Skip to content

Capture and compare original source Display request geometry - #269

Draft
wieslawsoltes wants to merge 19 commits into
feat/native-source-display-textfrom
feat/source-display-comparison-probe
Draft

wieslawsoltes wants to merge 19 commits into
feat/native-source-display-textfrom
feat/source-display-comparison-probe

Conversation

@wieslawsoltes

@wieslawsoltes wieslawsoltes commented Oct 2, 2026 •

Copy link
Copy Markdown
Owner

Scope

Adds independent original-Windows Typeface request-metric capture before TextFormatter/FormatLine, and a separate native source-double comparison probe over LayoutHintedSourceParagraph. Original 192-case and 288-case receipts, glyph order, assertions and 60-second/90-second/eight-minute bounds stay intact. New schemas 2/4 are separate CreateNew outputs; failure cannot erase prior primary observations. Exact internal diagnostic reflection is signature/assembly/face checked with no fallback.

Depends on #266 for the actual retained source producer and on #261 for the full original midpoint/Hebrew inventory. Both histories are preserved through normal merges. This draft is stacked on #266; dependencies must qualify/merge before retargeting to main.

The new source probe retains original doubles and full case/run/UTF-16/occurrence ownership, compares actual native source-run offsets/advances and line geometry without repair, and executes both explicit TT35/TT40 choices. All numeric policies are caller-selected. Old raw probe remains separate. Source-input allowlist now admits only the four exact schemas2/4 captured at92f4778b0 by successful original Windows workflow37011550318. Independent font/em/DPI/method/assembly identities were reviewed, all original case payloads exactly match prior immutable receipts, and cross-architecture input metrics agree. Old192/v3 schemas still cannot substitute missing metrics. No ordinary Display registration or parity claim.

Bounded postcommit checks

  • New device-free parser/comparison controls: initial31/31 against each original x64/ARM64 192/288 receipt; after input qualification37/37 against each paired old/new schema1/2 and3/4 receipt (148 executions, zero skipped).
  • Existing raw-probe64/64 and original midpoint17/17 controls pass unchanged.
  • Original Windows reference project source compile: zero warnings/errors against cached references; no Windows execution.
  • Six executable C# files parsed; YAML and all4 PowerShell blocks parsed without execution; git diff --check passed.
  • Native-linked probe type build/execution awaits qualified coherent producer packages. No local native/renderer/full-source builds, GPU/VM execution or runtime staging.

Hosted original-Windows workflow37011550318 succeeded on both architectures with every original primary and independent input capture. New head35d0a5cce has reference workflow37012562360 queued; the capture implementation is unchanged by its parser/hash-only update. Complete exact successful native Build qualification is still required before any native comparison execution.

@wieslawsoltes

Copy link
Copy Markdown
Owner Author

Hosted probe type compilation passed at bc048af829aa8fa248008c9fdb254da4b26a3021 in Linux job110865946418, step11: actual Backend.Native/Text outputs, Release/linux-x64/CI, BuildProjectReferences=false, zero warnings/errors, 1.46seconds. No native probe execution or producer qualification is claimed.

The same job later failed only the existing Boolean vector-mask allocation fixture at sample grid4: 3,168bytes versus zero; grids1/8 passed. Suite result: 5,553passed, one failed, nine existing skips. The allocation origin is not established.

Head b6ef52b84007b43ae75aac87bbe5cddb02ff6487 now carries only the already-reviewed test isolation and receipt via cherry-pick -x from fb7e1fd2e/055d1b0c9; no #264 shader feature ancestry. The helper, original payload, one warmup, 10,000 calls, exact-zero threshold, positive controls and validation match the reviewed files byte-for-byte; the new compile gate is unchanged. No redundant local graph/native validation or runtime staging occurred.

Replacement whole Build37019061789 is confirmed at the new exact head. The superseded failed run is not qualified. The independent source-policy comparison still awaits whole-successful producer #266 and will initially run only its explicit actual source-policy triple against all four reviewed input receipts, preserving both TT35/40 and every rejected/unmatched observation.

@wieslawsoltes
wieslawsoltes added this pull request to stack #314 October 2, 2026 21:48

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant