fix(dht): BEP44 mutable signing — channel feeds now work — + channels app two-column layout, native streaming player, file detail, checker guard

[SethMorrowSoftware]

0. THE BUG: mutable DHT values were signed wrong, so no channel feed ever worked

Reported live: followers saw none of a publisher's files and publishers reached no one - symmetric, total failure.

btx_dht_put_mutable's signing callback set the entry e to the value but signed the raw value bytes. BEP44 signs the bencoded form (a string "hi" bencodes to "2:hi"), and every storing DHT node and every getter's libtorrent verifies the signature against that. Signing raw bytes yields a signature that verifies against nothing, so storing nodes reject the put and getters drop the item before emitting dht_mutable_item_alert - the dhtMutableItem event never fires and no feed is exchanged, in either direction. The put still returned BTX_OK (it only confirms the request wired up; signing runs later on the network thread), which is why nothing looked broken. Fix mirrors libtorrent's own examples/dht_put.cpp: bencode e, then sign. The GET side (entry_to_bytes) already returns string entries verbatim, so the round-trip is clean once signing is fixed.

Regression guard: the signing is factored into one helper used by both the production callback and a new btx::test::dht_mutable_sign_verifies hook, which signs through that helper and checks the result with ed25519_verify against the reconstructed BEP44 canonical message (verify_mutable_item itself is TORRENT_EXTRA_EXPORT, not in the shared lib). torrent_smoke_test asserts it verifies for empty and non-empty salt. Confirmed the guard bites: reintroducing the raw-value sign fails exactly those 3 checks; with the fix, 552 checks / 0 failures under gcc ASan+UBSan. Internal change only - no btx_* ABI symbol, record field, or LCB-visible behavior changed, so no ABI bump.

Note: the fix is in the native shim, so it only takes effect with a rebuilt torrentxt library. The committed src/code/ binaries refresh automatically on merge to main (CI commit-binaries); to test before then, rebuild the shim locally.

---

The rest of this branch (the channels-app work that surfaced the bug while testing it):

1. Self-test harness fix - is a string is not valid LiveCode

is a accepts only number/integer/boolean/point/rect/date/color. Replaced with btClearError + (btLastError() is empty). The rest of that OXT run then came back all-green (71/75 handlers).

2. Channels demo - file detail, streaming, inline progress

• Inline progress bars in the transfers table (chProgressBar).
• Selected-transfer file detail via btFileList: name, size, live per-file progress bar; refreshed each tick (repaint only on change), selection preserved by handle.
• Sequential download via btSetSequentialDownload.

3. Two-column 720p layout + native streaming player

• Two-column layout (1180x648) fits a 720p screen; supersedes the 880x900 single column. tabStops re-fit.
• Native OXT player (chPlayer), hidden by default, appears over the transfer pane only while streaming, dismissed with Close. Stream points it at the largest video/audio file (chMainMediaFile) on the torrent's save path (btTorrentStatus["savePath"] + btFileList) and plays as the in-order download fills. Built inside try with an sHasPlayer gate. New chOpenPlayer/chCloseStream, wired into mouseUp.

4. tExt spells text + a static guard

tExt in chMediaKind is t-e-x-t = text; renamed to tSuffix. check-livecodescript.py gains rule 6: any t/p/s/k-prefixed name whose full lowercased spelling is a reserved token. CLAUDE.md gotcha 2 records it.

Verification

• Shim: 552 checks / 0 failures under gcc ASan+UBSan; static gates (lint + golden + registry/ABI sync) green.
• check-livecodescript: 5 files OK (incl. the new shadow rule).
• The two-column layout and native player still want an OXT pass to confirm the player object on the target build; the DHT fix wants the cross-machine re-test that surfaced it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01371AXB4CUUke7enHHS8okc