Repository navigation
Conversation
…an invalid signature On macOS 27 (arm64, Bun 1.3.13) the compiled binary keeps the linker's ad-hoc signature after Bun appends the payload. codesign --verify reports "code or signature have been modified" and the kernel kills the binary on launch with status 137. compile.sh now verifies the signature on darwin and re-signs ad hoc only when verification fails.
A Nano 7 synced by macOS carries iTunes Library.itlp and iTunesCDB but no iTunesDB; libgpod parses the CDB, yet IpodDatabase.hasDatabase only probed iTunesDB. device scan reported "iTunesDB not found" and sync refused with IPOD_NEEDS_INIT, pointing at device init for a device with 21 tracks. hasDatabase now accepts either file.
…line runRepair ran its own refusal check with the injected assessIpodIdentity but called runRepairPipeline without deps, so the pipeline's second check assessed the real device path. The supported-device test therefore failed on any host without an iPod at /Volumes/iPod. The assessor is now forwarded through the pipeline's existing deps; both checks stay, and the test counts that both used the injected assessor.
…abases The nano 6th and 7th generation sign iTunesCDB and Locations.itdb.cbk with HashAB. libgpod 0.8.3 only computes it through an external blob loaded from LIBGPOD_BLOB_DIR, which podkit does not ship. dstaley/hashab is an Unlicense C implementation of the same calcHashAB function. The C sources, headers, licence and the 100 upstream reference vectors are copied byte for byte from the pinned commit; UPSTREAM.md records the source. Nothing builds them yet.
…B signer libgpod 0.8.3 loads calcHashAB from an external hashab module in LIBGPOD_BLOB_DIR and fails nano 6G/7G writes without it. A small patch replaces that loader with a direct call into the vendored sources, which prepare-libgpod-nano.sh copies into the libgpod tree before autoreconf. They are built as an internal convenience library with -fwrapv, and their global symbols are renamed and hidden, so that they neither clash with other code linked with libgpod nor become part of its interface. The macOS developer build and the static prebuild dependencies both apply the patch. The signer only accepts an eight-byte FireWire GUID, so the patch checks for 16 hex digits first and fails the write otherwise. The static dependency cache now records a fingerprint of the patch and the vendored sources, and the CI cache keys hash all of tools/prebuild, so either change rebuilds libgpod. macOS static links name the system sqlite3 and libxml2 that libgpod already uses, and Linux static links group the archives, so standalone executables link too. tools/nano7 holds the checks: a read-only probe that verifies the signatures of a copied Locations.itdb/.cbk pair and iTunesCDB, including the hashing scheme libgpod writes, tests against the upstream vectors and a reference checksum book, and a libgpod roundtrip for nano 6G/7G that the prebuild workflow runs. ADR-030 records the decision. Both generations stay read-only.
The support notes, unsupported-device reasons and docs still said libgpod cannot produce the nano 6G database format or needs an external hashab blob. With the signer linked in, the remaining blocker is different: on a nano 7G the firmware shows no music for a library libgpod writes, even though its signatures verify. The nano 6G write path is covered only by synthetic tests. Both generations stay read-only. The notes and reasons now say that syncing is not supported yet and, for the nano 7G, what the firmware shows; the generated support matrix, the supported-devices page and the device notes follow, the generation notes in docs/formats/generations.md and docs/test-devices.md point to ADR-030, and the classify/identify tests track the new reason. The generation comments drop the test devices' serial numbers.
…evel A nano 7G (firmware 1.0.4) shows no music for a library libgpod writes, although its signatures verify. Replacing the databases on the device showed that the firmware needs the hidden category containers together with iTunes' schema level; the containers alone are not enough. For devices that sign with HashAB and are not iOS devices, the patch now - adds the columns iTunes' Library.itdb has beyond libgpod's schema, appended after the tables are created; the inserts that relied on column positions name their columns, - writes version_info 1.117 as iTunes does and sets the Genius.itdb user_version to 2, - runs the device's post-process command set 25, which fits that schema, adds is_home_video and sets user_version 26, instead of the newest set. Two commands of the set add a column that is already there: item_stats.rental_expired, which libgpod's schema has, is skipped once the column is confirmed, and the repeated AddGenreMapCompilationCount only after its first run succeeded. On these devices a missing command set, a missing, malformed or failing command, an insert that fails while the databases are generated or a failed commit now fails itdb_write() before any library file is replaced; failed post-process commands are rolled back. Other devices keep logging such failures. Attached database paths are quoted. The tools/nano7 roundtrip checks the schema level for nano 6G/7G, that each kind of command failure, a rejected insert and a commit that fails as on a full disk leave the library unchanged, that the first of the repeated genre commands runs, and that a nano 5G keeps libgpod's layout. Both generations stay read-only.
The nano 6G/7G firmware browses six hidden smart containers: Music, Videos, Movies, TV Shows, Audiobooks and Rentals. iTunes keeps them as dataset 5 playlists in the iTunesCDB and as container rows in Library.itdb. libgpod parsed and rewrote the dataset 5 playlists but wrote no SQLite rows for them. On devices that sign with HashAB and are not iOS devices: - Dataset 5 headers keep +0x52, the category kind that SQLite stores as distinguished_kind, and +0x54 as read. libgpod copied +0x50 into +0x52, which turns the Videos category (+0x50 0, +0x52 1) into an ordinary playlist for iTunes. - Library.itdb gets one hidden, live smart container per category, with iTunes' smart settings and the rules as smart_criteria, serialised by the existing mhod 51 writer. The firmware evaluates the rules itself, so no members are written; Music gets the master playlist's media kinds, as in iTunes. Dynamic.itdb gets their container_ui rows. - A library that lacks a category, such as a new one, gets it with iTunes' rules. - The Rentals rule has an action libgpod cannot classify; it is written as read instead of passing through itdb_splr_validate(), which only warned about it. The roundtrip checks the rows and rules in both files, that they stay the same over resaves, that missing categories are added while the others keep their ids, Music's media kinds with audio and a movie, that the three header fields round-trip independently, and that a nano 5G still writes dataset 5 as before.
…writes The nano 6G/7G firmware records plays, skips, ratings, bookmarks and whether a track was started in Dynamic.itdb, and reports plays and skips in Play Counts as well (the count minus play_count_user_original). libgpod rebuilt Dynamic.itdb from the iTunesDB and Play Counts only, so a write dropped what the device had recorded there. SQLite dates were also off by the host's offset, and an unset one was written as -978307200 minus the offset. On devices that sign with HashAB and are not iOS devices: - itdb_parse() reads the device's Dynamic.itdb by item pid after the Play Counts merge. Counts take the larger of the two totals, as the parser has added the Play Counts delta already; ratings, bookmarks and recent counts take the device's value, and play and skip dates the later of the two reports. A track the device marked as played, which it does when a track is started or skipped, stays marked. The exact UTC dates of items and playlists come from Library.itdb and Locations.itdb when they can be read. Edits made after parsing are written as usual, and the post-process commands reset the *_original counts to the totals. - A write keeps the audio and subtitle track choices the device recorded for items still in the library; the post-process commands would reset them. It fails, rather than replace the file, when Dynamic.itdb could not be read at parse time. - SQLite dates are UTC seconds since 2001 and an unset one is 0; the value libgpod wrote for an unset date before reads as unset. Locations.itdb records the date added, as iTunes does, and an unset one stays 0. - Play Counts dates use the database's fixed offset and the file's "no date" value sets none; a reported skip date is taken over like the play date. The roundtrip covers plays counted once over repeated writes, Play Counts dates with and without a Dynamic.itdb row and when the two reports differ, activity recorded only in Dynamic.itdb including the four track choice columns and the played mark, host edits that change or clear it, kept, removed and new items, a Dynamic.itdb without track choice columns, dates in four host time zones, and an unreadable Dynamic.itdb. Both generations stay read-only.
With the firmware categories, iTunes' schema level and the device's play data kept, podkit's libgpod writes the iPod nano 7G library in the form its firmware needs: on firmware 1.0.4, the categories together with that schema level made a library show and play, also after a reboot. A nano 7G on that firmware showed and played libraries written by this code, kept the plays, skips and ratings it had recorded, and took added and removed tracks and playlists, so nano_7g becomes syncable with hardware provenance. Its USB product IDs leave the unsupported table, so detection, readiness and sync treat it like the other syncable nanos. Its videos use the ipod-classic profile, as for other generations without a profile of their own. The nano 6G stays read-only: podkit writes its library the same way, but only synthetic tests cover that. Its note and refusal say so now instead of claiming that writes are unsupported. Tests that used the nano 7G as their example of a read-only or refused device use the nano 6G. The generated support matrix, the supported-devices page, the device notes and tools/nano7 follow, and the changeset covers the libgpod-node changes of the previous commits.
Pred8or
force-pushed
the
pr/4-nano7-preserve
branch
from
October 28, 2026 13:33
5dc2b79 to
9c7d699
Compare
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part 4 of 8 for iPod nano 7th generation support (overview: #75). Stacked on #78: until that merges, this PR also shows its commits; only the commits named below are new here.
Branch
pr/4-nano7-preserve(2 commits: 27f33d0, 9c7d699). Review after PR 3.Dynamic.itdb. libgpod rebuilt that file from the iTunesDB and Play Counts only, so every write threw that away. Nowitdb_parse()readsDynamic.itdbby track id (plays counted once, later dates win), takes exact UTC dates fromLibrary.itdb/Locations.itdb, keeps the device's audio/subtitle track choices and its "played" mark, and writes SQLite dates as UTC with 0 for unset. A write fails rather than replaceDynamic.itdbwhen it could not be read.nano_7gbecomes syncable, withverified: 'hardware'. The nano 6G stays read-only (synthetic tests only). Nano 7G videos use the ipod-classic profile until PR 5.Hardware test on my iPod nano 7G (firmware 1.0.4), starting from the library macOS had put on it, with macOS's device agents paused for the whole session so nothing else touched the iPod:
iTunesCDBandDynamic.itdbagreed after every write;That run used an earlier build of this branch. It showed that a write reset the "played" mark, which the iPod sets as soon as a track is started or skipped. I fixed that and repeated the check with the final code: the marks of four tracks the iPod had just started or skipped stayed set over two writes, nothing else changed, and the iPod played everything after a reboot.
Not covered by hardware: adding missing categories to an empty library, and bookmarks (synthetic tests only).
Notes, all unchanged libgpod behaviour: normal smart playlists lose their rules in
Library.itdb(the category playlists keep theirs); libgpod numbers the location kinds its own way (playback is not affected); when the iPod changes a rating, libgpod keeps the earlier computer rating inuser_rating_common(the iPod doesn't show it).Testing: round trip: plays counted once over repeated writes, Play Counts dates with and without a
Dynamic.itdbrow, activity recorded only inDynamic.itdb(track choices and the played mark), edits and clears from the computer, kept/removed/new tracks, four host time zones, an unreadableDynamic.itdb. Integration tests of libgpod-node and core against the patched libgpod.