feat(coordinator): testers' details are for BigBeans; everyone else sees the testers in numbers #120
No reviewers
Labels
No labels
kind
bug
kind
chore
kind
docs
kind
feature
kind
performance
kind
question
priority
critical
priority
high
priority
low
risk
high
risk
low
risk
medium
status
blocked
status
confirmed
status
duplicate
status
invalid
status
needs-info
status
wontfix
ci: approved
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Beanstalk!120
Loading…
Reference in a new issue
No description provided.
Delete branch "HumanBean/Beanstalk:feat/tester-privacy"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
The Testers page (
/agents) showed every tester's login, PC name, GPU model, driver, status and history to anyone. Now:/agents, the health page's tester table and its "Needs a look" notes that name testers, and any tester's page./agents/<id>).web.tester_counts): registered, seen this week and online, what the online ones are doing (testing, ready, waiting, paused, a problem), and how many have each OS, GPU maker and kind of PC (dGPU, iGPU, Steam Deck, handheld), online and this week. No login, PC name, GPU model or driver. Hidden testers count, as they always did. A tester's page answers 403 with a page that says who may see it and offers sign-in. The health page's notes become counts ("3 testers are on an app older than nightly-204").Hidden testers: a BigBean sees them too, marked
hidden, on/agents, the health page and their own page (with their Discord name). For everyone else they stay out of every page, as before, and PR pages and reports say "Hidden tester" for all.PR pages and reports still name the tester who ran each job, as before. The How page no longer says build changes need two approvals (#119).
Screenshots
The dev server's made-up data (beanstalk.devserver), on a throwaway database.
Testing
Tested: the coordinator suite in a python:3.13-alpine container with Postgres 17 on the farm's server: 593 passed, 3 failed, 1 skipped. The 3 (
test_regression.py) fail the same way on main. Newtest_tester_privacy.py: anyone sees no login, name, GPU or link on/agentsand/health, with the counts (the hidden tester counted); the health page counts every registered tester and a PC with two setup warnings once; a hidden tester sees their own page; a tester page is 403 for anyone, 200 for its own tester (who still sees counts on the list) and for a BigBean, who sees both tables. Tests that check tester details now view as a BigBean (bigbean_view).Still to test: after deploy,
/agentssigned out, signed in as a non-BigBean and as a BigBean.🤖 Generated with Claude Code
Recommendation: Request changes. The new health counts misstate registered testers and setup warnings, and hidden testers cannot access their own pages despite the stated access policy. The visual change also lacks the screenshots required by the repository.
The role checks and aggregate view reduce direct exposure of tester details on the affected pages, but the aggregates must be accurate and the promised owner access must work for hidden testers. Add desktop and phone-width screenshots using fictional data.
Does it help: Unproven: the PR reports 590 passing tests, three failures also seen on main, and one skip; the new tests cover basic signed-out, owner and BigBean access, but do not exercise hidden owners, old registrations or multiple warnings per tester. No independent run was performed.
Other systems: The shared counts appear on both /agents and /health; health fetches a different population. Hidden testers are intentionally still included in aggregates. PR and report identities remain public as described. Authentication uses the existing Forgejo session and membership checks.
Switches: None added, changed or removed.
Reviewed:
65a1bc334cagainstd6e717d10d: Read the base AGENTS.md and CONTRIBUTING.md; no docs/review-checklist.md or .forgejo PR template was present. Read the diff, new templates, health template, privacy tests, conftest, health.py, agents.py and auth.py; web.py was truncated by the file tool and the remaining changes, other changed tests and documentation were reviewed from the diff only. No code was run.; nothing run (a code review).Checked and fine:
Blocking (to change before merge)
Testing: Reported suite: 590 passed, 3 failed as on main, 1 skipped. Post-deploy role checks are still outstanding; this review did not execute tests.
Overlaps and merge order: No overlapping open PR found in searches for tester privacy, agents and health.
Automated code review by BeanBot (gpt-6.1-terra), on the pull request's opening. Nothing was run; a maintainer decides.
@ -0,0 +2,4 @@{% from "_tester_counts.html" import tester_counts with context %}{% block title %}Testers{% endblock %}{% block main %}<header class="page-head">Please add the required before/after desktop and phone-width dashboard screenshots to the PR description, using fictional tester data.
@ -433,0 +433,4 @@full = sees_testers(request)if not full:s["tester_rows"], s["hidden_testers"] = [], 0s["setups_n"], s["old_n"] = len(s.get("setups") or []), len((s.get("testers") or {}).get("old") or [])setupshas one entry per warning, not per tester. One PC missing both sandbox and game files is reported as two testers; count distinct agent IDs instead and cover multiple warnings on one PC.@ -433,0 +436,4 @@s["setups_n"], s["old_n"] = len(s.get("setups") or []), len((s.get("testers") or {}).get("old") or [])s["setups"] = []s["testers"] = dict(s["testers"], old=[])return page(request, "health.html", s=s, full=full, counts=None if full else tester_counts(rows))These rows were restricted to the last seven days, so Registered and reports in all on /health omit older active testers and disagree with /agents. Fetch all non-retired rows for these totals and test an older registration.
The hidden check runs before owner authorization, so a hidden tester gets 404 on their own page, contrary to the stated owner access. Allow the signed-in owner while keeping it hidden from other viewers, with a regression test.
65a1bc334cb7bdc6cc50b7bdc6cc508e93dc6cf7WIP: feat(coordinator): testers' details are for BigBeans; everyone else sees the testers in numbersto feat(coordinator): testers' details are for BigBeans; everyone else sees the testers in numbers8e93dc6cf7d0bb8d15b5Answered in
d0bb8d15:8e93dc6c(after the reviewed commit): a hidden tester signed in sees their own page; 404 for everyone else. Tested.