[coneilen]

Summary

New native Windows workspaces follow macOS naming, preview, and validation rules, including Unicode transliteration and 32-character truncation. Existing directories retain their names, paths, and identities; explicit Rename keeps its existing behavior.

Changes

• Use built-in Windows ICU through a System32-only loader for Latin transliteration, diacritic removal, lowercasing, and grapheme-aware ASCII mapping; fail explicitly rather than falling back to approximate ASCII normalization.
• Share creation-only naming across preview, submit-time collision checks, and directory creation. Preserve file/directory collision detection after transliteration or truncation, Windows path safety, legacy rename rules, and raw directory discovery.
• Reuse the existing Foundation-whitespace helper for preview only; validate every nonempty raw draft. Share owned naming messages between inline validation and creation-race errors.
• Cover ICU error-handle cleanup, allocation failures, malformed UTF-8, Windows path restrictions, collisions, Unicode/length/whitespace behavior, and existing identity preservation. Document finite evidence limits.

Test plan

RED: & $env:GRAPHCODE_ZIG0152 test src\WorkspaceLifecycle.zig --test-filter 'workspace creation naming parity' -> 0/3 passed; expected cafe-zurich, found caf-z-rich; expected NameTaken, created a different directory; legacy-preservation scenario returned NameTaken
GREEN: & $env:GRAPHCODE_ZIG0152 test src\WorkspaceLifecycle.zig --test-filter 'workspace creation naming parity' -> 6/6 passed, including 27 Foundation naming fixtures
REGRESSION: & $env:GRAPHCODE_ZIG0152 test src\WorkspaceLifecycle.zig -> 22/22 passed on 040ba32

Direct Zig commands ran from graphcode-windows with repository-bootstrapped Zig 0.15.2 and owned temporary/cache roots. The original slug RED/GREEN logs are preserved separately from the later full regression.

& $env:GRAPHCODE_ZIG0152 test src\NewWorkspaceForm.zig --test-filter 'new workspace' -target x86_64-windows-msvc -lc -luser32 -lgdi32 -ladvapi32 "-I$env:GRAPHCODE_WINGHOSTTY_ROOT\include"

Result: 21/21 passed. With the same flags and --test-filter 'new workspace naming whitespace parity', the additional RED was 0/3: a space-only draft incorrectly stayed quiet, NBSP was not trimmed from preview, and collision wording omitted the normalized slug. After the fix, that focused command passed 5/5, including no-directory effects, owned preview/message lifetimes, and creation-race reporting.

Actual shared Workspace.slug, macOS preview, and shared naming-error source were compiled with pinned Swift/Foundation 6.3.3: 27 slug fixtures, 12 preview/validation examples, and an exhaustive 19-scalar Foundation .whitespaces set. These are finite oracle checks, not all-Unicode or cross-version equivalence.

The sole desktop executor ran the following admitted current-source gates using owned roots and the already-built canonical guard:

Exact command	Result
Invoke-NamingAcceptance040ba325-CurrentWindow.ps1 -Mode Run -LaneStateSha256 3A65B799BDCC68F0549179AD9810D2E7FA6E86E73D35AE9143CDCEFB05E5B3B8	Exit 0; nine facts; two UI create/open operations with selected new workspaces; unchanged raw paths/names/native file IDs/semantic UI IDs for three legacy directories; all six cleanup fields true
Invoke-AcceptanceFullNaming040ba325.ps1 -Mode Run -LaneStateSha256 168F51E4822925612EECD664735B3EA9FBB2C374C143A4D34C3154D42232D51D	Exit 0; complete unfiltered 684/684 units; 2,238 direct executions plus one derived proof; exact 77-case/seven-group ownership aggregate; fresh 14-artifact/seven-control fixture inventory; all six cleanup fields true
pwsh -NoProfile -File Tools\windows\Test-ShellSectionManifests.ps1 -Directory D:\namingunits040\sections -ShardCount 3	Exit 0; all 66 sections executed exactly once across original 23/18/25 shard manifests

The first two script paths are under C:\Users\coneilen\.copilot\session-state\481117da-1883-41d1-b2d3-bb19ae883cf4\files.

Canonical shell commands executed under those owned supervisors:

& .\Tools\windows\validate.ps1 -Task windows-shell -ShellPart unit -ShellTestShard 0 -ShellTestShardCount 3 -ShellTestManifest D:\namingunits040\sections\shard-0.json -ShellValidationRoot D:\nu0400
& .\Tools\windows\validate.ps1 -Task windows-shell -ShellPart unit -ShellTestShard 1 -ShellTestShardCount 3 -ShellTestManifest D:\namingunits040\sections\shard-1.json -ShellValidationRoot D:\nu0401
& .\Tools\windows\validate.ps1 -Task windows-shell -ShellPart unit -ShellTestShard 2 -ShellTestShardCount 3 -ShellTestManifest D:\namingunits040\sections\shard-2.json -ShellValidationRoot D:\nu0402

All three passed their positive section manifests and cleanup checks. Shard 1 includes lifecycle 22/22 and form 21/21. Shard 0 includes App 1,715 passed/19 skipped and native accessibility 280 passed; skips are not counted as passes. Aggregate output SHA: CCC33285A2F5796D00EEDC9EC851D36CA08F8B10B2DF2BE5AD27D5119AB4F719.

Default integration: the earlier admitted attempt failed waiting for a later Edit edge modal, despite an earlier editor opening successfully. Its 6F83DCA6211EB9A7858E9CCE4B25410CBFEE3A5395F1890C221744332E4BA0BF terminal remains a failure. WM_COMMAND fallback transport success is not semantic editor-dispatch proof; no naming coupling, physical-user activity, RDP cause, or product root cause is inferred.

A separately user-approved unchanged default attempt ran:

& 'C:\Users\coneilen\.copilot\session-state\481117da-1883-41d1-b2d3-bb19ae883cf4\files\Invoke-AcceptanceDefaultIntegrationNaming040ba325-NewGrant.ps1' -Mode Run -LaneStateSha256 8D539D411F160FB55DD8140EC3A1D6161E0E610EECA94193D35D726D4D230D45

Its canonical workload was Tools\windows\validate.ps1 -Task windows-shell -SkipTrayLive -ShellPart integration -ShellValidationRoot D:\n -SwiftExecutable with pinned Swift 6.3.3. Workload exit 0 in 434.5490696 seconds, all six cleanup fields true, isolation/provider/smoke/scrubbed-startup/real-daemon/tray-contract markers present, and complete default UIA JSON with two fixtures, seven action patterns, seven surface patterns, edge workflow, node creation, and multiproject rename.

The original wrapper remains passed=false solely because it stopped for a new untracked runtime batch, not a workload failure. That exact new 14-file/2,573-byte batch was separately user-approved for nonoverwriting verified preservation, with individual journals. No previous batch permission was reused. The coordinator accepted saved evidence, without rerunning the workload or rewriting its original wrapper receipt:

• Raw default UIA: 9DE30D315302CC8C8EC0F79BB29E6ED51DDACE9FD5E437D5511746D3A8B5B462
• Original policy-stop terminal: 9AE83B3056FC7D8688355A6DB983622AB9D66AD062F35DD56182F4FBD87CD054
• Separate preservation completion: 88A31DA49EEDCF83D63B3F1936DADB84737D6EB9D4297AA986550BD629662C42
• Separate saved acceptance: 827351E0BED00C3D81F3184C87D5E5C31CA2DF7A85EABFF785A83A13B6A773FB

These receipts are under D:\namingintegration040b. Saved acceptance verifies deadlines, all markers/cleanup, 876 historical archived files, 448 protected outputs, 23 source controls, preservation, and clean source. Producer custody and raw compiler-file EOF remain unproved. Earlier harness, activation, partial, and integration failures—including the old unknown outer cleanup—remain separate preserved failures; none was relabeled. Initial toolchain/JIT/build-argument errors and private mock-preparation errors are also retained, not used as behavioral RED or native qualification.

Current-head applicability: runtime evidence belongs to 040ba3252b754311de48376aa6772e8b5fec10d5. Current head c30b789c2cb70a970960550b6b9e01b5f4f1af7e merges main 2073e81af4f89f6eb11f27366f732b391808b3a8; the entire tree differs from the tested head only in docs/ramps.json. All 480 compiled-input occurrences (229/126/125), actual guard MVIDs, seven guard-file bindings, 36 qualified product/runtime bindings, and accepted provider/tool/source inputs were rehashed unchanged. This establishes unchanged-input applicability, not new-head native execution or a new guard build. Required PR CI must run on this actual head.

At current head:

& .\Tools\windows\validate.ps1 -Task @('privacy','tdd-evidence')
pwsh -NoProfile -File Tools\tdd\Test-TddEvidence.ps1 -BodyPath C:\Users\coneilen\.copilot\session-state\c7ce035c-0558-4176-8a38-bab05a8cfaeb\files\naming-published-pr-body.md

Results: canonical privacy passed with 121 investigation inputs; TDD validator scenarios passed 4/4; full-template body validator passed. Four changed files match committed UTF-8/LF, git diff --check passes, and all three commits have DCO/coauthor trailers.

Limits: no all-Unicode/version guarantee, original-window focus-restoration claim, paired visual/physical-pointer qualification, injected native creation-race proof, or complete parity-ledger promotion. Shared Swift and macOS code are unchanged. make test/make check were not run on this Windows host. Packaging/scheduler lifecycle work remains CI-only under the existing local exception because it can mutate user settings; it is not claimed as a local pass. No checks, merge protections, or policies are bypassed; no release or auto-merge is requested.

Checklist

• I have read the Contributing Guidelines
• I have signed off my commits (git commit -s) per the DCO
• Tests pass locally (make test)
• Code follows the existing style (make check)
• I added the test/contract before the implementation and observed the intended RED failure