External/5chan
1
0
Fork 0
mirror of https://github.com/bitsocialnet/5chan.git synced 2026-08-03 07:41:04 +02:00
Code Issues Packages Projects Releases Wiki Activity

chore(portless): upgrade local dev URLs

Browse Source
This commit is contained in:
Tommaso Casaburi
2026-04-28 14:25:16 +07:00
parent 8ba1761d07
commit 34b60c7d53
23 changed files with 92 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/agent-playbooks/known-surprises.md
+5 -5
View File
@@ -92,10 +92,10 @@ If uncertain, ask the developer before adding an entry.
- **Date:** 2026-03-30
- **Observed by:** Codex
- **Context:** Running `PORTLESS=0 yarn start` while another local service already owned port `1355`
- **What was surprising:** The non-Portless dev fallback forced Vite onto `5chan.localhost:1355` with `--strictPort`, so the fallback path could fail immediately even though the main Portless flow is collision-safe.
- **Impact:** Contributors could lose the fallback dev path or interrupt their startup flow when `1355` was already busy.
- **Mitigation:** Keep the fallback behind `scripts/start-dev.js`, which now probes from `1355` upward and starts Vite on the next free port instead of exiting.
- **Context:** Running `PORTLESS=0 yarn start` while another local service already owned port `3000`
- **What was surprising:** The non-Portless dev fallback forced Vite onto `5chan.localhost` with `--strictPort`, so the fallback path could fail immediately even though the main Portless flow is collision-safe.
- **Impact:** Contributors could lose the fallback dev path or interrupt their startup flow when `3000` was already busy.
- **Mitigation:** Keep the fallback behind `scripts/start-dev.js`, which now probes from `3000` upward and starts Vite on the next free port instead of exiting.
- **Status:** confirmed
### Fixed Portless app names collide across 5chan worktrees
@@ -105,7 +105,7 @@ If uncertain, ask the developer before adding an entry.
- **Context:** Starting `yarn start` in one 5chan worktree while another 5chan worktree was already serving through Portless
- **What was surprising:** Using the literal Portless app name `5chan` in every worktree makes the route itself collide, even when the backing ports are different, so the second process fails with `"5chan.localhost" is already registered`.
- **Impact:** Parallel 5chan branches can block each other even though Portless is meant to let them coexist safely.
- **Mitigation:** Keep Portless startup behind `scripts/start-dev.js`, which now uses a branch-scoped `*.5chan.localhost:1355` route outside the canonical case and automatically increments a `-2`, `-3`, ... suffix when that branch-scoped route is already occupied.
- **Mitigation:** Keep Portless startup behind `scripts/start-dev.js`, which now uses a branch-scoped `*.5chan.localhost` route outside the canonical case and automatically increments a `-2`, `-3`, ... suffix when that branch-scoped route is already occupied.
- **Status:** confirmed
### Toolchain model names are not interchangeable
docs/agent-runs/pretext-feed-sizing/progress.md
+6 -6
View File
@@ -62,7 +62,7 @@ Append one entry per session.
- Summary: Tightened the desktop board-feed height model by matching the actual OP body render path more closely. The board estimator now adds the desktop blockquote padding, models the board-view truncated-comment notice, exposes an explicit multiboard board-label height hook for `/all`-style feeds, and applies a small desktop preview-reply calibration instead of treating preview replies like full thread rows. That moved the board benchmark from “measurement-cheaper but visibly off” to near-parity runtime with much lower DOM reads.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/components/post-desktop/post-desktop.tsx`, `src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `src/lib/utils/pretext-height-estimates.ts`
- Verification: `corepack yarn type-check`, `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run`, Playwright desktop board benchmark at `http://127.0.0.1:1356/?e2e=pretext-benchmark&surface=board&variant=production&count=800&seed=42&mode=dom` and `mode=item-size`, fresh-route Playwright smoke on `#/all` and `#/biz`
- Blockers: Fresh browser route smoke still cannot validate deep live `/all` content because RPC/cached-feed availability is inconsistent in this environment, and `./scripts/agent-init.sh --smoke` is not yet worktree-aware because it waits for the canonical `http://5chan.localhost:1355` URL instead of the branch-scoped Portless route.
- Blockers: Fresh browser route smoke still cannot validate deep live `/all` content because RPC/cached-feed availability is inconsistent in this environment, and `./scripts/agent-init.sh --smoke` is not yet worktree-aware because it waits for the canonical `https://5chan.localhost` URL instead of the branch-scoped Portless route.
- Next: Re-test cached live `/all` and `/biz` against the current branch to confirm the board-feed geometry feels better in real browsing, then decide whether the board path is finally strong enough to merge along with catalog/replies.
## 2026-04-02 17:23
@@ -70,7 +70,7 @@ Append one entry per session.
- Item: F002, F004
- Summary: Reprofiled live `/all` with the private Node RPC, found that desktop board cards were still underestimating almost exactly `14px` per preview reply, removed that stale feed-level subtraction, and added a regression test so desktop feed cards now add the preview reply estimates directly. The live result changed materially: a 99-card `/all` sample went from `39.7px` mean absolute error before the fix to `7.1px` after the fix, with the controlled board benchmark regaining a cleaner `item-size` win.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `src/lib/utils/pretext-height-estimates.ts`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `corepack yarn test --run`, live Playwright `/all` sampling on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all`, live same-branch A/B on `#/all` vs `#/all?pretextFeed=off`, controlled Playwright board benchmark on `?e2e=pretext-benchmark&surface=board&variant=production&count=800&seed=42&mode=dom` and `mode=item-size`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `corepack yarn test --run`, live Playwright `/all` sampling on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all`, live same-branch A/B on `#/all` vs `#/all?pretextFeed=off`, controlled Playwright board benchmark on `?e2e=pretext-benchmark&surface=board&variant=production&count=800&seed=42&mode=dom` and `mode=item-size`
- Blockers: Live `/all` now looks much healthier, but the route-level perf delta versus `pretextFeed=off` is still modest/noisy rather than a dramatic blowout, and the master-origin comparison is not trustworthy yet because the master session is not loading the same feed depth as the worktree even after the settings pass.
- Next: Do another user-driven real-feed pass on `/all` and `/biz` with the new desktop board fix in place, then decide whether the remaining work should focus on squeezing a larger live-route win or on preparing the branch for merge with the current measurable improvement.
@@ -79,7 +79,7 @@ Append one entry per session.
- Item: F002, F004
- Summary: Confirmed that the remaining single-board feed glitching was not primarily a bad Pretext formula. Desktop board preview replies were still rendering under `content-visibility:auto` with the `120px` intrinsic placeholder while the feed card itself was relying on Pretext sizing. I disabled deferred layout for board preview replies when the feed estimate path is active, added dev-only preview reply audit attributes on desktop, and reran live `/biz` sampling. That change dropped `/biz` from `52px` mean absolute feed-card error to `6.7px`.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/components/post-desktop/post-desktop.tsx`, `src/components/post-mobile/post-mobile.tsx`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `corepack yarn test --run`, live Playwright `/biz` sampling before and after the deferred-layout fix on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/biz`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `corepack yarn test --run`, live Playwright `/biz` sampling before and after the deferred-layout fix on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/biz`
- Blockers: The branch is now much closer to “merge-ready on real feeds,” but the strongest user-facing proof is still the live geometry/jank sampling plus the deterministic harness rather than a clean master-origin A/B, because the master session is still not consuming the same feed depth reliably.
- Next: Do one more human-eye browsing pass on `/all`, `/biz`, and catalog with the current branch. If the route now feels consistently better than production, prepare the branch for merge with the rollback query flags retained.
@@ -88,7 +88,7 @@ Append one entry per session.
- Item: F002, F004
- Summary: Tracked the mobile `/all` stutter to two separate regressions. First, the Pretext worktree was missing `master` commit `99c0bbfac perf(board): reduce mobile reverse-scroll jank`, so `board.tsx` was still snapshotting Virtuoso state on every `scroll` tick and `comment-media.tsx` was double-running the GIF first-frame hook per card. After porting that, live mobile `/all` audits still showed huge feed-card misses on preview-heavy media posts, so I reprofiled the live cards and found the mobile board preview calibration was stale: cards with five preview replies were underestimating by `+371px` / `+367.9px`. I replaced the old aggressive polynomial calibration with a bounded per-preview-count table, then skipped the full thread-reply estimate hook on board previews and memoized the expensive derived reply maps. That collapsed the live mobile `/all` mean absolute feed-card error from about `135px` to about `18.8px` and pulled the reverse-scroll probe much closer to master.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/components/comment-media/comment-media.tsx`, `src/components/post-desktop/post-desktop.tsx`, `src/components/post-mobile/post-mobile.tsx`, `src/hooks/use-reply-height-estimates.ts`, `src/lib/utils/__tests__/pretext-height-estimates.test.ts`, `src/lib/utils/pretext-height-estimates.ts`, `src/views/board/board.tsx`
- Verification: `corepack yarn type-check`, `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts src/components/__tests__/post-community-address-compat.test.tsx src/views/board/__tests__/board.test.tsx`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), live Playwright mobile `/all` geometry audit on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all`, live mobile reverse-scroll probe on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all`, and matched master control probe on `http://5chan.localhost:1355/#/all`
- Verification: `corepack yarn type-check`, `corepack yarn test --run src/lib/utils/__tests__/pretext-height-estimates.test.ts src/components/__tests__/post-community-address-compat.test.tsx src/views/board/__tests__/board.test.tsx`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), live Playwright mobile `/all` geometry audit on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all`, live mobile reverse-scroll probe on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all`, and matched master control probe on `https://5chan.localhost/#/all`
- Blockers: Mobile `/all` is no longer catastrophically wrong, but the worktree still trails the current master control in the dev reverse-scroll probe. Latest matched sample: worktree `seenCount=23`, `scrollHeight=18209`, `maxFrameMs=133.3`, `slowFrames16=11`, `slowFrames32=9`; master `seenCount=23`, `scrollHeight=17735`, `maxFrameMs=108.3`, `slowFrames16=8`, `slowFrames32=6`.
- Next: Keep tuning the mobile board path until the reverse-scroll probe and human-eye feel are at least on par with master, likely by profiling rerender/commit churn on mobile `/all` now that the geometry bug is fixed.
@@ -97,7 +97,7 @@ Append one entry per session.
- Item: F002, F004
- Summary: Found the remaining mobile `/all` board regression in the board-level Virtuoso prop wiring rather than the Pretext formula. In `off` mode the worktree was still rendering `<Virtuoso itemSize={undefined} />`, which overrides React Virtuoso's internal default DOM measurer. That left mobile multiboard cards stuck on the `defaultItemHeight` fallback (`420px`) instead of reconciling to real DOM sizes, which explained the bad `data-known-size` values, wrong scroll height, and the remaining stutter. I changed board and catalog to only pass `itemSize` when the explicit `item-size` mode is active.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/views/board/board.tsx`, `src/views/catalog/catalog.tsx`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/views/board/__tests__/board.test.tsx src/lib/utils/__tests__/pretext-height-estimates.test.ts`, fresh Playwright mobile `/all` load on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all`, and persistent-session Playwright checks on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run src/views/board/__tests__/board.test.tsx src/lib/utils/__tests__/pretext-height-estimates.test.ts`, fresh Playwright mobile `/all` load on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all`, and persistent-session Playwright checks on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all`
- Blockers: The biggest correctness bug is fixed, but the live mobile `/all` scroll probe in this noisy dev environment is still not a clean proof of parity versus master. The good signal is that board cards now reconcile correctly again: on the persistent RPC-backed session the first mobile `/all` card went from stale `data-known-size=420` before the fix to `data-known-size=413` after the fix, matching the real measured height.
- Next: Re-run human-eye A/B on mobile `/all` and `/biz` against master. If the feel issue is gone, the branch is much closer to merge-ready; if not, the remaining work is runtime churn rather than height reconciliation.
@@ -106,6 +106,6 @@ Append one entry per session.
- Item: F002, F004
- Summary: Re-ran matched live mobile A/B probes against `master` after the Virtuoso prop fix and confirmed the branch is no longer trailing on the board routes that mattered. On the persistent RPC-backed `/all` sessions, the worktree now renders the same visible card heights as `master` while posting slightly better reverse-scroll numbers in the probe (`maxFrameMs 83.3` vs `99.1`, `slowFrames16 6` vs `7`, `slowFrames32 2` vs `3`). On `/biz`, the branch keeps the Pretext path active (`data-pretext-height` nodes present) while matching the same visible card heights and total scroll height as `master`; probe timing is now in the same range instead of the clear regression the user saw earlier. I also added an inline comment in board/catalog so the `itemSize={undefined}` pitfall is documented in code.
- Files: `docs/agent-runs/pretext-feed-sizing/feature-list.json`, `src/views/board/board.tsx`, `src/views/catalog/catalog.tsx`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run`, matched Playwright mobile A/B on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/all` vs `http://5chan.localhost:1355/#/all`, matched Playwright mobile A/B on `http://codex-feature-pretext-feed-sizing.5chan.localhost:1355/#/biz` vs `http://5chan.localhost:1355/#/biz`, and catalog smoke on both origins at `#/all/catalog`
- Verification: `corepack yarn type-check`, `corepack yarn build`, `corepack yarn lint` (existing warnings only), `corepack yarn doctor` (`86/100`, repo baseline), `corepack yarn test --run`, matched Playwright mobile A/B on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/all` vs `https://5chan.localhost/#/all`, matched Playwright mobile A/B on `https://codex-feature-pretext-feed-sizing.5chan.localhost/#/biz` vs `https://5chan.localhost/#/biz`, and catalog smoke on both origins at `#/all/catalog`
- Blockers: None in the implementation itself. Remaining work before merge is review/commit/PR hygiene, plus the follow-up GitHub issue about future custom virtualization once this branch lands.
- Next: Prepare the branch for merge, keep `FeedCacheContainer`, and open the deferred “Pretext unlocks custom virtualization” issue only after the Pretext rollout is merged into `master`.