check_platform_folder.py, the script the site publishes, still accepted
a cut md5 by prefix. It now compares whole digests as Batocera does
(batocera-systems:1088), like verify since cut_md5_owner.
Mini vMac opens no MinivMacBootv2.dsk at babdf7b5: disks come from a .cmd
(libretro-core.c:181-195) or disk1.dsk to disk9.dsk
(OSGLUERETRO.c:563-579). The name is the one Recalbox's configgen writes
into its .cmd, already recorded in its launchers: layer. Nine manifests
no longer send the 24 MB image to platforms that never open it.
IPL_USA.n64 and IPL_DEV.n64 are the names RetroBat's mupen64 generator
writes for Rosalie's Mupen GUI (Mupen64.Generator.cs:232-238).
Mupen64Plus, Mupen64Plus-AE and M64Plus FZ fix no IPL name: the path
comes from the IPL-ROM key or --dd-ipl-rom, then the Android front ends
copy the chosen file to dd_rom.n64.
Six places read an md5 shorter than 32 digits, each with its own bound.
digest_verdict accepted any prefix down to one digit, while Batocera
compares the whole string (batocera-systems:1088). A cut value now names
a dump only when one held md5 starts with it (cut_md5_owner, read by the
scraper, the resolver and --from-md5), and every verdict compares whole
digests.
The reason named only MSX and FPseNG fonts, while fm7.zip and fm77av.zip
hold a Fujitsu kanji ROM (MAME kanji.rom, crc32 62402ac9). Rendered at
the codes kanji.c:353-380 swaps between JIS78 and JIS83 (0x3646, 0x445B,
0x4768), it draws the JIS83 forms: the ROM is a first standard JIS83
table, not the JIS78 one XM7 reads as KANJI.ROM.
gitignored_paths read a missing file as an empty registry: nothing to
restore, nothing unsatisfied, exit 0 with every release asset absent.
The run now stops with the path it could not read.
_pin_verified returned an existing <sha1>/<name> link without checking
its inode. A link whose cache name was gone had one link left, and the
prune of the same download removed it after it was returned, so
fetch_large_file handed out a path that no longer existed. The link is
now replaced when it holds another inode, and a revision directory a
prune removed before the link is made again.
_collect_emulator_extras recorded a losing claim's path without the
declared root, while the winner's was rewritten with it. On RetroDECK
verify --region kept Rustation and Mednafen STV copies of withdrawn
bios/ files in undeclared_files as packed (7 under japan, 4 under
north-america, 3 under europe).
_pack_version took the text after the last underscore of the pack name.
AmberELEC's prerelease-20261019_1835 read as 1835 and the page would
announce a version gap that does not exist. The site now removes the
platform part the builder writes (platform_pack_prefix).
--emulator handy --required-only removed the empty ZIP, printed the
path it had just removed and exited 1, while verify reported 0/0 OK and
exited 0. The builder now says no file is needed and succeeds.
Geargrafx drops a 512-byte header before its crc32 (src/media.cpp:754-773
at 7c447b0b), and its profile lists both sizes. slots compared the crc32
of the whole file, so EmuELEC's headered syscard3.pce read as contested.
MiSTer and BizHawk declare cores: []. --region, --one-per-slot and
--no-game-data read the profiles of the cores a platform runs, so their
packs and manifests withdrew nothing under a name and README that said
they had. generate_pack and verify now refuse the flags for a named
platform and leave it out under --all.
dsi_bios7.bin and dsi_bios9.bin hold the DSi Panda devkit boot ROMs, eMMC
key-Y zeroed. Every md5 and crc32 declared under those names designates
a retail dump in the same directory, yet hashless entries (RetroBat,
dsperate) got the devkit ROMs by name. by_proven_name maps such a name to
the proven primaries beside it. kick13.rom, dsi_firmware.bin and
spectrum.zip follow the same rule.
_ships and _emulator_ships resolved without the data directories, the
archive index and the slot override the builder uses: 400 platform
declarations and 178 emulator entries read as missing while the pack
held them.
A search that found nothing sent any short text, which a name could be.
Its words now go out only as a file name or a hash, every other search
is counted as "other" without them.
Pages, referring domains and a few actions (searches that found
nothing, downloads, missing addresses, "Did this page help?") are sent
to GoatCounter. Nothing is stored on the device, and nothing is sent
under Do Not Track or Global Privacy Control. wiki/privacy.md says what
is sent.
EmuELEC's scripts/image writes COREELEC_DEVICE to os-release (v4.8, line
189), as a GameForce image shows. The installer reads it the way it does
for AmberELEC and applies install/targets/emuelec.json. MiSTer FPGA takes
EmuELEC's place in the test of a platform without targets.
The v4.8 release publishes four images (Amlogic-ng, GameForce,
OdroidGoAdvance, OdroidM1, read from its asset names). The metapackage
(packages/sx05re/emuelec/package.mk:8-77) builds the S922X cores, dolphinSA
and mame only on Amlogic-ng, Amlogic-no, RK356x and OdroidM1, and removes
duckstation on the handhelds with a sed that replaces the first match only:
pcsx_rearmed, written twice, keeps its 64-bit build. Each package is mapped
to its cores by the .so or program its install step copies, PKG_ARCH is
applied as scripts/build:28-30 does, and es_systems.cfg is edited per
device as emuelec-emulationstation/package.mk:114-143 does. Amlogic-ng and
OdroidM1 list 124 cores, the handhelds 117.
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.
check_source_paths read the trees profiles pin from source and upstream
only, so the two platform trees the record now holds for pin_build
files read as orphaned and the record as stale.
reading-reports quotes a RetroDECK run, which now prints the bundled count and 30 missing game data instead of 38. The comment introducing the bundled names also loses its semicolon.
test_twin_declaration_status and test_verify_data_dir_notice join the testing guide table, and test_large_file_cache calls unittest.main after its last class.
Walking the systems sorted reordered the declarations doubled.py
reports from, so the export report named the CoCo reader first. The
declarations keep the platform's order again, and the packed file is
picked from the sorted walk the builder makes.
A search for gba_bios.bin or scph5500.bin landed on a page described as
"It reads 4 files, none of them required". The snippet now names the
console and the files, required first, in the longest form that fits.
The --core answer added 14 words to a page held at 450. It now names the
command in as many words as the line it replaced, and the --region line
goes: what --region keeps is on the installer page.
Every source link on the site leads to a path the source record saw at
a commit. The Build file row linked pin_build files unchecked, eight of
them at commits the record did not hold. build_source_paths now records
the build files read at a commit, and the row links only those: a tag
can move and stays plain text. The record run also picked up two cited
paths it had not seen before (pom2 CMakeLists.txt, pcem configure.ac).
generate_manifest called slot_choices, and pack_overrides called it again with the same five arguments. pack_overrides takes the choices when the caller holds them. Measured on generate_pack.py --all --manifest --offline: 48 calls and 7.0 s in slot_choices before, 24 calls and 4.3 s after (wall 45.7 s and 42.4 s, one run each on a loaded host). The manifests written are identical.
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).
A data archive cut mid-download was deleted on every failure: the retry and the next run started from zero, while the run said nothing was lost. The archive is now kept beside its tree with the validator its server sent, and the rest is asked for with Range and If-Range. Checked against the release: fbneo-samples stopped at 16 MB, the next run downloaded 34.6 MB of 50.5.
Abdessamad Derraz