Every profile is loaded whatever the command, so one misindented key in
an unrelated profile stopped verify.py --emulator handy and generate_pack
on a 37-line yaml traceback. The sources common.py reads (profiles,
platform files, registry, shared groups, targets) now raise
SourceUnreadable, which run_cli prints on one line with the file, the
line and the validate_schemas command. It is not a FileNotFoundError:
ten callers skip a platform whose file is absent, and a damaged one must
not be skipped the same way.
--db databse.json, with database.json in the same folder, was told to run
generate_db --force --output databse.json: a rescan of the whole
collection into a second file under the typo. load_database now names the
closest .json in that folder first, and gives the rebuild command only
when none is close.
Only generate_pack.py and pipeline.py checked the interpreter. verify.py,
slots.py, validate_pr.py, cross_reference.py and the probe tools started
under 3.8 and stopped halfway on a str.removeprefix AttributeError. The
check now lives in common.py, which every build script imports, and in
core_probe.py and region_check.py, which do not. generate_site.py imports
common before siterender, which raised a TypeError at load under 3.8.
RetroPie was the only platform with status: archived, which kept it out
of verify --all, slots, the pipeline and list_platforms unless
--include-archived was typed. release_pack already decides which packs a
release publishes and the installer serves every platform, so the status
set nothing apart. The field, its schema entry and the flag are gone, and
RetroPie is covered by every full run like any other platform.
SelectionIndex.build and system_group called _norm_system_id 1.7 million times per manifest run for about a thousand ids, each through the manufacturer prefix loop. The function reads only its argument and two constant tables, so its answer is cached. Measured on the call volume of one manifest run (630 ids, 1,686,618 calls): 1.53 s uncached, 0.07 s cached. The end-to-end manifest run moves inside its own noise on this host (45.9 to 50.6 s across four runs).
artifact_lock created its directory even for a shared reader, so a
mistyped path given to --verify-packs or release_record was left as a
directory holding a .lock, and split_pack created the parent before
printing a bare errno. A reader now raises ArtifactDirectoryMissing,
which run_cli prints on one line, and split_pack checks its target
before taking the lock.
A platform entry names no profile, so bios/Other/<emulator>/ was never foreign to it. EmuELEC's ports/VVVVVV/data.zip resolved to xrick's data.zip, then to xrick's data directory copy. A system that lists its cores now takes no name match only another emulator's tree answers.
Demul 0.7 reads 44 NAOMI BIOS rows from naomi.zip, the HOTD2, F355 and Airline Pilots BIOS among them. No copy held them all. The set is built with build_torrentzip from members already in the collection, each checked against the crc32 and sha1 of demul.exe's table, and kept in bios/Other/demul/. Batocera, KNULLI and RetroBat place it at bios/naomi.zip, the file MAME cores there also open, so it carries the nine MAME members the Demul table does not list.
The resolver left another emulator's own copy out of the member and superset ranking, so Flycast and the MAME cores keep the copies they were served.
segadimm.zip holds fpr23905.ic36 and fpr23905c.ic36 under the same crc32, and segasp.zip four fpr-*.ic72 images. Ranked by crc32 alone, a copy missing some of them tied with the full set and the index order chose it. A crc32 that two declared members share now also requires the member name.
AmberELEC's nec-pc-88 declares DISK.ROM with an empty md5. QUASI88 reads
that name as n88sub.rom, the 2 KB PC-88 disk ROM (libretro.c:1085), but
the name step served the 16 KB MSX disk ROM, the only primary of that
name. The quasi88 alias never reached by_name, which refuses an alias a
primary already carries.
load_platform_config now writes each entry's system as platform_system.
A platform entry with no hash and no owning emulator, whose system id
names a maker, takes no name candidate filed under another maker, in
the name step and in the hash_mismatch step after it. When nothing is
left, by_shadowed_alias, the profile aliases by_name refuses, answers
within the system's maker folder. names_manufacturer and BUCKET_FOLDERS
move to common.py, and generate_site imports them.
Across every registered platform the only resolution that moves is
AmberELEC nec-pc-88 DISK.ROM, from bios/Microsoft/MSX/DISK.ROM to
bios/NEC/PC-88/N88SUB.ROM. Manifests, truth and exports are otherwise
byte-identical.
Without --platform, cross_reference and the site's gap report merged every
platform's standalone set, MinUI, NextUI and MyMinUI coredir: markers
included. A marker says which cores one platform hands a folder, so in the
union every core those three do not list got no layout: px68k showed 0 of
6 files and the global report lost 238 profiles and 1201 files.
merged_standalone_set drops the markers, and both readers use it.
--emulator mednafen_psx refused with 'use --emulator beetle_psx' although the profile names its target. The alias now selects that profile and says so on stderr, in verify, generate_pack and cross_reference alike. Launchers are still refused, and pipeline step 5 skips aliases since they compare under the profile they name.
AmberELEC and EmuELEC declare blueMSX's Machines/<model>/config.ini and
Shared Roms/RS232.ROM with no hash. The name step served msx-emu's
ColecoVision config.ini in 192 machine folders and the 16 KB RS232.ROM,
while data/bluemsx held blueMSX's own file at that very path. A platform
declaration with no hash now reads its destination's tail in the data
directories before the name step. Profile entries keep the foreign-tree
redirect, so Dolphin's dsp_coef.bin dump still outranks the free copy.
Manifests: amberelec and emuelec omit 192 config.ini entries (held in
data/bluemsx only, so the installer has no URL) and serve colecodb.xml,
MSX2+/config.ini, MSXturboR/config.ini and RS232.ROM from blueMSX's bytes.
verify --all --target swich ended on 'no platform has it' after seventeen INFO lines, two of which held the suggestion. The refusal now names the closest targets across every platform with the platforms that have them, and the per-platform lines print only when some platform keeps the target.
Messages in check_freshness, generate_pack, common, profile_sync and fileless_audit named commands as export_native.py --platform x, which fails from the repository root. They now read python scripts/<name>.py like the rest of the report.
install.py took --list-platforms, generate_pack.py --list and verify.py
nothing. Both tools now take --list-platforms and print the same list
from common.py. generate_pack.py keeps --list (finding 80).
A system the target leaves no core for was verified 0/0 with exit 0,
and generate_pack blamed the platform for lacking it. Both read
platform_scope_systems and system_scope_error now (findings 13, 14).
A one-member archive whose member carries the zipped_file md5 under
another name has that md5 as its composite. The name step returned it as
md5_composite_exact, which slots and the site take as content proof, while
Batocera looks the member up by name and finds nothing.
reminiscence's Flashback module reunion resolved to skyline's
zoneinfo/Indian/Reunion through the case-insensitive name step, and the
emulator pack shipped the time-zone file. Another casing found only in
another emulator's tree says nothing about the entry.
The rom folder survey joined 463 folders to their default core's whole
profile, no system of it matching. SYSTEM_ALIASES, the one table
_norm_system_id reads, now folds ten more spellings of one machine:
apple2gs, astrocade, astrocde, creativision, gamepocketcomputer,
nintendo-super-game-boy, cps1, cps2, cps3 and pgm. The profiles that
used them take the canonical id (mame, groovymame, mamemess, mame0258,
mame0278, bsnes, mesen-s, mesence, jollycv, fbalpha2012 and its three
CPS cores, fbneo_cps12), as test_system_ids requires.
With bsnes and bsnes-hd now filing SGB1.sfc under one system, their
claims fold onto one report entry. _names_content read only the first
claim, hashless, and nSide's SGB1.sfc/ tree took the path before the
file. A hash on any folded claim now counts: every pack keeps the
destinations it had.
The site drops the eight display names the canonical ids replace and
names tvgames. README counts 533 systems instead of 541.
install/<platform>.json wrote the constant "1.0" as the version of the
list it was built from. The manifest now carries the version the platform
YAML records, and generate_pack refuses a YAML without one.
inherits: no longer passes a parent's version to a child: Lakka was
published as RetroArch v1.22.2 in its manifest, pack name and site page.
lakka.yml records v6.1, the Lakka-LibreELEC tag its target list is read
at. The MiSTer scraper refuses a BiosDB without a timestamp instead of
writing an empty version.
A pack's manifest.json records list_version, so an install from the pack
alone knows it too. Manifests regenerated twice and compared: only
version and generated changed.
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.
The name step tried the destination's basename alongside the entry's
own names. With dsi_nand.bin absent from a checkout, nand.bin from
dsi/nand.bin let the Wii NAND outrank the entry's own DSi variant, and
the site named melonDS Nightly a reader of it. The basename is now tried
only when no own name answers.
FileConsumers also resolved profile entries without the archive members
the emulator pack passes, so gngeopi and neo-emu were credited with the
MAME neogeo.zip instead of the archive their pack carries.
bios/Bungie holds the Marathon scenarios Aleph One reads, entries its profile files as category game_data. Listed with the other publishers' trees, they leave the system files count and sit under Game engines and PC games on the site.
A platform can hand each libretro core a system directory of its own
(core_system_dirs). platform_layouts then gives one view per directory,
so the gap report, the builder, the manifest, slots and cross_reference
place a core's files below each folder it is handed, and a core handed
none gets no file. Resolution strips the folder again and searches the
collection by the core's own path.
Each OS publishes Batocera's systems dict (knulli-systems, and
batocera-systems at AmberELEC's newest prerelease commit and EmuELEC v4.8),
read with the shared parser. Cores come from each OS's own ES list.
AmberELEC and EmuELEC declare neogeo/ and ports/ beside bios/ from
/storage/roms, so their packs extract there and core_destination puts
core files under bios/.
standalone_path is relative to the directories a standalone build
resolves it against, now declared per profile (standalone_directories).
A launcher record that links or passes a whole folder of the BIOS tree
under one of them gives that folder, and platform_layouts rebases the
standalone view once for the builder, the manifest, verify and slots
(launchers.standalone_root, common.standalone_destination).
Flycast on RetroPie and ROCKNIX reads bios/dc, on EmuDeck
bios/flycast/bios, on Batocera (FLYCAST_BIOS_PATH) and RetroBat
(Dreamcast.BiosPath) bios/dc. The ROCKNIX, Batocera and RetroBat
scrapers now record the patched data directory, the environment
variable and the setting that say so.
The -listxml names, for every machine, its parent (romof), the devices of its default configuration (device_ref) and the cards its slots offer (slotoption). A machine takes the system of the first set on its romof chain a written entry lists, and passes it to each device it can instantiate. A device set reached from one system gets system:, from several the new per-entry systems:, from none neither, and keeps serving every system of the profile. The gap report, the core extras and slots read the entry's own systems, so generate_pack --system and the installer pick by them.
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.
_holding_members kept every same-named copy when the best held nothing,
so galgames.zip, a lone assembler listing, resolved by name for mame,
groovymame and mame4droid-2024, and awbios.zip for mame4droid. Such a
copy now resolves as hash_mismatch.
mame2003_plus declared its contents by name only, while fileio.c:1031-1040
loads a member by CRC when the name is absent. The 61 members now carry
the CRC their driver declares, so stvbios, megatech and the six ZN sets
keep matching the archives the collection holds.
_foreign_tree compared owner tokens by equality, so cxbx-reloaded-2023
read bios/Other/cxbx-reloaded/ as another emulator's tree and the
case-folded fallback served xemu's encrypted EEPROM. The leading whole
words of a profile key now name its own tree too. Across every profile
entry this moves cxbx-reloaded-2023's EEPROM.bin to the decrypted dump,
turns eight name matches on the same files into path matches, and gives
solarus-2.0.0 the Solarus tree's gamecontrollerdb.txt.
Abdessamad Derraz