id(web): nested marks corrupt on save — em/strike/code inside strong/em gain   entities and multiplied delimiters, several non-fixpoint

Summary

In the id web editor's markdown ↔ ProseMirror codec (pkgs/id/src/web/markdown.rs), any nested mark combination corrupts on save: when the active mark set changes mid-run (emphasis inside strong, strong inside emphasis, strikethrough or inline code nested inside strong/em, emphasis inside a link), the emitted markdown gains   HTML entities in place of spaces and multiplied emphasis delimiters. Several cases are not a fixpoint — every subsequent save makes the output worse.

Single-mark and sibling cases are fine (plain **strong**, *em*, ~~strike~~, `code`, and ***both*** all round-trip cleanly), which is why the existing tests miss this: the corruption only appears when adjacent text nodes carry different overlapping mark sets.

Evidence (verified this run on main `aa0d6d1`)

Standalone probe crate compiling the verbatim codec from pkgs/id/src/web/markdown.rs (current main), markdown → ProseMirror → markdown, twice:

Input	After 1st save	After 2nd save
`bold _em_ end`	`bold *em* end`	`bold ***em*** end` (keeps growing)
`em strong* end*`	`em *strong** end*`	stable in corrupted form
`bold ~~strike~~ end`	`bold ~~strike~~ end`	`bold ~~strike~~ end` (changes again)
`bold \`code` end`	`bold \`code` end`	—
`em \`code` end`	`em \`code` end`	—
`_em_ rest`	`*em* rest`	—
`[link em text](https://e.com)`	`[link ](https://e.com)[em](https://e.com)[ text](https://e.com)`	`[em]` migrates inside the link on the next save (not a fixpoint)
`*a b \`c` d* e**`	`a *b \`c` d* e**`	—

The parsed ProseMirror is correct in every case (e.g. **bold _em_ end** parses to text nodes bold [strong], em [strong, em], end [strong]) — the corruption is introduced on the emit side.

Probe logs: goal workspace hidden_files/probe-2026-10-26-0815/out.log and out2.log.

Suspected area

prosemirror_to_markdown's inline emission: create_marked_text (markdown.rs:1117) and build_mark_chain (markdown.rs:1167). When consecutive text nodes have overlapping-but-different mark sets, the chain appears to close/reopen marks in an order that leaves boundary text nodes whose leading/trailing spaces comrak then renders as  , and re-parsing the result produces yet more mark boundaries — hence the per-save growth.

Related but distinct: #944 (strong+link mark-order oscillation — no nesting required), #953 (code+link drops the link via the code early-return at markdown.rs:1122). This issue reproduces with no link and no code mark at all (pure em+strong), so it is not a duplicate of either; a fix should be checked against both.

Why it matters

Nested emphasis is ordinary writing (**bold with *italic* inside**). Today, opening such a file in the web editor and saving it — even without edits — rewrites it into entity-escaped, delimiter-doubled text, and repeated saves compound the damage. This is silent data corruption of user documents.

Acceptance criteria

**bold _em_ end**, *em **strong** end*, **bold ~~strike~~ end**, **bold \code` end**, em rest, and link em text` each round-trip open → save byte-identical (modulo the codec's documented normalizations), and a second round-trip is a fixpoint.
No   (or other HTML entities for ordinary spaces) appear in emitted markdown for these inputs.
New unit tests in pkgs/id/src/web/markdown.rs cover nested-mark round-trips, including the fixpoint (double round-trip) property.
Existing codec tests still pass, and the id(web): bold links flip spelling on every save — strong+link marks have no round-trip fixpoint #944 / id(web): inline code combined with another mark drops the other mark on save — code+link loses the link #953 cases do not regress.

CODE-165

🤖 Agent note · filed

Agent: new-issue-creator job
Session / ID: new-issue-creator run 2026-10-26 08:15 CDT
Found via: filed this run
Task: Filed nested-mark save corruption in the id web markdown codec (verified probe on main aa0d6d1, verbatim codec in a standalone crate: **bold _em_ end** -> **bold *****em***** end**, growing again on the second save; 8 nested combinations reproduced); queued as a triage packet for the queue worker.
Expectation: effort/m — done when nested-mark round-trips are byte-stable fixpoints with no   entities, with double round-trip unit tests.
Links: packet enqueued to repo-health-code-pilot this run

🤖 Agent note · update

Agent: new-issue-creator job
Session / ID: new-issue-creator run 2026-10-26 08:15 CDT
Found via: same probe batch, after filing
Task: Two more reproduced cases in the same nested-mark family, recorded here instead of as separate issues: (1) x *em *nested* end* y -> x *em **_nested_** end* y (em inside em, same symptom); (2) [~~old~~ new](https://e.com) -> ~~[old](https://e.com)~~[ new](https://e.com) (strikethrough nested in a link leaks outside the link text on save). Both are mark-set-change boundaries in the same emit path; a fix for this issue should be regression-checked against them. Probe log: goal hidden_files probe-2026-10-26-0815/out3.log.

🤖 Agent note · parked

Agent: queue worker (repo-worker-code)
Session / ID: queue task 1116, cron shift 2026-10-26 13:24 CDT
Found via: queue claim
Task: Triage parked with two material findings. (1) The probe evidence in this issue predates PR fix(id-web): preserve other marks on inline code on save (#953) #999 (merged today, task 1102), which changed exactly the suspected emit area: create_marked_text no longer early-returns on code, and a new append_inline_child merges adjacent text nodes sharing the same outer mark wrapper. Several id(web): nested marks corrupt on save — em/strike/code inside strong/em gain   entities and multiplied delimiters, several non-fixpoint #968 cases share an outer mark across all nodes, so current main MUST be re-probed before any fix is written — the fix may be partly or wholly landed already. (2) Local verification route: a fresh-worktree cargo test --features web --lib first fails on the pre-existing RustEmbed prerequisite (pkgs/id/web/dist must exist), and after creating dist the id lib-test compile was SIGKILL'd (OOM) in this sandbox. The creator's standalone-crate probe (verbatim markdown.rs + comrak only) is the proven lightweight route.
Expectation: resume — re-probe the 8 table inputs on current main via the standalone-crate route; if clean, close as no-longer-reproducible with the probe log and add fixpoint tests; if still corrupt, implement the emit fix. Probe test code (probe_968_nested_marks, 8 inputs) is uncommitted in worktree ~/git/wt-1116 on branch queue/1116-968-nested-marks-triage, target dir warm.
Links: PR fix(id-web): preserve other marks on inline code on save (#953) #999 · issue id(web): inline code combined with another mark drops the other mark on save — code+link loses the link #953

🤖 Agent note · parked

Agent: queue worker (repo-worker-code), lane B shift
Session / ID: queue task 1173, run 2026-10-26T23:02Z
Task: id(web): nested marks corrupt on save — em/strike/code inside strong/em gain   entities and multiplied delimiters, several non-fixpoint #968 nested-mark emit fix implemented: append_inline_child peels pure mark chains and tail-inserts the leaf, reusing the tail wrapper for the shared mark (subsumes the id(web): inline code combined with another mark drops the other mark on save — code+link loses the link #953 merge; all three append sites). PR fix(id/web): re-nest nested-mark runs on save instead of fragmenting spans (#968) #1026 @ b37a2314.
Verification so far: standalone codec crate (module verbatim): 67/67 web::markdown tests pass, including the 5 previously-corrupt cases byte-identical + fixpoint and the clean-set regression guards; rustfmt clean.
Expectation: parked 20 min for PR CI. On resume: if CI green, merge fix(id/web): re-nest nested-mark runs on save instead of fragmenting spans (#968) #1026, verify id(web): nested marks corrupt on save — em/strike/code inside strong/em gain   entities and multiplied delimiters, several non-fixpoint #968 auto-closes, complete; if red, fix on queue/1173-968-nested-mark-emit.
Links: PR fix(id/web): re-nest nested-mark runs on save instead of fragmenting spans (#968) #1026

Resumed by queue worker (Lane A), task 1173: PR #1026 CI was fully green on b37a231, but the PR had become unmergeable — main moved under it and pkgs/id/src/web/markdown.rs conflicted. Merged current main (5709db9) into the branch and resolved both hunks: main's newer #1031 footnote-definition block handling and #1032 bullet-list coalescing were kept, adapted to this branch's 3-arg append_inline_child (the #968 mark-chain re-nesting signature). Verified the merged file in a standalone verbatim-codec probe (comrak 0.56): 75/75 unit tests pass (nested-mark matrix, footnote, coalesce, table and task-list suites), rustfmt clean. Merge pushed as cdf990f; parked for the CI rerun.

{"v":1,"kind":"note","state":"parked","task_id":1173,"summary":"PR 1026 was green but conflicted with main in markdown.rs. Merged main 5709db9f, resolved 2 hunks keeping main #1031/#1032 logic on the branch 3-arg append_inline_child; probe suite 75/75 pass, rustfmt clean. Merge commit cdf990f1 pushed; parked for CI rerun.","artifacts":["https://github.com/developing-today/code/pull/1026"],"next":"merge PR 1026 when CI green on cdf990f1, verify #968 auto-close, complete task 1173"}

Merge pull request #1026 from developing-today/queue/1173-968-nested-…

Done — queue worker (Lane A), task 1173: PR #1026 merged (merge commit 20e1a23), issue auto-closed. CI on the main-merged head cdf990f was fully green — all 27 checks pass, including test-unit, all shards, clippy, fmt and the web suites — on top of the local verbatim-codec probe at 75/75 from the conflict-resolution pass.

{"v":1,"kind":"note","state":"done","task_id":1173,"summary":"PR 1026 merged 20e1a23c; #968 auto-closed. CI fully green on cdf990f1 (27/27); probe 75/75 at merge-resolution time.","artifacts":["https://github.com/developing-today/code/pull/1026"],"next":"none — task 1173 complete"}

fix(id): hoist delimiter-mark edge spaces touched by word chars on sa…

Input	After 1st save	After 2nd save
`bold _em_ end`	`bold *em* end`	`bold ***em*** end` (keeps growing)
`em strong* end*`	`em *strong** end*`	stable in corrupted form
`bold ~~strike~~ end`	`bold ~~strike~~ end`	`bold ~~strike~~ end` (changes again)
`bold \`code` end`	`bold \`code` end`	—
`em \`code` end`	`em \`code` end`	—
`_em_ rest`	`*em* rest`	—
`[link em text](https://e.com)`	`[link ](https://e.com)[em](https://e.com)[ text](https://e.com)`	`[em]` migrates inside the link on the next save (not a fixpoint)
`*a b \`c` d* e**`	`a *b \`c` d* e**`	—

Navigation Menu

Repository navigation

id(web): nested marks corrupt on save — em/strike/code inside strong/em gain   entities and multiplied delimiters, several non-fixpoint #968

Summary

Evidence (verified this run on main `aa0d6d1`)

Suspected area

Why it matters

Acceptance criteria

29 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

id(web): nested marks corrupt on save — em/strike/code inside strong/em gain &#32; entities and multiplied delimiters, several non-fixpoint #968

Description

Summary

Evidence (verified this run on main aa0d6d1)

Suspected area

Why it matters

Acceptance criteria

Activity

developing-today commented on Oct 26, 2026

dezren39 commented on Oct 26, 2026

dezren39 commented on Oct 26, 2026

dezren39 commented on Oct 26, 2026

29 remaining items

dezren39 commented on Oct 26, 2026

dezren39 commented on Oct 27, 2026

dezren39 commented on Oct 27, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

id(web): nested marks corrupt on save — em/strike/code inside strong/em gain entities and multiplied delimiters, several non-fixpoint #968

Evidence (verified this run on main `aa0d6d1`)