https://gitlab.synchro.net/main/sbbs/-/commit/d22b21d57a0d129e795878a1
Added Files:
src/syncterm/scripts/syncterm.wren
Modified Files:
src/syncterm/re1/parse.y regexp.h y.tab.c src/syncterm/scripts/runtests.wren src/syncterm/wren_bind.c wren_host.c
Log Message:
SyncTERM: extract syncterm Wren module to scripts/syncterm.wren

The 740-line foreign-class-declaration string literal in wren_host.c
moves to scripts/syncterm.wren as plain Wren source. Now editable
with syntax highlighting, comments stay readable, no escape-quote
gymnastics. The build embeds it via the existing scripts/*.wren
glob alongside connected.wren and console.wren.

Load order: wren_host_init() now resolves "syncterm" first (preferring
a user override at scripts/syncterm.wren on disk, falling back to the
embedded copy), captures the Cell/Cells/KeyEvent/MouseEvent class
handles, then walks the rest of EMBEDDED_SCRIPTS[] and the user
script dir Ä both passes skip the syncterm entry to avoid double-load.

Bonus fix shaken out by the move: parse() in RE1 unconditionally
prepended `.*?` to every pattern for match-anywhere semantics, which
defeated the streaming dispatcher's IMPOSSIBLE-shift trim. Add parse_unanchored() that returns the user's pattern wrapped only in
the implicit group-0 capture; Hook.onMatch uses it. parse() keeps
its original behavior for any future buffer-shaped caller.

scripts/runtests.wren needed an explicit `import "syncterm" for Hook`
since it loads as its own "runtests" module (not embedded, not in
the syncterm scope).

Co-Authored-By: Claude Opus 4.7 (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/378af3b080bb6209bc6a3649
Modified Files:
src/syncterm/scripts/syncterm.wren src/syncterm/wren_bind.c wren_host.c wren_host_internal.h
Log Message:
SyncTERM: lazy-cache foreign class handles, move Cache into Wren

Cleanup after rev to per-module entry scripts:

* The four foreign-class handles (Cell, Cells, KeyEvent, MouseEvent)
were captured eagerly in wren_host_init, which forced an
`import "syncterm"` bootstrap to run first so the wrenGetVariable
lookups had a loaded module to read from. None of that was
necessary Ä the handles are pure caches around symbol-table
lookups, and by the time any allocation site runs, the calling
script has already imported syncterm. New load_class_into_slot
helper does the fetch-and-cache lazily; six allocation sites
(push/resume key & mouse, fn_Screen_readRect, cells_make_view)
call it at the slot they were already filling. The `if (... ||
st->X_class == NULL)` bailouts on those sites were guarding a
case the lazy fill makes impossible.

* `Cache` (the singleton Directory pointing at the script-cache
directory) was injected into the syncterm module from C via
wren_bind_define_cache, reaching into vm->modules and calling
wrenDefineVariable. Replaced with a Wren-side `var Cache = ...`
in syncterm.wren backed by a new foreign static
Host.cacheDirectory. Cache is now an ordinary module variable
instead of a C-injected ghost. A user override of syncterm.wren
has to include the same boilerplate (one line) to keep Cache
available Ä that's the trade-off, and matches the rest of the
module's "the override is responsible for the contract" model.

* The entry-script iteration's special-case skip for "syncterm"
is gone; both loops now use wrenHasModule to skip any module
another entry script's import already loaded. Same effect for
syncterm, plus handles any future library-also-embedded module
generically.

wren_host_init shrinks to: make VM  wire callbacks  cache
call/call(_) handles  set ownership  glob  run embeds  run
user scripts. No bootstrap, no eager class capture, no Cache
injection, no name-specific skip.

Co-Authored-By: Claude Opus 4.7 (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/44e412b47990ef7178a62032
Modified Files:
src/syncterm/scripts/syncterm.wren
Log Message:
SyncTERM: Cells is Sequence

The iterate / iteratorValue protocol Cells already implemented makes
it Sequence-compatible Ä declaring it `is Sequence` brings in all of
Wren core's iteration helpers (each, where, map, all, any, contains,
count(f), reduce, join, take, skip, isEmpty, toList) without any
extra C-side bindings.

The O(1) foreign `count` getter overrides Sequence's default
(which would otherwise walk the iterator).

Co-Authored-By: Claude Opus 4.7 (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/5f7ae2341062bd086a309ade
Modified Files:
src/syncterm/scripts/syncterm.wren src/syncterm/wren_bind.c
Log Message:
SyncTERM: Wren foreign toString overrides

The default Object.toString returns "instance of <ClassName>" Ä fine
for opaque values, useless for data-bearing ones. Adding overrides
on the foreigns where it earns its keep.

Natural-form (no class prefix; the value has an obvious string form):
Directory  path with trailing slash, or "Cache/" for the lazy
BBS-cache directory; "(dead)" suffix when invalidated
File  path with " (open)" / " (dead)" suffix
SFTPEntry  server's longname (ls-style) when present, else name

Debug-form ("ClassName(field=val, ...)"; no canonical string form):
KeyEvent  "KeyEvent('a' 0x0061)" with text when printable,
"KeyEvent(0x3B00)" otherwise
MouseEvent  "MouseEvent(event=N at X,Y)" or "...at X,Y-X',Y'"
Cell  "Cell(0xNN attr=0xNN)"
Cells  "Cells(count=N)" Ä metadata only; scripts can use
cells.join(", ") for content via Sequence
HookHandle  "HookHandle(calls=N)"
SFTPStat  "SFTPStat(size=N, mode=0NNN, mtime=T)"
SFTPError  "SFTPError: <name>: <message>" using the new
sftp_err_short_name helper for library codes and
sftp_get_errcode_name for SSH_FX_* server statuses

SFTPHandle deliberately keeps the Object default Ä the alive case
is no improvement, and "is the handle closed" wants an explicit
getter, not state buried in toString. ExtAttr / LastColumnFlag /
FlowControl / CustomCursor / VideoFlags also keep the default Ä
they're niche bitfield snapshots that scripts query field-by-field.

Co-Authored-By: Claude Opus 4.7 (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/6d6869c23e820d62f42fff6d
Modified Files:
src/syncterm/scripts/syncterm.wren src/syncterm/term.c wren_bind.c wren_host.c wren_host.h wren_host_internal.h
Log Message:
SyncTERM: useful wren-cli module pieces Ä Platform + Timer

Two small classes lifted in spirit from wren-cli, adapted to our
fiber-arg / result-queue model rather than wren-cli's libuv +
Fiber.transfer + Scheduler.await_ shape.

Platform.name returns the uname(2) sysname on POSIX (e.g. "FreeBSD",
"Linux", "Darwin"), "Windows" on Windows, "Unknown" elsewhere. Win32
checked first so POSIX-overlay environments (Cygwin, MSYS) report
the native OS rather than uname's emulated reply. No isPosix or
shell-exec / process surface Ä scripts on this host don't need it.

Timer.trigger(fiber, ms) registers a one-shot resumption: pushes a
TimerElapsed onto the result queue once xp_timer() reaches the
absolute due-time, then the standard drainer resumes the fiber.
Multi-fire / event-loop dispatch by type works straight out of the
existing pattern:

Timer.trigger(Fiber.current, 100)
SFTP.stat(Fiber.current, "/foo")
while (true) {
var x = Fiber.yield()
if (x is TimerElapsed) { spinner.tick(); Timer.trigger(Fiber.current, 100) }
if (x is SFTPStat) { ... break }
if (x is SFTPError) { ... break }
}

For recurring fire-and-forget callbacks (no fiber resume), use
Hook.every(ms, fn) Ä different pattern, both useful.

Implementation: bounded array of (fiber, due_s) on wren_host_state.pending_timers, swept once per doterm() iteration
just before wren_result_drain. Past-due entries push TimerElapsed
to the result queue with NULL data; the deliver fn builds a
TimerElapsed foreign on demand. Cap is 32; overflow throws on
Timer.trigger so a runaway script fails loudly rather than silently
dropping registrations.

Skipped: wren-cli's Scheduler (redundant with our doterm + result
queue), io (async File/Directory; ours is intentionally synchronous
on the owner thread), Process / shell-exec, Stdin/Stdout/Stderr
(SyncTERM scripts don't have a CLI).

Co-Authored-By: Claude Opus 4.7 (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/781b4e9f132487e70c47b9f6
Modified Files:
src/syncterm/scripts/syncterm.wren
Log Message:
SyncTERM: force CP437 fonts inside Screen.modalRun

Every Wren App enters via Screen.modalRun, which already snapshots
the screen on the way in and restores it on the way out. When the
cterm emulation has selected ATASCII / PETSCII / Prestel / BEEB,
all five font slots point at that emulation's character set, so
panes drawing with CP437 codepoints (box-drawing chars, shaded
blocks, ANSI accents) render as the wrong glyphs and the typed-
character codepage translator -- which keys off slot 1 via
ciolib_getcodepage() -- pipes input through the wrong mapping.

Stamp slots 0..4 to font 0 ("Codepage 437 English") right after
Screen.save() captures the originals. No per-App changes
required: every modal pane (online_menu, capture_menu, music_menu, scrollback_view, sftp_app, disconnect_flow, ...) gets the fix at
the chokepoint.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

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