Text column edges

Photo details Step 4 implementation summary

Feature plan: Step 1 · Step 2 · Step 3 · Step 4

Debug mode: enabled (inherited from Step 3)

Effort study: railspike-effort / v3/image_metadata; host ledger. Standing P2 enrollment; extended estimates enabled. Stage estimates below are copied from the approved Step 3 plan; Claude single exposed rater throughout.

Cycle estimate: 0.50

Branch: feature/image-metadata, from origin/main at 14beeaaa. Planning commit d8e0060.

Study snapshots

  • Step 4 entry, 2026-11-15 UTC: 0.50, Step 3 reaffirmed unchanged. Written after the planning commit and before Stage 1 code.

Stage 1 - Read photo details in the browser

Original estimate: 0.15

Retrospective stage work: 0.15

  • Changes:

    • public/assets/image_prepare.js: a bounded EXIF reader finds the EXIF block in JPEG (APP1), PNG (eXIf) and WebP (EXIF) sources. It reads IFD0, the Exif IFD and the GPS IFD in either byte order.

    • ForumImagePrepare.prepare now returns details: {key, label, value, display, line} for Camera, Lens, Exposure, Taken and Location. line is the signed text, such as Photo-Taken: 2026-11-14T16:12:08-05:00.

    • Guards: 512 entries per IFD, 4096 values per tag, a visited-offset set against loops, bounds checks on every read, control characters folded to spaces, 100 characters per text value, and a catch that returns no details.

    • The prepared file still carries no metadata.

  • Verification:

    • Passed: node tests/browser/image_prepare.mjs (Chromium). The new case covers big-endian JPEG, little-endian PNG, no EXIF, truncated, looping, oversized IFD count, and a newline inside a value.

    • Passed: ./v3 test smoke set, 464/464. The baseline before any change was also 464/464.

    • Passed: python3 -I tests/frm_client_test.py; it only mentions image_prepare.js in a docstring.

    • Not run: WebKit (IMAGE_BROWSER=webkit); the iPhone replay in Stage 4 covers Safari.

  • Notes:

    • Reuse: the new reader sits beside the existing dimensions marker walk in the same file. No third-party library.

  • Estimate details: original scope as Step 3; uncertainty: low; revisions: none.

— Claude · 2026-11-15 UTC · Model: claude-opus-5-5

Stage 2 - Choose details at attach time and sign them

Original estimate: 0.20

Retrospective stage work: 0.22 (one rendering fix and a secure-context test harness)

  • Changes:

    • templates/partials/image_compose_fields.php: a details.compose-disclosure with data-compose-disclosure="photo" under the preview, hidden until a prepared photo has details. Every composer gets it through this partial.

    • public/assets/image_upload.js: builds one unticked checkbox per detail, with the value in plain words and a warning on Location. Collapses again for each new photo. New ForumImageUpload.chosenLines(form).

    • public/assets/compose_disclosure.js: the photo kind summarises ticked labels, like Options.

    • public/assets/browser_signing.js: imageSignedFields(fields, image, description, photoLines) checks each line against the five known keys and adds them after Image-v1:. Thread, board and reply submits pass the chosen lines.

    • public/assets/reply_continuation.js: the reply item stores photoLines in its saved record, and every prepare, including Retry, sends them.

    • public/assets/compose.css: row styles for the checklist.

  • Verification:

    • Passed: new node tests/browser/image_photo_details.mjs (Chromium). It covers collapsed and unticked by default, the Location warning, the summary, chosen lines, a photo without metadata, and a reply whose first prepare fails then retries with the same lines.

    • Passed: node tests/browser/image_compose_transport.cjs, extended. Nothing ticked still gives today's body; two ticked lines follow Image-v1:; an injected newline or unknown key is refused.

    • Passed: node tests/browser/compose_disclosure.mjs and node tests/browser/image_prepare.mjs.

    • Passed: ./v3 test smoke set, 464/464.

    • Checked by screenshot: Gloss, Light, Dark and Word97 at 390 and 1280 pixels, collapsed and open. Values first rendered bold through theme label styles; fixed in compose.css.

    • Not run: WebKit and a live server upload; Stage 4 covers the iPhone replay.

  • Notes:

    • Deviation from the Step 3 wording: the reply item stores the final lines, not the detail keys. The stored attachment is the prepared file, which has no metadata, so keys could not be resolved again on retry. The lines are already validated text.

    • Reuse: the existing compose disclosure, its summary hook and the image partial. No new composer or disclosure style.

  • Estimate details: original scope as Step 3; uncertainty was the reply retry path, now covered by a test; revisions: none.

— Claude · 2026-11-15 UTC · Model: claude-opus-5-5

Stage 3 - Show details under the image

Original estimate: 0.15

Retrospective stage work: 0.17 (two shared consumers added: thread titles and board excerpts)

  • Changes:

    • src/ForumRewrite/Images/ImageReference.php: PHOTO_DETAIL_LABELS, photoDetails(string $after) and withoutReference(string $body). Only Photo- lines directly after the Image-v1: line, with a known label and at most 120 characters, count. Taken shows as "Nov 14, 2026, 4:12 PM" (the camera's own clock time) and Location as "42.3536° N, 71.1167° W".

    • src/ForumRewrite/View/MediaEmbedRenderer.php: the image card ends with a post-photo-details grid of label and value; those lines leave the body text. Plain and Markdown posts behave the same.

    • public/assets/offline_reader.js: the offline image placeholder is followed by "Label: value" lines, formatted like the server.

    • src/ForumRewrite/Support/ThreadTitle.php and templates/partials/thread_row.php: titles and board excerpts skip photo lines, as they already skip the image line.

    • public/assets/site.css: the post-photo-details grid.

  • Verification:

    • Passed: new MediaEmbedRendererTest::testPhotoDetailsAfterTheImageLineRenderUnderTheImage. It covers plain and Markdown, escaping, formatting, no details, a Photo- line away from the image, an unknown key, and thread titles.

    • Passed: OfflineSnapshotPresentationTest::testThreadDetailRendersPostBodyLineBreaksLikeOnlineView, extended with a photo-details post.

    • Passed: ./v3 test --smoke MediaEmbedRendererTest ThreadTitleWithdrawalAndConsumersTest ThreadRootCardRepeatedTitleTest ImageContentTest, 494/494; ./v3 test --smoke OfflineSnapshotPresentationTest, 472/472.

    • Checked by screenshot: the rendered post card in Gloss at 390 pixels.

    • Inapplicable: RSS and ApiTextController show the body as plain text, so the lines read as Photo-Camera: …; no code change was needed.

  • Notes:

    • Reuse: the existing image card, ImageReference parsing and the offline nl2br mirror.

  • Estimate details: original scope as Step 3, plus the two title and excerpt consumers found during inspection; uncertainty: low; revisions: none.

— Claude · 2026-11-15 UTC · Model: claude-opus-5-5

Stage 4 - Document and verify the whole flow

Original estimate: 0.08

Retrospective stage work: 0.08

  • Changes:

    • New concept entry docs/concepts/photo_details.md: purpose, scenario, line format, sources and boundary.

    • docs/concepts/shared_composer.md and docs/concepts/post_body_formatting.md: one dated line each, linking to it.

    • docs/plans/effort_estimation_trial.md: ledger index row and dated note for v3/image_metadata.

  • Verification:

    • Passed: ./v3 test --smoke MediaEmbedRendererTest OfflineSnapshotPresentationTest ThreadTitleWithdrawalAndConsumersTest ThreadRootCardRepeatedTitleTest ImageContentTest ImageUploadValidatorTest, 504/504 (smoke set plus affected classes).

    • Passed: node tests/browser/image_prepare.mjs, image_photo_details.mjs, compose_disclosure.mjs and image_compose_transport.cjs (Chromium).

    • Passed: git diff --check against origin/main.

    • Not run: the full suite (./v3 fulltest); the maintainer runs it.

    • Pending: iPhone Safari replay on zenmemes.com with a real camera photo. This is the Step 2 known-environment check. Chromium results do not satisfy it.

    • Final diff review: 27 files, all within the approved stages. No dependency, schema, read-model or feature-flag change. No navigation destination added, so nav icons are inapplicable.

    • Feedback items: none in docs/feedback/items/ mention photo metadata, EXIF or location, so no Feedback: trailer applies.

  • Notes: concept entries match the delivered behavior; their source links resolve.

  • Estimate details: original scope as Step 3; uncertainty: the iPhone replay; revisions: none.

— Claude · 2026-11-15 UTC · Model: claude-opus-5-5

Cycle closeout

  • Technical status: Stages 1–4 complete and committed on feature/image-metadata. Not merged or deployed.

  • Delivered scope: Option B as approved, plus the collapsed-disclosure revision. Two shared consumers, thread titles and board excerpts, were added in Stage 3 because they would otherwise show raw Photo- lines.

  • Corrective work during the cycle: the checklist rebuilt too early, before preparation finished (Stage 2), and values rendered bold under theme label styles (Stage 2). Both were fixed before their stage commits.

  • Retrospective whole-cycle scope assessment: 0.55. This is not a measured actual; no labor measurement was recorded.

  • Pending: iPhone Safari acceptance and the maintainer's comfortable-fit judgment. Study acceptance stays pending.

  • Follow-ups not in this cycle: frm.py photo details, keeping metadata inside the image file (Option A), and editing details after posting.