https://gitlab.synchro.net/main/sbbs/-/commit/cb3e68711619be177d113e6a
Added Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncretro: parse a cartridge's title, year and publisher from its name

A ROM set names files "Title (Year) (Publisher).int", so the launcher
needs no database to show a shelf. Titles carry their own parentheses
-- "Adventure (AD&D - Cloudy Mountain)" -- so the year is the LAST
four-digit group, not the first, and dump markers like "[!]" belong to
neither the publisher nor the title.

A name that does not fit the pattern keeps its whole self as the
title. A wrong guess reads worse than no guess.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/8f872c3eea13c7cb9817d226
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncivision: handle stacked dump markers and year ranges in titles

sv_parse_title() strips only one trailing "[...]" dump marker, so a
name with two stacked markers (e.g. "[a1][!]") keeps the inner one
and misses both parse regexes. It also expects a bare 4-digit year,
so a range like "(1982-83)" doesn't match either.

Measured against the real 235-name cartridge set, these two gaps
account for all 4 titles that fell back to the whole filename. Strip
markers repeatedly, and accept an optional "-NN"/"-NNNN" suffix on
the year, taking the first year of a range. Fallback count on the
real set drops from 4 to 0.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/28a76a9bbcdb834d88db856e
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncretro: discover cartridges in a dumped ROM set

A sysop drops a ROM set in and it is not a curated folder: it carries
XML and torrent litter, BIOS images that are unplayable as cartridges,
and short aliases of the same bytes. So the launcher decides what is a
game rather than trusting the directory.

Whitelist the cartridge extensions, reject anything outside the
Intellivision's 2 KB to 64 KB range, and reject the two BIOS images by
CONTENT, so a re-dropped set cannot smuggle one back in under a
cartridge's name. Dedupe by size plus a hash of the first 4 KB: over
the install's SMB view a symlink is indistinguishable from a regular
file, so readlink cannot help and only the bytes can. The more
descriptive of two names wins.

md5_calc() returns base64 unless asked for hex. Ask.

A name carrying a double quote, a dollar, a backtick or a backslash is
reported unsafe: the door command line wraps the path in quotes, and
those are the only characters that escape them.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/eb57a904e8d59589281aaeef
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncretro: keep a play-activity log

One JSON object per completed play, appended. No aggregate counts
file: data_dir is an SMB mount a second host also writes, so a read-modify-write of a counts file is a lost-update race waiting to
happen. Most Played and Your Last Game are derived by scanning, which
at BBS scale is a few thousand lines.

A torn or hand-edited line is skipped rather than fatal, and a play
that cannot be logged returns false rather than throwing: losing the
record is not worth failing a player's session over.

No in-game scores. Nothing in the libretro ABI reports one, and
reading Astrosmash's points means a per-ROM RAM-address table that
works for this core alone.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/3da06552004331704a338949
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncretro: collapse dump-quality variants in ROM discovery

A ROM set carries several dumps of the same game that differ only by
a trailing marker the title parser strips -- "... [!].int" beside
"... [a1].bin". Their bytes differ, so the content dedupe (size plus
a hash of the first 4 KB) cannot merge them, and the picker showed
two or three identical-looking lines per game, some of them known-
inferior rips. 12 of the 221 cartridges in the reference set were
affected.

sv_dump_rank() ranks a name's best trailing marker: 0 verified [!],
1 no marker, 2 alternate [a*] (or anything unrecognized), 3 overdump
[o*], 4 bad dump [b*]. sv_collapse_variants() groups by title + year
+ publisher (publisher stays in the key so different ports, e.g. the
Intv Corp and Atarisoft Pac-Man, are never merged) and keeps the
best-ranked entry per group, tying on the lexicographically smallest
name so the result never depends on directory() order. sv_discover()
calls it after the content dedupe and before the final sort.

This hides files that exist on disk -- it is a display-layer pick,
not a delete. The sysop's escape hatches are unchanged: [roms]
exclude= in syncretro.ini, or deleting the losing dump directly.

Tests cover the rank table (including stacked markers, where the
best marker present wins), the collapse itself, the Pac-Man-ports must-not-merge case, and a same-rank tie.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/694d7815119d36ed9bbbacff
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncivision: dedupe sv_discover() on the full-file hash, not 4 KB

sv_discover()'s dedupe key was size + md5(first 4 KB). Against the
real 221-cartridge set that collides across genuinely different
cartridges of the same size -- "Pac-Man (1983) (Atarisoft).int" and
"Pac-Man (1983) (Intv Corp).int" are two different ports that share
their first 4 KB but diverge later in the file, so the dedupe was
silently dropping one of two real games from the picker.

sv_discover() already computes the full-file md5 for every size-gated candidate, to check it against SV_BIOS_MD5, so keying the dedupe on
that same full hash costs nothing extra. Compute it once into `full`,
use it for both the BIOS guard and the dedupe key, and drop the
now-unused 4 KB head hash.

Added a discovery-section test that builds two same-size files whose
first 4096 bytes are identical but whose tails differ (named after the
real Pac-Man case) and asserts both survive sv_discover(), plus an
assertion that genuinely identical full-file bytes still collapse to
one entry.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/44e8f4b74654807c3ad4a759
Modified Files:
xtrn/syncivision/syncivision_lib.js xtrn/syncivision/tests/test_syncivision_lib.js
Log Message:
syncretro: lay the cartridge list out in columns

Two hundred titles do not fit a screen, so follow xtrn_sec.js: read
down each column, collapse to one column below eighty, and derive the
column count from the terminal's width rather than hardcoding two, so
a wide terminal gets three.

A cartridge with no year in its name shows no year, rather than the
year zero.

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net