[sanyamsingh001]

TTML and DFXP exports could emit duplicate XML IDs or replace valid Unicode cue IDs. A cue named s1 could collide with the document’s style ID, and a generated c1 could take an ID belonging to a later cue.

The writer now reserves valid explicit cue IDs before generating replacements for missing, invalid or repeated IDs. Style, region, cue and speaker IDs remain unique across the document. Valid Unicode IDs retain their exact spelling, including combining characters, under these collision rules. Speaker IDs are allocated after cue IDs, preserving a valid cue named agent1.

This can change exported IDs for existing files: IDs accepted by the old check but invalid under XML NCName rules (for example x², containing U+00B2) now receive deterministic replacements. Later duplicate IDs are also replaced; the first usable occurrence is kept. These identifier changes do not alter cue text or timing.

The README documents the behavior and the Namespaces in XML 1.0 Third Edition and XML 1.0 Fifth Edition character rules.

Validation on Linux:

• Focused regression: 2 passed / 10 failed before the fix; all 12 TTML/DFXP cases passed afterward. Coverage includes reserved/generated/duplicate IDs, Unicode preservation, invalid name characters, deterministic output and caption round trips.
• All six required gates passed: formatting, strict workspace Clippy, workspace tests, dependency layers, asset checks and WASM. Workspace tests: 3,500 passed, 0 failed, 44 ignored; WASM checked 49 crates.
• External-reader interoperability and other operating systems remain unverified. Test totals do not imply every optional external oracle ran.

Depends on #1749, including its unchanged #1725 and #1639 dependencies. This follow-up changes three files: the caption writer, its cue-ID regression tests and the caption README. Against upstream main at cee1feb185e675a108f78067f71ad86a640ace3d, the full comparison contains four commits and 14 files, including those dependencies. Original dependency authorship is preserved.

[FredApps]

This is stacked on #1749 (with its #1725 and #1639 dependencies), so I reviewed only the last commit, 353fa96e, which is ttml.rs, its tests and the README. The first three commits are the same ones carried by #1905, #1857 and #1911, and they need their review on #1749, which I'd land first. The change is correct; I'm commenting for three reasons that are not about the logic.

What I checked in 353fa96e:

• The collision bug is real on main. cue_id returns c{index+1} for a missing or invalid ID without looking at later cues, and nothing keeps an explicit s1 or r1 from colliding with the style and region IDs (ttml.rs:130/154/182). A cue 1 with no ID and a cue 2 named c1 get the same xml:id.
• cue_ids fixes it in the right order. It reserves every valid explicit ID first (plus s1 and r1), keeps the first use of a valid ID, and gives every other cue c{n} with _ appended until it is in neither the reserved nor the used set. The loop ends, since each pass adds a character to a name that is checked against a finite set. Speaker IDs are still allocated after, from ids, which now starts with the final cue IDs.
• The name check follows the specification. The start ranges and the extra name characters (-, ., digits, U+00B7, U+0300–036F, U+203F–2040) match XML 1.0 Fifth Edition [4]/[4a] without :, and the old check (is_alphabetic/is_alphanumeric) was wrong in both directions (it allowed ² and other characters that are not name characters, and it refused valid names such as ones starting with a letter outside the old test's idea of alphabetic). Valid Unicode IDs are kept byte for byte, with no normalisation.

Points:

1. Behaviour change for existing files. A cue ID that the old check accepted but the XML rule does not (for example one containing ², or Ⅷ) is now replaced by c{n}, and a duplicate explicit ID now keeps only its first use, where before both were written. Both are the point of the fix and produce documents that other readers can load, so I'm not objecting, but the text should say that exported IDs can change for such files.
2. Merge state. The PR is DIRTY: merging it with main conflicts in crates/captions/README.md. It will need a rebase once #1749 settles.
3. Same file as other open PRs. #1723, #2049 and #2394 also change ttml.rs. I did not check them against this one line by line, but the first to land will mean a rebase for the rest.

Hosted CI does not build or test filmcraft, so I'm leaving this as a comment. The text reports six passing gates on the author's Linux machine, and says external readers were not tested. I'll approve once #1749 is settled, the branch is rebased and the receipt for this exact head is in.

Validation request (head 353fa96eeb4c57de63602258a6d4ed221276cee5):

• os: linux
• commands: cargo fmt --all -- --check; cargo clippy --workspace --all-targets --locked --offline -- -D warnings; cargo test -p filmcraft-captions --locked --offline; cargo test -p filmcraft-captions --test ttml_cue_ids --locked --offline
• expected: all pass, 0 ignored; ttml_cue_ids should report 12 passing cases
• holds: stacked on #1749 (needs its own review); merge conflict in crates/captions/README.md; clippy nonminimal_bool on main (#2007) is not caused by this PR