_MISSING_DATA_DIRS is module state: a test that called main() after one
whose data directories were not cached exited 1 with the earlier run's
notice. main() now clears it first.
release_pack: true in the registry is the list a release builds: RetroArch,
EmuDeck, RetroBat, Batocera, RomM and MiSTer, 83% of the downloads of
v2026.10.10. generate_pack --all, --verify-packs, the pipeline consistency
step, release_record and check_release read it. Manifests are still written
for every platform.
Each manifest carries system_titles, the name the site prints for every system its files list (Sony PlayStation, not sony-playstation or psx), so the installer can show it. Older installers ignore the field.
batocera-systems lists bios/saves/dolphin-emu/Triforce/segaboot.gcm, so the pack folder for the save directory is userdata-saves/. validate_schemas refuses a pack folder the BIOS tree already uses.
Batocera, KNULLI, Recalbox, RetroBat and RetroDECK hand a libretro core <saves>/<system> at each launch. Each copy lands below every system that runs the core. EmuDeck keeps one unsorted saves/retroarch/saves. The unsupported platforms record why.
load_from: save_dir entries of a libretro build were left out of every
pack. They now go below the platform's save folder (registry
save_directory: RetroArch saves/, Lakka savefiles/, by core), the
manifest marks them root: save, and verify replays the same placement.
Platforms keeping saves per system or per game list them as omitted.
The install manifest marks each game data file "category": "game_data":
a platform entry only game data profile entries read, or a core file its
profile declares game data. Firmware carries no category. Every manifest
says "categorized": true so the installer can refuse an older list.
generate_pack.py --no-game-data cuts those files from the platform list
and the core profiles before the build, names each one with the cores that
read it as game data, and tags the pack _NoGameData through _narrowings.
verify.py and the pack check take the same cut, so --verify-packs and the
report describe the pack the builder wrote. Modes without a platform
(--emulator, --system alone, --from-md5) and --all-variants refuse it.
The pack check failed a pack for a declared file nothing in the collection resolves to, though the builder cannot ship it and verify already reports it missing. Such destinations are now counted as not in the collection; a held file absent from the pack stays an error.
minarch hands every core SDCARD_PATH/Bios/<TAG> as its system directory
and opens no file in it. The scrapers read each pak's launch.sh (core,
tag, devices), each device's core makefile (the project a library is
built from: MyMinUI's dosbox_libretro.so is dosbox-pure) and the minarch
statement, at a pinned commit. The installer recognises the card from
.system/version.txt and refuses a folder inside it.
The builder keeps a held regional fallback when the requested file is
missing (since 2026-10-17), and the comment beside the manifest's
omissions still said the opposite. A file the builder holds but the
manifest cannot download sat in omitted_files counted unheld, so the
installer kept a fallback the pack drops. Such an omission now carries
held: true and narrow() counts it held, as the builder does.
After --all-variants, --verify-packs printed "hashes checked during the
build" for the platform-only, truth and required-only packs and compared
none of them, so one missing a file its narrowed list requires passed.
Each pack is now held to the platform list narrowed the way it was built:
source and required_only, read from its name, reach the baseline, the
region grouping and the replay of the core extras.
generate_pack and verify read --system beside --platform through common.resolve_system_arg, built on platform_system_names, the mapping the manifest's system_names comes from. verify --platform --system now reports on the narrowed pack. Unknown names are refused with suggestions before the database is read.
A core extra resolved by name shipped on existence platforms even when
the core's own size check turns the copy away. The RetroPie pack carried
the 128 KB Apple IIGS ROM as Basilisk II's ROM (min 512 KB). The builder,
the manifest and verify now read validation.core_size_refusal, which
accepts a copy any version of the same profile slot takes.
The pack README's one-line install was written per platform: ROCKNIX and MiSTer had none, EmuDeck's Windows edition was told to open Konsole, and RomM users were sent to library/bios/ whatever their firmware template. The automatic section now follows each platform's install.detect systems, the EmuDeck and RomM manual steps name every layout, and the download page lists EmuDeck under Windows and macOS.
The platform's systems are known once its config is loaded. The filter ran inside the build, after the output lock and a refresh of every data directory, and ended on a WARNING with exit 1. It is now a usage error before any work.
The check ended on the last platform's line: a failing pack was named only above its own errors. It now names every failing pack on stderr at the end, as the build path does, or counts the packs found OK.
CommentsSayWhy read three scripts while 22 comments in 13 other files
named the next statement. They are removed or replaced by the reason,
and the test reads every tracked script and install.py.
The manifest now carries system_names, each system's native_system names from the platform's own file, and --system, its suggestions and --list-systems read them: --platform batocera --system psx installs sony-playstation.
download_external stayed in generate_pack, outside the offline switch: it fetched under --offline, wrote to the system /tmp and held the body in memory. It now lives in packresolve, honours the switch, streams into the repository tmp/ and hashes on the way.
BizHawk is both a platform and a profile display name: --from-md5 with --platform bizhawk and with --emulator bizhawk wrote one file in two layouts. The emulator mode now carries _Emulator in the name.
verify resolved platform entries with common.resolve_local_file and put
the slot override first, and the region pass added a destination hint.
packresolve.resolve_declaration now serves the builder, the manifest,
verify and the region pass: release assets cached for a file absent from
bios/ count as shipped in all four, and a release asset keeps its answer
over a slot override.
The refusal tables compared each flag with its default value, so --source full and --group-by system passed beside the modes that refuse every other value. Both now default to None and take their effective value after the checks, in generate_pack and pipeline.
mkstemp creates its scratch 0600 and os.replace kept that mode, so
database.json, the manifests, the packs, the recipe snapshots and the
large-file cache were readable by their owner only. set_plain_mode
gives the scratch 0o666 less the umask, as open() would, before each
rename. A test walks every function that renames a mkstemp file.
Started outside the repository root, both ended on a FileNotFoundError
traceback: the database and the data registry name files from the root.
They now move to the root after reading the paths the caller typed, which
stay relative to where they were typed.
_enrich_hashes repeated "ground truth" five times with capitalised
insistence for a three-step lookup, and the MAME ZIP rebuild listed
three guarantees of which two were the same.
The lock refusal, "no pack for", "no manifest for" and "no pack to
verify" went to stdout, so a run redirected to a log exited 1 with
nothing on the terminal. A test reads both scripts and refuses an ERROR
line printed to stdout.
generate_pack refreshed the data directories before probing its output
lock. The refresh waits for every run holding a tree, and a pipeline holds
them all: a second build sat silent for hours instead of being refused.
The output lock is now taken first, and a refresh that has to wait for a
tree names it.
A value of another length was looked up as an MD5 and reported NOT FOUND,
a SHA-256 was dropped, and an input with no hash ended on "No valid
hashes found in input". Digests are now read by length (CRC32, MD5 and
its 29-digit Batocera form, SHA-1, SHA-256), SHA-256 is looked up in the
database, and any other value is refused with the formats accepted.
--platform retroarch --system sony-playstaton listed the first ten systems
of the alphabet. The warning now names the closest systems and the
command that lists the platform's own, as --system alone already did.
The region grouping and phase 2 of each pack and manifest called
find_undeclared_files with the same arguments: 30 reports for the 15
manifests of --all --manifest, 2.8 s, now 15 and 1.7 s. The 15 manifests
are byte-identical before and after.
Step 5 compared (ok, total) and the core file count. RetroArch, Lakka
and RetroPie read 526/526 on both sides while verify credited four paths
below SGB1.sfc and SGB2.sfc that the pack never writes: a declared file
blocks them as a directory. verify now places the platform files in the
builder's order with the same path rules, and a user_provided or
external declaration keeps its slot on both sides, so no core file takes
it in the report only.
verify --destinations writes the paths each pack holds by the report,
and step 5 compares them with the archive members (notes and data tree
members set aside). Every emulator profile, in each build it has, is
built in memory (PackRecord, nothing read or written) and compared with
verify --emulator path by path. Cost measured here: about 1 s per
platform archive and 11 s for the 930 emulator packs.
Every member was deflated, archives and PNGs included. On the 2026-10-20
RetroArch pack the .zip, .7z, .chd, .png, .ogg and .pk4 members are 2.1 GB
that took 28 s to deflate again for 9.6 MB saved (0.45 %), and about 16 GB
for 72 MB across all packs. Those formats are now stored. The choice reads
the member name only, so a pack stays byte-reproducible.
_stable_hashes retried through the hash cache, so both retries returned
the first reading and the file was read once before the manifest run
stopped. The retries now read the file again.
Both are written as <platform>.json. --manifest-targets aimed at install/
replaced install/retroarch.json with a target table, and --manifest aimed
at install/targets/ did the reverse, both with exit 0. Each mode now stops
when the output directory holds the other kind and names the directory to
use.
gngeopi and neo-emu declare neogeo.zip loose and, apart, the ROMs they
open inside it. The archive pass took the copy holding them, the loose
entry resolved by name to another copy, and both verify --emulator and
the emulator pack reported the archive missing while the ZIP held it.
with_archive_members() gives the loose entry the members in verify, the
pack, slots and the gap report.
The Required packs and manifests carried every data tree, the FreeIntv
overlays included. A data_directories entry now accepts required: false,
and the platform, emulator and manifest walks leave such a tree out under
--required-only. FreeIntv reads its overlays only with the multi-screen
option and draws a placeholder keypad without one (src/libretro.c:294-296,
350-369).
--verify-packs after an --all-variants build rebuilt only the full pack's
name and opened it alone. The five narrowed packs of the same run passed
without a line. Each one is now named with the hash check the build ran
on its members.
--emulator accepts a buildbot core name from a profile's cores:, and a
mistyped target gets the closest names instead of every target. diff_truth
shows destinations and reports unpaired entries as evidence, not removals.
The verify summary counts discrepancies, a shared destination names both
systems, a truth hash conflict names both cores, and pipeline steps carry
their place in the real run instead of N/8. Every GitHub API caller takes
the token from github_token(), which falls back to gh auth token.
BlastEm opens 32X_M_BIOS.bin and PicoDrive 32X_M_BIOS.BIN. On a case-insensitive platform the manifest kept the first and dropped the second core's claim, so an install narrowed to PicoDrive (RetroArch on Switch) lost the 32X BIOS its pack carries. The claim now joins the entry kept.
Abdessamad Derraz