https://gitlab.synchro.net/main/sbbs/-/commit/bd8561267a9eaa29c4c2601a
Modified Files:
src/doors/syncretro/main.c syncretro_audio.c syncretro_audio.h syncretro_io.c
Log Message:
syncretro: stop the audio dead on exit, and swallow the last keystrokes

Two things the player sees after quitting.

The audio kept playing. SyncTERM's do_flush() clears a channel only
when no fade is asked for; given O=, it queues a silent crossfade
against the head buffer and leaves the rest of the FIFO in place (syncterm/audio_apc.c). The door runs three chunks deep, so a faded
flush let ~300 ms of game audio play on after the terminal was handed
back to the BBS. Flush without a fade. An abrupt stop is what a
player expects at the moment they quit, and it is the only spelling
that actually empties the queue.

The BBS typed at them. The door restores translated-key mode before
it exits, so any key still held is re-sent as a plain character that
nothing reads; Synchronet's input thread resumes, finds the bytes,
and echoes them at its prompt. Build a fort in Utopia, quit, and the
BBS greets you with "3141". Swallow input for a moment after the
restore, bounded by a deadline and a byte cap.

Also: Ctrl-R now flushes the channel, which the design promised and
the code did not do -- the queued audio belongs to a game that no
longer exists. Pausing drops any chunks held mid-prime for the same
reason. And the flush is skipped entirely for a terminal that never
confirmed libsndfile, so it really does hear nothing from us.

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

https://gitlab.synchro.net/main/sbbs/-/commit/e6b9c8adcfa880a43a5420be
Modified Files:
src/doors/syncretro/main.c syncretro.h syncretro_config.c syncretro_input.c syncretro_io.c
Log Message:
syncretro: answer an unbound key, and trace what a terminal sends

Two changes to the input path, one a feature and one the tool that
explains it.

A player who presses a key that does nothing is asking what the keys
are, so answer: an unbound printable key now raises the key legend
instead of being silently dropped. That makes the pause screen and the
help screen the same screen -- both list the keys, and only the banner
and the footer differ -- so they collapse into one sr_screen_keys(),
and '?' behind the pause screen has nothing left to add. The legend
also carries the two hints a first-time player needs: many cartridges
read the RIGHT hand controller, and most games start on a keypad digit
rather than an action button.

The diagnostic is the more useful half. `keytrace.on` in the door's own directory turns on a log of every inbound byte, tagged with the key
mode in force, alongside the frame-pacing decisions (sent, dropped,
de-duped, acked) and a once-a-second sample of the pad the core sees.
It answers what no amount of source-reading can: what a given terminal
actually sends for a given key, and whether a frozen picture is a
stalled door or a core that has the input and draws nothing. Both of
those questions came up chasing a paddle that moved one tick per
press-and-hold in Brickout!, and the trace settled each of them after
static reasoning had guessed wrong three times.

A touch-file rather than an environment variable because bbs.exec()
starts a door through execvp() with the BBS's own environment, so a
sysop cannot set a variable without restarting the BBS -- and under
EX_BIN the door's stderr is discarded, so the output has to be a file.
Both paths resolve against the new sr_config_launch_dir() rather than
cwd: sr_config_apply() has already chdir'd into the per-user sandbox by
the time a key arrives, so a relative path would quietly land under data/user/NNNN/. The launch directory was already captured there; it
just had no accessor.

Off unless the switch file exists -- not even a file handle.

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/9c924cc2abf225d820686b01
Modified Files:
src/doors/syncretro/main.c syncretro_binds.c
Log Message:
syncretro: stop the PAUSED badge bleeding red, and make the version legible

Three fixes to the pause/help screen, all of them visible in a screenshot.

THE BADGE BLED. A background attribute is STICKY, and the SGR that followed
the badge (A_FRAME) names only a foreground -- so the red stayed set, and the rule glyphs after the badge, up to and including the corner, were drawn on
it. Reset before returning to the frame color, and never leave a background live across a glyph. Verified on the wire: zero box glyphs are now emitted while a background attribute is set.

THE VERSION WAS DARK BLUE ON BLACK, which is to say it was not readable at
all. Light gray, with synchro.net in cyan.

THE KEY SEPARATOR WAS '|', WHICH IS A KEY. In a legend that says "press
this", a pipe between two key names reads as a third thing to press --
"W A S D | arrows" invites the question. It is now the charset's box-drawing vertical bar, which cannot be misread as something to type.

That last one is why the separator is not simply a character in the bind tables: it is one byte in CP437 and three in UTF-8, and the tables are shared by both. They carry a marker ('\1') instead, and the help renderer expands it to the client's glyph and counts it as one column, so the padding still lands.

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

https://gitlab.synchro.net/main/sbbs/-/commit/3f084da749bab8f33df32ce4
Modified Files:
src/doors/syncretro/main.c syncretro.h syncretro_binds.c syncretro_binds.h syncretro_input.c syncretro_io.c
Log Message:
syncretro: add the text render tiers and the F4 tier cycle

SyncRetro drew sixels or nothing. A terminal without them -- conhost, PuTTY, anything old -- got a door it could not play. termgfx already has the block-character renderer the sibling doors use; wire it in, and let F4 walk
the tiers so a player can pick what looks best on his own screen rather than taking ours.

The cycle is built from what the CLIENT can do, never from a menu of
everything we can emit:

sixel only if DA1 answered param 4. A terminal that did not say so
starts on TEXT instead of having sixels printed at it as
garbage.
half-block one glyph, two pixels. Renders in EITHER charset (CP437 0xDF,
UTF-8 U+2580), so every client gets it.
blocks+shades 2x2 detail from the classic CP437 block/shade glyphs only --
it renders on font-limited terminals where the quadrant
glyphs come out as missing-glyph boxes.
quadrant 2x2, exact (U+2596-259F). UTF-8 CLIENTS ONLY.
sextant 2x3 (U+1FB00). UTF-8 CLIENTS ONLY.

The last two are withheld from a CP437 client on purpose: cycling SyncTERM through them would step it through two screens of missing-glyph boxes. The charset comes from termgfx/charset.c ($SBBSNODE/terminal.ini), so the door follows the client rather than guessing.

F4 has no byte form. It arrives as SS3 "ESC O S", as a CSI ending in 'S' (xterm's modified form, and kitty's), as ESC[14~, or -- on SyncTERM with physical-key reports -- as evdev keycode 62 and no escape sequence at all.
All four spellings land in one place. The tier switch erases the screen and throws away the cell-diff shadow: a sixel is pixels laid over the grid and
the text tiers write cells, so neither overwrites the other, and a shadow
that still believes it knows the screen would leave half a picture behind.

The text path takes the same DSR-ACK backpressure and the same byte
accounting as the sixel path -- pacing is about what the TERMINAL has drawn, not about what we drew it with -- and reserves the bottom row, which belongs
to the stats strip.

The stats strip now names the live tier (it used to say "sixel" always, which with a tier cycle would simply be a lie), and it erases to end-of-line with
its own background set, so BCE paints the row into a clean full-width bar and wipes the tail of a longer previous readout in the same stroke. Erasing first -- which is what it did -- cleared the row to black and then painted a short cyan stub on it. The sibling doors have always had this the right way round.

The pause screen sheds ornaments when the terminal is too short: adding F4 to the key list was, by itself, enough to push the bottom border off an 80x24 screen.

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

https://gitlab.synchro.net/main/sbbs/-/commit/b226d595468ca7f97f7006da
Modified Files:
src/doors/syncretro/main.c syncretro_audio.c syncretro_audio.h src/doors/syncscumm/door/sst_io.c src/doors/termgfx/audio_stream.c audio_stream.h
Log Message:
termgfx: make the audio stream dB-native (volume_db, dB stepping)

The shared PCM stream module still carried its channel level as an int
0..100 percent, converting to dB only at the A;Volume call -- the last
percent holdout after the volume-is-dB migration. Move it to a float
volume_db (0 dB = unity); the mute sentinel moves from `volume <= 0` to `volume_db <= TERMGFX_DB_MUTE`, still stopping the uplink entirely.

termgfx_stream_volume_step now takes a dB delta and returns dB, capped at
unity (a pre-mixed stream gets no boost from + / -); stepping down past a near-silent floor (-36 dB) snaps to real mute so the player need not press
all the way to the -60 dB floor.

The two stream doors adapt: the sysop [audio] volume knob stays a friendly 0..100 percent (converted to dB at read); the + / - hotkeys step 6 dB; and SyncRetro's on-screen toast now reads dB. Behavior at the defaults is
unchanged -- SyncSCUMM -6 dB, SyncRetro unity -- the live control simply
steps in even dB now rather than percent.

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/f15366d5db3bfad5bfc56f1a
Modified Files:
src/doors/syncretro/main.c
Log Message:
syncretro: the key screen says PAUSED however the player opened it

One screen serves Space, '?' and an unbound key, and the core is stopped
behind all three of them -- the door does not run it while any screen is
up, and it pauses the audio too. Only the Space path drew the PAUSED
badge, so a player who hit '?' saw a key list with no badge on a game
that had, in fact, stopped: the badge was reporting which key opened the
screen rather than what the core was doing.

Draw it unconditionally. The closing prompt still differs, because only
the Space pause needs a particular key to leave, and that is now the parameter's only job -- renamed from_pause so it cannot drift back into
meaning "the core is stopped".

Verified headless (-stdio, fceumm) by scripting '?' then Space and
decoding the wire bytes: both screens carry the badge, both top rules
measure 76 visible cells with the closing corner intact, and the prompts
still read "any key to return" and "Space to resume" respectively. The
fill arithmetic was re-checked with a long -console name, which is the
tightest the rule gets.

Reported by Rob Swindell.

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/3a9fb3c143e7a079afd76189
Modified Files:
src/doors/syncretro/main.c retro_core.c retro_core.h syncretro.h syncretro_binds.c syncretro_config.c syncretro_door.c syncretro_profile.c syncretro_profile.h xtrn/syncarcade/lobby.js
Log Message:
syncretro: an arcade controller profile, and controller-device selection

TWO CHANGES, both about input, neither about the wiring.

The arcade profile binds EXACTLY what `pad` binds -- a control panel is a
stick and a row of buttons, which is what a RetroPad already is -- and exists for the words. `pad`'s labels are a console's: it calls Backspace "Select"
when on a cabinet it is the COIN SLOT and nothing happens at all until it is pressed, calls the stick a "d-pad", and names six buttons "B/A/Y/X/L/R" after
a gamepad's face rather than the six a fighting game has. It also drops the
Tab controller-swap, which on a cabinet only moves the player onto player
two's controls -- where his coin and start are different inputs again -- and leaves him with a dead machine and no clue why.

Buttons are numbered 1-6, not named. MAME 2003-Plus reports what a driver
wants ("Supports 6 distinct button controls" for Street Fighter II, 0 for Pac-Man) but sends no SET_INPUT_DESCRIPTORS, so the core cannot say WHICH arcade button each RetroPad id became. Numbering them in the core's own order is the most the door can honestly claim; "punch" and "kick" would be a guess, and a wrong guess on a help screen is worse than a vague one.

Separately, retro_set_controller_port_device() is now resolved and callable.
It never was, so a core offering alternative controllers stayed on its default forever -- MAME 2003-Plus advertises RetroPad, Fightstick, 8-Button and 6-Button through SET_CONTROLLER_INFO and we could select none of them. Driven by [input] device, defaulting to 0 = say nothing, which is exactly the
previous behaviour: a wrong device id silently rewires every button, so this
is opt-in rather than a guess applied to every console.

Verified: -profile arcade resolves and an unknown profile still falls back to pad with a warning; [input] device = 769 selects 6-Button on sf2 and, with the section absent, the door emits no selection at all.

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

https://gitlab.synchro.net/main/sbbs/-/commit/67a84e4e47cc58919b79fb4a
Modified Files:
src/doors/syncretro/main.c
Log Message:
syncretro: load games.ini at startup for the help screen

Call sr_games_load() right after sr_profile_select(), passing sr_config_launch_dir() rather than relying on the cwd: by this point sr_config_apply() has already chdir'd into the per-user save sandbox,
so a relative fopen("games.ini") would silently look in the wrong
place. This is the last wire connecting the games.ini reader (already
built and tested) to the running door.

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