~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
import '../helpers/safe-fetch-with-auth-to-fetch.js';
/**
* Tests for the log-panel header badge's ERROR severity signal
* (`.ldr-log-indicator--has-error`), added by PR #5648 and corrected in
* its follow-up.
*
* All three defects below were introduced and corrected inside this branch;
* none of them reached main. There was no JS test for the log panel's badge,
* which is why they survived several revisions of the branch. Each `it` below
* names the defect it pins:
*
* 1. The badge selected on the RAW level (`[data-log-type="error"]`),
* but the repo's normalizer folds CRITICAL/FATAL into the `error`
* display category (utils/log-helpers.js). A CRITICAL row therefore
* left the badge purple and titleless while the "Error" FILTER badge
* counted it — the panel disagreeing with itself, on precisely the
* two severities the signal exists for.
* 2. The badge counted only entries currently in the DOM. The header
* number is deliberately floored to the server-known `fetched` count
* because the DOM holds fewer rows, so the badge could report a full
* count with no severity for a run whose errors had been paged out
* or pruned. The count is now a per-research high-water mark, and
* the tooltip says "at least N" while the view is truncated.
* 3. Two other writers of the indicator's textContent (`initializeLogPanel`
* and `recomputeCountersFromDom`) never cleared the error class or
* the title, so a re-init or a recompute could repaint a fresh count
* under a stale red badge. All writers now go through
* `paintLogIndicators`.
*
* Plus the accessibility contract: colour alone must not carry the state,
* and the explanation must not be hover-only (role + aria-label).
*/
let logPanel;
let emptyCounts;
beforeAll(async () => {
await import('@js/utils/log-helpers.js');
emptyCounts = window.LdrLogHelpers.emptyCounts;
window.escapeHtml = (s) => String(s ?? '').replace(/[&<>"']/g, '');
window.URLBuilder = {
researchLogs: (id, limit) =>
`/api/research/${id}/logs${limit ? `?limit=${limit}` : ''}`,
historyLogCount: (id) => `/api/research/${id}/log_count`,
researchLogsExport: (id) => `/api/research/${id}/logs/export`,
};
Object.defineProperty(window, 'location', {
configurable: true,
value: { ...window.location, pathname: '/', search: '', hash: '' },
});
await import('@js/components/logpanel.js');
logPanel = window.logPanel;
});
beforeEach(() => {
// The #research-progress wrapper makes initializeLogPanel treat this
// as a research page — without it the init path bails before it
// paints the indicator, and the two re-init tests below would pass
// vacuously.
document.body.innerHTML = `
<div id="research-progress">
<div class="ldr-collapsible-log-panel">
<div class="ldr-log-panel-header" id="log-panel-toggle">
<i class="ldr-toggle-icon"></i>
<span class="ldr-log-indicator" id="log-indicator">0</span>
</div>
<div class="ldr-log-panel-content collapsed" id="log-panel-content">
<div class="ldr-console-log" id="console-log-container"></div>
</div>
</div>
</div>
`;
window._logPanelState.counts = emptyCounts();
window._logPanelState.renderedIds = new Set();
window._logPanelState.currentFilter = 'all';
window._logPanelState.totalLogs = null;
window._logPanelState.fetchedLogs = null;
window._logPanelState.renderedLimit = null;
window._logPanelState.errorHighWater = 0;
window._logPanelState.initialized = false;
window._logPanelState.connectedResearchId = null;
// initializeLogPanel pre-fetches logs (and the row count) for a
// non-null research id; stub both so the tests never touch the
// network and never leave an unhandled rejection behind.
globalThis.fetch = vi.fn(() =>
Promise.resolve({ ok: true, json: () => Promise.resolve([]) })
);
});
/** Append one rendered row of the given RAW severity to the container. */
function seed(rawLevel, message = `msg-${rawLevel}`) {
const container = document.getElementById('console-log-container');
const entry = document.createElement('div');
entry.className = `ldr-console-log-entry ldr-log-${rawLevel}`;
entry.dataset.logType = rawLevel;
entry.dataset.logId = `${rawLevel}-${message}`;
const span = document.createElement('span');
span.className = 'ldr-log-message';
span.textContent = message;
entry.appendChild(span);
container.appendChild(entry);
return entry;
}
function badge() {
return document.querySelector('.ldr-log-indicator');
}
describe('header badge severity — display-category normalization', () => {
// Defect 1. Before the fix the selector was
// '[data-log-type="error"], .ldr-log-error'
// which matches neither of these rows, so both assertions failed.
it.each(['critical', 'fatal'])(
'turns red for a %s entry (folded into the error category)',
(level) => {
seed(level);
logPanel._updateLogCountIndicator();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(true);
expect(badge().title).toContain('1 error');
}
);
it('turns red for a plain error entry', () => {
seed('error');
logPanel._updateLogCountIndicator();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(true);
});
it('stays neutral — no class, no title, no ARIA — with no errors', () => {
seed('info');
seed('warning');
seed('milestone');
logPanel._updateLogCountIndicator();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(false);
expect(badge().title).toBe('');
expect(badge().hasAttribute('role')).toBe(false);
expect(badge().hasAttribute('aria-label')).toBe(false);
});
it('counts critical and error together, matching the Error filter badge', () => {
seed('error', 'a');
seed('critical', 'b');
seed('fatal', 'c');
seed('info', 'd');
logPanel._updateLogCountIndicator();
expect(badge().title).toContain('3 errors');
});
});
describe('header badge severity — entries outside the rendered window', () => {
// Defect 2. The engine-self-disable ERROR this feature exists to
// surface is emitted at init, so it is the row most likely to be
// dropped from the fetched window or evicted by the ordered prune.
it('keeps the signal after the error row leaves the DOM', () => {
const errorRow = seed('error');
seed('info');
logPanel._updateLogCountIndicator();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(true);
// The row is paged out / pruned; only routine rows remain.
errorRow.remove();
logPanel._updateLogCountIndicator();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(true);
expect(window._logPanelState.errorHighWater).toBe(1);
});
it('words the tooltip as a lower bound while the view is truncated', () => {
seed('error');
// The server holds far more rows than the DOM renders.
window._logPanelState.totalLogs = 9002;
window._logPanelState.fetchedLogs = 500;
logPanel._updateLogCountIndicator();
expect(badge().title).toContain('at least 1 error');
// The number itself is still the floored server count, unchanged
// by the severity work.
expect(badge().textContent).toBe('500');
});
it('does not claim a lower bound when the DOM holds everything', () => {
seed('error');
window._logPanelState.totalLogs = 1;
window._logPanelState.fetchedLogs = 1;
logPanel._updateLogCountIndicator();
expect(badge().title).toContain('1 error');
expect(badge().title).not.toContain('at least');
});
it('does not carry the previous research’s errors across a switch', () => {
seed('error');
logPanel._updateLogCountIndicator();
expect(window._logPanelState.errorHighWater).toBe(1);
// Research switch: initializeLogPanel resets the per-research state.
document.getElementById('console-log-container').innerHTML = '';
logPanel.initialize('research-b');
expect(window._logPanelState.errorHighWater).toBe(0);
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(false);
});
});
describe('header badge severity — every writer clears the state', () => {
// Defect 3. initializeLogPanel and recomputeCountersFromDom wrote
// textContent directly and left the class/title untouched, so a badge
// that had legitimately gone red for research A kept its colour and
// tooltip under B's count.
it('recomputeCountersFromDom clears a stale red badge', () => {
badge().classList.add('ldr-log-indicator--has-error');
badge().title = 'stale tooltip from a previous paint';
badge().setAttribute('role', 'img');
badge().setAttribute('aria-label', 'stale');
seed('info');
logPanel._recomputeCountersFromDom();
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(false);
expect(badge().title).toBe('');
expect(badge().hasAttribute('aria-label')).toBe(false);
});
it('initializeLogPanel clears a stale red badge', () => {
badge().classList.add('ldr-log-indicator--has-error');
badge().title = 'stale tooltip from a previous paint';
// No researchId → no count fetch, just the synchronous "0" paint.
logPanel.initialize(null);
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(false);
expect(badge().title).toBe('');
});
});
describe('header badge severity — accessibility', () => {
// The blocked state was conveyed by colour alone (--error-color vs
// --accent-primary: two similar-luminance mid-tones, both with white
// text) and the explanation lived only in a hover `title` on a bare
// span. The glyph lives in CSS (::after) so textContent stays a bare
// number for updateFilterCounters; the text alternative lives here.
it('exposes the severity to assistive tech, not just on hover', () => {
seed('critical');
logPanel._updateLogCountIndicator();
expect(badge().getAttribute('role')).toBe('img');
const label = badge().getAttribute('aria-label');
expect(label).toContain('log entries');
expect(label).toContain('1 error');
});
it('keeps textContent a bare number so the All badge can parse it', () => {
seed('error');
seed('info');
logPanel._updateLogCountIndicator();
expect(badge().textContent).toBe('2');
expect(badge().textContent).not.toContain('!');
});
});
describe('header badge severity — a header without a console container', () => {
// The success paths all repaint through recomputeCountersFromDom on the
// emptied container, so only a DOM that has the indicator but not the
// container isolates initializeLogPanel's own paint. Reverting that
// call leaves the previous research's red badge, tooltip and ARIA label
// sitting under a stale count.
it('repaints the badge on init when the container is absent', () => {
document.body.innerHTML = `
<div id="research-progress">
<div class="ldr-collapsible-log-panel">
<div class="ldr-log-panel-header" id="log-panel-toggle">
<i class="ldr-toggle-icon"></i>
<span class="ldr-log-indicator" id="log-indicator">7</span>
</div>
<div class="ldr-log-panel-content collapsed" id="log-panel-content"></div>
</div>
</div>
`;
expect(document.getElementById('console-log-container')).toBeNull();
badge().classList.add('ldr-log-indicator--has-error');
badge().title = 'at least 7 errors in this research';
badge().setAttribute('role', 'img');
badge().setAttribute('aria-label', '7 log entries, at least 7 errors');
// No researchId → no count fetch, just the synchronous "0" paint.
logPanel.initialize(null);
expect(badge().textContent).toBe('0');
expect(
badge().classList.contains('ldr-log-indicator--has-error')
).toBe(false);
expect(badge().title).toBe('');
expect(badge().hasAttribute('aria-label')).toBe(false);
});
});