[Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db

[bgexpert]

Summary

A project whose primary folder path contains a non-ASCII / Unicode character renders as empty in the Desktop sidebar (no sessions listed), even though the sessions exist in state.db with a cwd that correctly matches the project folder.

The backend (projects.tree / project_tree.build_tree) returns the sessions correctly — the project is only emptied in the Desktop UI layer, not in the data.

Reproduction (Windows 10, Hermes 0.18.2)

1. Create a project whose primary folder path contains a Unicode letter, e.g. a folder literally named bistånd (Swedish å) under D:\Projects\SV\.
2. Open that project in the Desktop sidebar and start one or more chats there (so state.db rows get cwd = D:\Projects\SV�istånd).
3. Send a message (a state.db row is created with that cwd).
4. Open the Projects view.

Expected: the project lists its chats.
Actual: the project shows zero sessions, even though the same data renders fine for an ASCII-only sibling project.

Empirical evidence (from a real state.db + projects.db, no PII / no real paths shown)

Two projects, both with valid, matching cwd:

project folder (normalized, case-folded)	session cwd (normalized)	backend match	Desktop UI
d:/projects/sv/bistånd	d:/projects/sv/bistånd	MATCH (True)	EMPTY
d:/sud/alimonyluxembourg	d:/sud/alimonyluxembourg	MATCH (True)	shows sessions

• The backend project_tree._FolderIndex.match() uses Python casefold() over path segments, so the accented path matches correctly and build_tree returns the 2 sessions for the accented project.
• A sibling project with an ASCII-only path and identical structure shows its sessions normally.
• The only structural difference between the two projects is the Unicode letter å (å) in the folder name.

Root-cause hypothesis

The Desktop/Electron renderer resolves and compares the project's folder path against session cwds using a path comparison that is not Unicode-normalization-safe (likely comparing the raw on-disk path / a differently-encoded variant of the accented name against the projects.db / state.db string). Because the two byte-strings never compare equal at the UI layer, the session-to-project grouping produces an empty lane — while the Python backend (Unicode-aware casefold) groups them correctly.

This is adjacent to #64629 (Windows D:\ vs D:/ normalization duplicate), but distinct: there the project appears twice; here an accented-path project appears empty despite a correct backend match. It is also distinct from #50438 (where cwd is NULL and never recorded) — in this case cwd is present and matches.

Suggested fix direction

Make the Desktop path-key comparison Unicode-normalization-form-stable (e.g. NFC/NFD normalize both sides, or compare via the same casefold/segment logic the backend uses) before grouping sessions into a project lane.

Related: #64629 (Windows path normalization), #50438 (TUI sessions not recording cwd).

[GottZ]

This was generated by AI during triage.

Summary

Fifteen PRs were evaluated in this issue complex, but only #65030 directly addresses #65014: it NFC-normalizes path identity in both the backend project tree and the Desktop overlay so canonically equivalent accented paths retain their sessions. The remaining PRs address adjacent but distinct causes such as non-Git lane identity, Windows separator/case handling, stale repository metadata, lane ordering, or branch switching.

Related pull requests

• feat: add 'No workspace' bucket for cwd-less sessions in project tree #53098 unrelated — (+89/-2) — unrelated: Adds a synthetic “No workspace” bucket for cwd-less sessions; [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014 has valid cwd values whose Unicode spellings fail Desktop identity matching. The keep_open review concerns salvaging that separate bucket while excluding junk cwd rows.
• fix(projects): use _branch_lane_id for non-git folders to prevent duplicate lanes (#53329) #53345 unrelated — (+33/-1) — unrelated: Changes non-Git fallback lane IDs from raw paths to ::branch::main, addressing duplicate lanes rather than canonically equivalent Unicode paths rendering empty. Its keep_open review also requires preserving persisted raw-path identifiers or providing a compatibility migration.
• fix(cli): prevent stale git_repo_root from hijacking session placement in sidebar tree #55834 unrelated — (+45/-5) — unrelated: Validates that a persisted git_repo_root contains the session cwd so stale metadata cannot hijack placement; it does not normalize Unicode path segments. The keep_open review identifies a separate incorrect expectation that rejected rows become unscoped instead of falling back to their cwd workspace.
• fix(desktop): use main lane for non-git project folders #58048 unrelated — (+26/-1) — unrelated: Aligns non-Git fallback lanes with the Desktop ::branch::main identity and label to prevent duplicate folder/main lanes. The high-salvageability keep_open verdict confirms that non-Git mismatch, not [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014’s NFC/NFD ownership mismatch.
• fix(desktop): preserve legacy non-git workspace groups #61950 [closed] partial — (+236/-33) — partial but not a [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014 fix: Adds backend Windows case/separator identity and restores legacy non-Git grouping, but does not NFC-normalize path segments. Although its review initially kept the salvageable backend work open, this closed PR remains relevant because it was explicitly superseded by fix(desktop): preserve legacy non-git workspace groups (supersedes #61950) #63077, which preserved that work and added Desktop parity.
• fix(desktop): preserve legacy non-git workspace groups (supersedes #61950) #63077 [merged] unrelated — (+316/-41) — adjacent merged reference implementation: Establishes mirrored backend/Desktop path-identity helpers for Windows case and separator equivalence, providing the shared comparison boundary that fix(projects): NFC-normalize path identity so accented folders keep their sessions (#65014) #65030 extends with NFC normalization. It does not itself handle canonically equivalent Unicode spellings.
• fix(desktop): order project-tree lanes by recency in the overview, not alphabetically #63547 unrelated — (+45/-2) — unrelated: Delays removal of overview session rows until after lane sorting so recency ordering matches drill-in ordering; it does not affect project ownership or Unicode path comparison. Its keep_open verdict is for that independently verified ordering defect.
• fix(desktop): normalize Windows workspace identity 🩷 #64648 unrelated — (+87/-8) — adjacent Windows fix: Replaces raw main-checkout and worktree comparisons with existing Windows-aware path keys, addressing separator/case duplicates rather than NFC/NFD mismatch. Its keep_open review confirms these additional Windows reconciliation sites and does not cover [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014’s normalization form.
• fix(projects): NFC-normalize path identity so accented folders keep their sessions (#65014) #65030 best fix — (+71/-2) — fixes: NFC-normalizes every path segment at the shared backend _comparison_segments() and Desktop comparisonSegments() boundaries, with backend and live-overlay tests for NFC/NFD equivalence and combined Windows case/normalization differences. This directly matches [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014’s cause, and the maintainer-bot keep_open review rates the focused salvage path highly.
• fix(desktop): allow new chats in non-git projects #68414 unverified — (+654/-31) — unrelated: Introduces explicit Git capability metadata and gates branch switching so new chats work in non-Git projects; it does not normalize accented path identity. Its keep_open verdict confirms the non-Git branch-target boundary and associated capability-propagation coverage.
• fix(projects): keep Windows main checkout in branch lane #71858 unrelated — (+20/-1) — unrelated subset: Uses _path_key() instead of raw equality when classifying Windows main checkouts, preventing separator variants from becoming worktree lanes but leaving Unicode normalization untouched. The keep_open review calls it incomplete for fix(desktop): duplicate branch lanes in project sidebar on Windows — backend/frontend lane-id mismatch #71837 and identifies fix(tui_gateway): normalize lane roots so Windows main checkout isn't duplicated (#71837) #71889 as the consolidation path.
• fix(desktop): reconcile backend main-lane ids with the frontend overlay convention (#71837) #71860 [closed] unrelated — (+92/-3) — unrelated superseded fix: Normalizes Git probe roots and changes non-Git fallback lane identity, addressing Windows and non-Git duplicate lanes rather than [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014. This closed PR remains relevant because its author explicitly closed it as a lossless subset of fix(tui_gateway): normalize lane roots so Windows main checkout isn't duplicated (#71837) #71889.
• fix(tui_gateway): normalize lane roots so Windows main checkout isn't duplicated (#71837) #71889 unrelated — (+138/-4) — unrelated consolidation: Combines Git probe normalization, _path_key() main-checkout comparison, and canonical non-Git lane IDs for fix(desktop): duplicate branch lanes in project sidebar on Windows — backend/frontend lane-id mismatch #71837. Those changes resolve Windows separator and lane-contract mismatches, not canonically equivalent accented path strings.
• fix(desktop): create sessions in non-git projects #75459 unverified — (+467/-22) — unrelated broader non-Git fix: Distinguishes successful non-repository probes from failed probes, folds fabricated non-Git lanes, and skips branch switching only for confirmed plain folders. The keep_open review’s blocking probe-state objections were addressed in the shown follow-up diff, but none of this supplies NFC normalization for [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014.
• fix(tui_gateway): normalize git probe paths so main checkout isn't misclassified as a worktree #77219 unverified — (+13/-1) — unrelated narrow Windows fix: Applies os.path.normpath() only to Git probe roots so slash variants compare equal and main checkouts are not misclassified. It neither touches the backend/Desktop Unicode identity helpers nor fixes the NFC/NFD mismatch reported in [Bug]: Desktop sidebar renders a project EMPTY when its folder path contains a non-ASCII (Unicode) character — sessions exist in state.db #65014.

Duplicates

#53345 and #58048 substantially overlap on canonical non-Git fallback lane identity. #61950 was superseded by the merged #63077. #71858 and #71860 are component fixes consolidated by #71889; #77219 overlaps only #71860/#71889’s Git-probe-normalization component. #68414 and #75459 overlap on allowing session creation in non-Git projects, while #75459 uses probe-derived folder state instead of #68414’s backend gitKind contract. None of these is a duplicate of #65030’s Unicode-normalization fix.

Suggested consolidation

#65030 is the best existing fix for #65014 because its diff updates both authoritative identity boundaries and adds direct NFC/NFD regressions; #77219 is not a substitute because it only normalizes Git probe path separators. Consistent with the maintainer-bot keep_open verdict on #65030, author action: rebase #65030 onto main, or split out the backend/Desktop NFC-normalization changes and their four focused tests as the part that can merge. Do not consolidate the unrelated non-Git, Windows separator, ordering, or branch-switch PRs into #65014; close only their documented duplicate chains under their own issues, notably #53345 versus #58048 and #71858/#71860 in favor of #71889.

Complex graph

flowchart TD
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I53329(["issue #53329 (open)"])
    I57715(["issue #57715 (open)"])
    I59228(["issue #59228 (open)"])
    I61362(["issue #61362 (open)"])
    I62165(["issue #62165 (open)"])
    I64629(["issue #64629 (closed)"])
    I65014(["issue #65014 (open)"])
    I68192(["issue #68192 (open)"])
    I71837(["issue #71837 (open)"])
    P64648["PR #64648 (open)"]
    P65030["PR #65030 (open)"]
    P77219["PR #77219 (open)"]
    subgraph Dup53345 ["PRs duplicating each other"]
        P53345["PR #53345 (open)"]
        P58048["PR #58048 (open)"]
    end
    subgraph Dup61950 ["PRs duplicating each other"]
        P61950["PR #61950 (closed)"]
        P63077["PR #63077 (merged)"]
    end
    subgraph Dup68414 ["PRs duplicating each other"]
        P68414["PR #68414 (open)"]
        P75459["PR #75459 (open)"]
    end
    More["+3 more PRs not shown"]
    P61950 -.->|partial| I65014
    P65030 -->|best fix| I65014
    P68414 -.->|unverified| I65014
    P75459 -.->|unverified| I65014
    P77219 -.->|unverified| I65014
    P53345 -.->|partial| I53329
    P58048 -->|best fix| I53329
    P68414 -->|best fix| I53329
    P75459 -->|best fix| I53329
    P53345 -.->|partial| I57715
    P58048 -.->|partial| I57715
    P68414 -->|best fix| I57715
    P75459 -->|best fix| I57715
    P53345 -.->|partial| I59228
    P58048 -->|best fix| I59228
    P68414 -->|best fix| I59228
    P75459 -->|best fix| I59228
    P53345 -.->|partial| I61362
    P58048 -.->|partial| I61362
    P68414 -->|best fix| I61362
    P75459 -->|fixes| I61362
    P53345 -.->|partial| I62165
    P58048 -->|best fix| I62165
    P68414 -->|best fix| I62165
    P75459 -->|best fix| I62165
    P61950 -.->|partial| I64629
    P63077 -.->|partial| I64629
    P64648 -->|best fix| I64629
    P77219 -->|fixes| I64629
    P68414 -->|best fix| I68192
    P75459 -->|fixes| I68192
    P53345 -.->|partial| I71837
    P58048 -.->|partial| I71837
    P61950 -.->|partial| I71837
    P63077 -.->|partial| I71837
    P64648 -.->|partial| I71837
    P75459 -.->|partial| I71837
    P77219 -.->|partial| I71837
    class I53329 open
    class I57715 open
    class I59228 open
    class I61362 open
    class I62165 open
    class I64629 closed
    class I65014 open
    class I68192 open
    class I71837 open
    class I65014 target
    class P53345 open
    class P58048 open
    class P61950 closed
    class P63077 merged
    class P64648 open
    class P65030 open
    class P68414 open
    class P75459 open
    class P77219 open
    class P58048 best
    class P58048 best
    class P58048 best
    class P64648 best
    class P65030 best
    class P68414 best
    class P68414 best
    class P68414 best
    class P68414 best
    class P68414 best
    class P68414 best
    class P75459 best
    class P75459 best
    class P75459 best
    class P75459 best
    class More closed
    click I53329 "https://github.com/NousResearch/hermes-agent/issues/53329"
    click I57715 "https://github.com/NousResearch/hermes-agent/issues/57715"
    click I59228 "https://github.com/NousResearch/hermes-agent/issues/59228"
    click I61362 "https://github.com/NousResearch/hermes-agent/issues/61362"
    click I62165 "https://github.com/NousResearch/hermes-agent/issues/62165"
    click I64629 "https://github.com/NousResearch/hermes-agent/issues/64629"
    click I65014 "https://github.com/NousResearch/hermes-agent/issues/65014"
    click I68192 "https://github.com/NousResearch/hermes-agent/issues/68192"
    click I71837 "https://github.com/NousResearch/hermes-agent/issues/71837"
    click P53345 "https://github.com/NousResearch/hermes-agent/pull/53345"
    click P58048 "https://github.com/NousResearch/hermes-agent/pull/58048"
    click P61950 "https://github.com/NousResearch/hermes-agent/pull/61950"
    click P63077 "https://github.com/NousResearch/hermes-agent/pull/63077"
    click P64648 "https://github.com/NousResearch/hermes-agent/pull/64648"
    click P65030 "https://github.com/NousResearch/hermes-agent/pull/65030"
    click P68414 "https://github.com/NousResearch/hermes-agent/pull/68414"
    click P75459 "https://github.com/NousResearch/hermes-agent/pull/75459"
    click P77219 "https://github.com/NousResearch/hermes-agent/pull/77219"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 15 pull requests and 9 issues in this complex. Each diff was read against this issue; Assessment working set: 164 kB of PR diffs, 72 kB of issue/PR text, 40 kB of discussion (43 comments), 81 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.