GitLab refused the hundred raw reads per IP. A GitLab directory is now
read from one archive request, else a shallow sparse git fetch, and the
cache keeps its file list so offline runs know what is absent.
Each board builds system.ini against its defconfig: a core whose package
the board does not enable is dropped (pisnes was the SNES default
everywhere), and a system with no core is not built. The row names the
default most boards share, board_defaults the others. A files: list
names whole files and is carried as files through the index and the
installer.
A folder no name resolves, run by a core of several machines, now holds
those machines: thomson, multivision and pcv2 had no fallback, so their
games got no file. Arcade-indexed cores keep naming files per set.
board_cores walked only the plain [core.N] sections and replaced a board
section whole. romfs2 systemholder.py reads a slot present as
[core.N.<board>] alone (gb and gbc bsnes on x86_64) and each key from the
board section, then <key>.<board>, then the plain key. The committed
target lists are unchanged on all seven boards.
A section shares its page's search fields, and "playstation 3 bios" found the PlayStation page before the Sony page that held the PS3. CONSOLE_PAGES splits Sony by console whatever the size. The search gate now expects the PlayStation 3 page in the top two for that query, the order of words being something Pagefind does not weigh, and the page first for "ps3 bios".
The games importers wrote through fixed .part names, the new options had
no help text, and romfolders stopped on the first setup the offline cache
lacked. RetroArch's rom folder map is regenerated from its source spec
(path null, as the schema says). The installer's Python 3.8 job imports
the games tests without pyyaml, so the region rules load lazily.
--for-games reads the card's rom folders through install/games/ and keeps
what the games need: per system what its default core reads, one region
where the content proves it (crc32 and size, a disc serial, a header the
core reads), an arcade set's BIOS parents. A game not recognised gets
every BIOS of its system and is named. --required-only keeps only what
starts. The plan and the line menu open on it when the card holds games,
everything one press away. A cache keeps what reading a card found.
mame2000, mame2003 and mame2003-plus carry a DAT their code moved past. Their parents and BIOS sets now come from src/driver.c and src/drivers/ at the pin, named by the src tree sha. The DAT is still read to report gaps. FBNeo DATs are read one namespace each, console sets keyed by driver. A BIOS with no dumped ROM is no longer listed.
The games index now names, per arcade core, the game sets that load no
BIOS set and every BIOS set, so a set on a card the DAT does not know is
told apart from one that needs nothing. Each board default core of a rom
folder gets its own fallback set, and index.json maps every folder name
a setup uses to its systems, for a games folder on a computer.
MAME's first card part showed 1,931 words against the 1,693 ceiling once
888 entries named their system: each row gained a "for <system>" badge
while parts were cut by card count. Parts are now cut by the words each
row shows with its card closed, balanced, at most 1,350 a part and still
at most 180 entries. MAME goes from 9 parts to 10, no part disappears.
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.
934 of the 1,563 mame.yml entries named no system and so served every
system of the profile: each MAME-run system's set was the whole profile.
The 65 BIOS root sets written by hand take the system groovymame.yml
gives the same 0.289 archive (sega-systemsp, konami-gx, taito-gnet and
41 more ids, added to systems:). The 19 XaviX sets take tvgames, the
directory MAME keeps xavix.cpp in. They seed the device walk.
mame_hash_scraper --devices files a device set under arcade when a
machine no entry places loads it: MAME starts such a machine from its
set name, as an arcade folder passes it. 1,106 of the 1,254 device sets
now carry arcade, alone or beside the systems of their placed machines.
qsound, ym2413 and ym2608, written by hand, take the same answer.
Left without a system: 41 device sets no machine of the listxml
instantiates (a software list slot selects them), and the hash/,
plugins/ and floppy/ trees every machine reads.
redump.info moved its BIOS DAT links from /static/bios/<name>.dat to one
route per system (/bios/PSX, /bios/PS2, /bios/GC, /bios/XBOX), served as
an attachment. The scraper matched only the old shape and found nothing.
Both shapes are read now.
Refreshed snapshot: the PS2 DAT moved from 2026-06-16 to 2026-10-10,
191 entries become 225 (34 mechacon, DVD and font ROMs added, the
ps2-dvd-302c.bin sha1 corrected, five descriptions completed). Only
the provenance fields of database.json change.
The gaps page said the exporter regenerates each platform's file, which it patches, and opened on 48 missing above a heading of 10 files. The provenance total sat above three lists that add up to more, and the cores page counted every profile as run by the platforms.
The sentence fits the page's 400-word ceiling and its 20-word sentences.
A test reading the site generator skips under Python 3.8, as the others
do, and the core plan test names the cores --core gave.
Abdessamad Derraz