes.

Two existing helpers (visible_usernames, column_values) now skip single-cell rows, since the partial has a second full-width row for them to trip over.

Test Results¶

Run from src/ (config: src/pytest.ini; e2e is deselected by default):

python -m pytest parodynews/tests/test_model_table.py
python -m pytest -m e2e --browser chromium parodynews/tests/e2e/test_model_table_e2e.py

The behaviour was additionally verified by executing the real table_utils.js against the rendered markup in a jsdom document. Against the new template row with the pre-fix filterTable, six checks fail — the message never clears, the two filters do not combine, clearing one filter reveals everything, and the message stacks on the server empty state — which is what shows the JavaScript change is required and not only the template row.

⚠️ Breaking Changes and Migration¶

filterTable's signature changed from (table, columnIndex, query) to (table). It is not exported and the only call site is its own input handler, so nothing else needs updating. Any hand-rolled table calling it directly must drop the extra arguments; the markup contract is unchanged and is documented in src/assets/js/README.md.

No migrations, no configuration changes.

🔄 Prevention Measures¶

Both empty states are now described side by side in src/assets/js/README.md and src/parodynews/templates/includes/README.md, with what each one means.
The template's own contract comment states that removing the row removes the message, since table_utils.js only toggles it.
The markup half is asserted in the default (non-e2e) suite, so it cannot regress unnoticed on a build without browsers.

Original bug report: #96 (parent #95)
Prior fix for the same partial: #161, opened against #87
Partial: src/parodynews/templates/includes/model_table.html
Script: src/assets/js/table_utils.js