The pack integrity check rebuilt the core extras with no taken paths and
tested presence alone, so it credited core files at paths the platform's
own file holds and printed a core count the builder contradicted. It now
replays the builder's placement through packextras.hidden_extras and
compares each placed member with the copy the builder resolves.
archive_members scanned the whole file list on every call, and callers ask once per entry of the same profile: 38 million entries walked in one verify --all over mame.yml. Lists of 64 entries or more are now indexed by archive once (held so the id is not reused, indexed again when the length changes), members still read from the entries at each call. archive_members drops from 2.30 s to 0.14 s tottime under the profiler. verify --all (text and --json), slots --json and the retroarch manifest are byte-identical.
Run from another directory, validate_pr.py read no database, platform or
profile and reported every file as referenced by no platform and outside
bios/. It now moves to the root as verify.py does, reading the files
named from where they were typed.
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.
A browser upload into bios/ lands on a fork and opens the pull request, and CI already hashes the file and names its folder. CONTRIBUTING asked for a fork, a clone, five packages and the whole suite, and promised a PR comment that a fork's read-only token cannot post. The issue chooser links the upload page.
A row said the declared hash was contradicted by the local file and
named neither. The table now prints the hashes the platform declares
and the path and hashes of the copy the collection holds.
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.
The quick install block gains the PowerShell form that passes options and a link to the installer page. The collection size uses the GB format of the download table and the site, counts get separators, and the sentences about RetroArch's existence check and the method describe the code instead of judging documentation.
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.
Step 2a advanced mame.yml's core_version and upstream_commit to a new
release and left the device ROM sets on the previous one. A plain run
that moves the pin now fetches that release's listxml into .cache/dats
and runs the device merge, which checks version and commit itself.
asset_candidates fell back to the bare name when no .gitignore line carried it: every online build asked large-files for aes.zip and five others and printed an HTTP 404 per name, pointing at an upload that never existed. The tests that fetch now register their names.
load_from, found_by, on_mismatch, built_by, pin_frozen and
version_source were in the profiles and on no page: Dolphin's
gba_bios.bin showed a path under system/ although the libretro core reads
it from its save directory, and mame0278 did not say only Recalbox ships
it.
The acquisition report never read a profile, so XM7's required DICROM.ROM
sat among 2 287 targets at the rank of dumps no core reads. A missing
entry that a profile reads and the collection lacks is now listed first
with the profile, matched by hash or, for a hashless entry with a size,
by name and size. 29 entries on the current snapshots.
A crc32 and a size can be met by crafted bytes, so a remote download
matched on them could land in bios/. verify_content now requires every
sha256, sha1 and md5 the entry declares to match, and an entry declaring
none is reported as not verifiable from the network and nothing is
written. crc32 and size still find a file the database already holds.
check_crypto_validation keeps its answer keyed by the file and keys file as compute_hashes keys them (path, inode, size, mtime, ctime). verify --all ran the verifiers 35 times for 5 files (0.88 s), now 5 times (0.13 s). verify --all --json output is byte-identical.
164 entries wore the in repo badge with no copy named: they are held as
a ZIP member or in a data directory, which the database does not index.
held_copy() finds that copy and the card names it. storage: large_file
no longer counts as held by itself: kenji-nx's user-picked firmware
package read as held although nothing collects it.
verify_content read sha1 then md5 only, so an entry pinned by sha256 or
crc32 could be searched and never accepted, and one without any hash was
announced as searched. The strongest declared hash now decides, crc32
counts with its size, the database lookup reads sha256 and crc32, and an
entry with nothing to prove a download by is reported, not searched.
The second pass, the archive-prefix pass and the agnostic scan built
extras without source_name or source_path, so verify found no gap report
entry for them and counted game data such as hiscore.dat as core files.
Every extra now carries its entry's category and key, and verify reads
the category from the extra the builder places.
Equal files have equal sizes, and the scan only uses the sha1. Files alone at their size are no longer read and the rest get one digest instead of five. Same 278 groups (645 paths), same dry-run output, 36.7 s to 1.9 s.
load_target_config raised out of cross_reference.py as a traceback. The hint now names the full verify.py command, since cross_reference and slots have no --list-targets.
download.py authenticated its release lookup and download.sh did not, so
past the anonymous quota the shell twin failed with 403. The header goes
to api.github.com only and is read from stdin, out of the process list.
check_freshness exited 0 when no upstream answered. An unknown row now
holds the exit code, and a profile skipped by nature is reported skipped
so it does not hold it forever. --only names the valid areas, a stale
native row names its platform, and the summary names refresh_stale.py.
The summary in exports/<platform>.json read '82 hashes corrected' and '854 additions left out'. It now counts what the written file carries, in the exports page's terms (hashes updated to the emulator source, required flags aligned with the code). Files the format cannot carry, withheld updates and content-found labels are printed on a line of their own and stay in the report's counts.
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.
refresh_stale ran a parent's rescrape and its children's truth and
export jobs side by side, so a child read the parent mid-rewrite. A job
now starts once the rescrapes of its ancestors have ended, and a child
no row names gets its export rewritten after them. A rescrape chain ends
on verify.py --platform, whose summary closes the report.
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.
verify --emulator listed only OK destinations, while the emulator pack
also writes a copy whose hash the profile contradicts, loose or archive.
pack_destinations now records the paths the replay places, and both
branches resolve through packresolve.resolve_file as the pack does.
When two cores declare different values for one file, the truth keeps the first and marks the entry hash_conflict. Its hashes pair an entry and never replace the platform's: armsx2's GameIndex.yaml no longer overwrites System.dat's line for LRPS2.
Bios::Scan takes Zip::Md5Composite for every .zip (Bios.cpp:64-68). The resolver already honoured zip_digest: composite, but digest_verdict accepted the container's own md5 first, so the pack check passed bytes Recalbox marks red. The page described that order.
Step 1b read 18 924 held files and members for 7 407 distinct contents and ran every transform on each read. A content already transformed is now skipped (the first path still keeps the match), the power-of-two crc shift operators are squared once for the pass instead of once per length, and a group pairs members by the sizes it holds. derivable.json and the step output are byte-identical, 33.4 s to 25.1 s.
export_native and cross_reference read data/ without the shared hold, so
a refresh could swap a tree mid-walk. hold_data_dirs skipped a cache not
yet on disk, so a refresh could install it between verify and the packs
of one pipeline run. Both now hold every registered tree.
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.
FileConsumers resolved platform declarations with their destination as
a hint, which the builder, the manifest and verify never pass. RetroBat's
63 mame/hash lists and the curated gc-ntsc-12.bin credited another copy
than the one the manifests install.
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.
The FBNeo import recorded the listing's blob sha for bytes fetched from a
branch URL a CDN caches, and --fetch <revision> still read the tip. The
revision is now resolved to a commit, listed and downloaded there, each
DAT is held against its blob sha, and a DAT read at a named revision is
labelled with it without replacing the tip's recorded blobs.
No platform YAML holds an md5 shorter than 32 characters: the Batocera scraper completes one when a single collection md5 starts with it. The pages said some entries carry 29 characters and that scrapers store them as-is.
The database keeps one path per content, so a system placed by the
copy deduplication kept linked to another maker: nintendo-nes and
nintendo-fds to Arcade > FBNeo, sega-mastersystem and
sinclair-zx-spectrum to Other > zesarux. Every folder holding a copy now
votes, an id that spells one console folder links to it, and a maker's
folder outranks the folders that file copies by emulator.
A module-level alias written with a PEP 604 union compiled under 3.9 and
raised TypeError when check_release_assets.py, and check_release.py
through it, were loaded. The floor test now imports every module under
scripts/ with a real 3.9 interpreter, which CI gives pyyaml and
jsonschema.
The pipeline locked only --output-dir while it also writes database.json,
install/, README.md and docs/. A run on a scratch output directory could
start beside the run on dist/ and rewrite them. Every run now holds the
repository's .lock first.
_patched_entry read '#{ ... }' lines as live items, found no match in the model and gave up, so atarist was written again whole (52 lines changed for 3 additions). A commented item is now left as written.
--system compared the raw id, so the normalized spelling printed by verify.py --list-systems selected 1 Saturn profile of 11. The hint now counts the spellings that listing prints.
release_record rebuilt nothing from the name builder: a substring match
took RetroArch_v1.22.2_AndroidArm64V8A_BIOS_Pack.zip for the RetroArch
pack. The record now refuses any pack that is not a registered platform's
full pack and stores the platforms each pack serves, which the readers
use instead of the pack name.
check_release_assets.py built a tuple[..., int | None] alias at import, a TypeError on the 3.9 floor tools.md states, and check_release.py imports it. A test reads every script for unions and match statements evaluated at runtime.
RomM lists the platform firmware folder without recursing and looks each file up as slug:file_name (models/firmware.py, base_handler.list_files at 5.4.0). An addition named with a directory (fuse/48.rom) is not written.
--detect-new-files and --watch-hashes were printed only by the text report and refused with --json, so check_freshness --profiles never saw a name or hash the cited code gained. Both now live on the report, and check_freshness passes them and marks such a profile stale.
Abdessamad Derraz