| Commit message (Collapse) | Author | Age |
| |\
| |
| |
| |
| |
| |
| |
| |
| | |
the walker's cheaper compile
A merge of term-phase1a and pardes-next, no conflicts. core-test (the core with no terminal) 411 passed, 535 skipped of 946; shell-test (the same tests, the terminal compiled in) 989/989, on tui.
Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
|
| | |\ \
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
terminal behind src/Term.zig
A merge of pardes-next, plumber and walk-fix. Only the changelog conflicted: both sides' entries kept, and the first plumber entry's "A look does not ask it yet" dropped, a look asking it now. core-test 990/990 on tui; behaviour-test 269/273 (the four addon ABI tests the base fails); the look, diff, pager and tree process cases pass, goldens unchanged; replicas pass.
Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
|
| | | | |
| | | |
| | | |
| | | | |
names has a file, plumb/<port> (0444), listed after rules and send in the order rules named them and never dropped (Plumb.ports, dumped: seeded at boot from the rules in force, grown by each apply, refused past 32), its node a third family in the tree beside panes and columns (Node.ofPort, Target.port; port_bit is 1 << 40, clear of the host tree's and the archive's high bits, which 1 << 57 hit); an open reads (Open.what.port, an events.Queue of its own, 64 KiB): the messages sent from then on, packed, one a read, a short read the front and the next the rest, a read with none held until one comes (p.fs.news, answered by the listener as event reads are); every open gets its copy (fanOut, plumber(4)'s queuesend), one over the queue is not sent to it and the log says so; a port pardes reads itself (edit, web, a loaded kind) hands its readers copies without asking anyone, so a look stays speculated whole; another port's readers are the authority's to know, asked through tape.authority (a replica's own look waits for the first answer and guesses the last), with none its rule set's start runs, or its client, the message held in p.fs.plumb_held for the port's first opener (Plumb.opened, plumber(4)'s queueheld); test: plumber.zig, two opens of plumb/edit each read a look's message while the file opens, an open made after reads nothing, a client's message reaches the first opener and the next goes to the reader with nothing started, a port is not written; monkey9p.zig gains the plumb/ files, rules and messages that start no program, the ports' reads as held reads, and plumb/rules' rest read at the close (whose refusal is the log's alone, as a line file's: restAtClose); replicas.zig: a look to a port a program reads, the second of a replica's own speculated whole and confirmed, digests equal; tree.zig's readdir test lists web and edit; the reference's tree and plumber section, fs-help and the changelog say so
|
| | | |/
| | |
| | |
| | | |
after the pane's own readings and once a name already open in a pane is focused (never sent), lookAt sends the trimmed name, type text, src pardes (term from a raw terminal, a kind's name from its pane), wdir the pane's directory, with what resolveLookTarget found of it (Plumb.Seen: isfile and isdir take it, so no second stat and Look's own readings of a URL or `//` stay no file); a message the rules give back to `edit` as sent is Look's own (Sent.look) and runs lookAt's switch on that target, one nothing takes or a started word declines is Look's own too, whose `.url` arm is now the word's search; the edit port's open of another message runs with look_text, which now means "unplumbed", so the edit port's lookAt never sends again (the recursion trap); the three diff hooks become the default rules of the record's §5.2 (a bare .diff/.patch to `Diff $file`, a commit hash in the editor to `Diff $0`, after the two edit sets), and `Diff` takes a file or a revision, a rule's Diff declining in silence (Pardes.plumb_declined) where it cannot (no hunk, not readable, no .git above): hunks.lookedAt and lookedAtCommit go, their bodies in hunks.word, the `needs` check of the bare form in the word, PaneKind.diff gone; the old one-line `Plumb` word, Core.plumb, look.Port, plumbRuleError, plumbTarget and resolve's rules block go: bare, `Plumb` opens <config>/plumbing as Config opens init (both through builtins.openConfigFile; `include basic` in a new one), an argument is refused in words, Save applies it (host_io.writeFile beside reapplyInitFile, Plumb.saved), and boot reads it (Plumb.boot in applyStartupConfig); a started program now runs detached as Plan 9's plumber's (the coordinator's decision for the user's question 4): Effect.start, every host's host_io.startDetached (double fork, setsid, /dev/null for 0-2, every other fd closed, argv exec'd in wdir with no shell, the child's errno back through a close-on-exec pipe, said as `plumb start: <prog>: <why>`), so exec_dir goes; resolveLookTarget's and Plumb.Send.search's directory walk is one look.Dirs; test: look.zig's Plumb test rewritten onto the rules (a rule's word, an open pane unsent, the hover's resolve, a URL with no web rule searched for) and a hash inside and outside a git work tree; addons.zig's Plumb test onto plumb/rules (a port no kind reads, its kind gone: Look's own); Plumb.zig: the word, the save and boot; replicas.zig gains a URL (speculated whole, confirmed, no rollback) and a hash, each the authority's and a replica's own; every look golden unchanged; the reference quotes src/plumbing whole and says the order, the changelog says so
|
| |/ /
| |
| |
| | |
src/term/term.zig roots it, importing std, ghostty-vt, wuffs, zstbi and plain, never the core, and gives the core one table, `ops` (plain.term.Ops: about 60 entries, the records they take and give and the opaque handles State, Encoder, Track and Replica, in src/plain/term.zig), its own functions put in it through a cast refused at compile time unless each parameter and result matches the table's but for the handles and every error is one the table names; src/Term.zig, the core's side, takes `ops` from the term module when the build's term_options say the core has one (every core module but core-test's), null otherwise, and Term.State is a handle with the walk's hooks, so a dump keeps a terminal as it did (its Parts by name through walk.save, then ghostty's snapshot) and the digest is the emulator's; echo.zig comes back to the core (src/echo.zig: predicted echo is the replica's presentation, its five reads of the emulator through Term.grid, modes, prompt and Term.echoCell); screen_state's Peer is a plain record both sides name (its noteSent a function of screen_state's), its Encoder, Track and Replica handles made and freed through the table (replica.Screens makes its encoder at the first encode); a prompt's cut is told in columns and measured in its row's text by the core (Term.promptPrefix, File.rawAtDisplay); TerminalImages' decode bound moves from memory.limits to the decoder it bounds; build.zig: core_lean, the core module without the terminal, is core-test's, and wireCore no longer gives any core module ghostty-vt, wuffs or zstbi (the term module has them), so core-test's compile names none of ghostty, wuffs, zstbi, simdutf, highway or -lc++ (from --verbose); term-test runs the module's own tests (the pager's, screen_state's, the luminance table's); the tests of a terminal through a core move as they were to test/terminal.zig, in pane-test, which imports both (`of` reads the emulator behind a handle; the dump round trip frees through State.dumpFree), and tests elsewhere that read the emulator do it through Term.emulatorOf, a test in core-test skipping first; core-test skips a test whose terminal Term.open refuses; Authority.init releases its tape when the core it boots fails (it leaked); tests: core-test (no terminal) 409 of 941, 532 skipped, shell-test (the same tests with the terminal) 984 of 984, pane-test 180 of 181 (1 skipped, as before), term-test 15 of 15, kitty-test 45 of 45, replicas checked/wrong 0, hxdiff 658 cases 0 mismatches, behaviour-test 268 of 272 (the base's four addon ABI tests), lib-test
|
| |\|
| |
| |
| |
| |
| |
| |
| |
| | |
plumber's first phase
A merge of term-phase0 and pardes-next. File.zig's recolorSyntax keeps the theme's grounds and names Term.Filter and Term.Span; the changelog keeps both sides' entries, the palette sentence in the place of the entry the terminal work cut down. core-test 985/985 on tui; the diff, pager and tree process cases pass, goldens unchanged.
Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
|
| | |
| |
| |
| |
| |
| |
| | |
patch into `pardes -` and snaps the +Diff it is restacked into (its hunk found two lines down where the file moved it, under the declaration it stands in, `a.zig:3` and `const S = struct {`, its label one level deep, `a.zig:5 *`; the changed word on a stronger ground in both lines, a line added whole on a faint one, the markers in their tints' hue), as text and as styles; replicas opens with a .patch looked at by the authority and as the replica's own input, the first stopped at its first read and confirmed with no rollback, the replica's digest and +Diff the authority's either way; the soak's fixture is a repository whose last commit changes code.zig, and `gitshow` pages `git show HEAD` into `pardes -` from a shell
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
|
| | |
| |
| |
| | |
terminal half is reached through the interface (Term.encode, Term.encodeRows and Term.untrack on the authority, Term.wasRead and Term.clearRead for the rows a step read, Term.apply with the prompt marks riding a delta, Term.applyRows, Term.shown and Term.blank on a replica, Term.Encoder, Track, Peer and Replica its handles), so replica.zig reaches no emulator state; its kind half (a kind's frame as a Shadow, its strips and Blobs, encodeShadow) is the core's and moves, as it is, with its tests, to Kind.zig beside the Slot it shadows; predicted echo is Term.Echo; the draw entries (Term.recolor, Term.images, Term.petscii, Term.togglePetscii, Term.hasImages, Term.Filter, Term.FilterPalette, Term.palColor) are reached through the interface with their signatures as they were, not yet plain data (Term.zig says so); src/Term.zig is the one file outside src/term/ that imports one inside it (fs.zig embeds terminal.zig's text as a served source), panes.terminal, pardes.screen_state and pardes.echo go, the comments that named ghostty for the emulator say the emulator, and the tests that read emulator state read the records but three (Core.zig's palette, pty.zig's output pin and page limit); the changelog says so; nothing else changes
|
| | |
| |
| |
| | |
program's keyboard protocol and modes ask, Term.mouse and Term.wheel a mouse event's and a wheel turn's, Term.paste a paste as xterm makes one, Term.modes is the plain record of the modes the core decides by (mouse, all motion, the wheel, focus, bracketed paste, the alternate screen) and Term.promptMoves how far a click on the prompt line moves the shell's cursor; the mouse's records (MouseMods, MouseButton, MouseAction) are the interface's; keys.forwardKey (the bytes queued, an error reported) and keys.enterTty (raw mode, the cursor handed to the shell) are the core's, in keys.zig, pardes.keys for the tests; bracketedPaste, reportsMouse, reportsFocus, ownsWheel, tracksAllMotion and onAlternateScreen go into the record; the tests that called them follow; nothing else changes
|
| |/
|
|
| |
screen_state.zig, echo.zig, TerminalImages.zig), as they are: names and signatures kept, only the import paths change (theirs to ../, pardes.zig's, panes.zig's, tape.zig's, replica.zig's and Kind.zig's to term/), the comments that name them by path follow, and the embedded sources serve terminal.zig at src/term/terminal.zig; the first step of docs/next-terminal.md's Phase 0, a move, nothing else
|
| |
|
|
|
|
|
| |
(the ~42 fields views.fields named: size, cell metrics, columns, rects, focus, jumps, pointer, gestures, prefixes, frontend), one of Pardes.views[4] (null when closed), the live one views[live] through Pardes.view() (index field `view` renamed `live`); going into another view is the index plus the panes' shares, which stay swapped (park and unpark now move only Pane.pv); views.zig loses fields, Part over Pardes, Parked, kept, emptied and the session half of every copy, and a pane closing leaves every view's layout at once (views.gone, at the two places a slot is freed) instead of the serials each parked view kept; drawOne and within put the drawn view's own share back whole; the views are in the digest now (tape's per-field parts name views[k].field), so a fresh view's col_x/col_w default to zeros, not undefined; the dump keeps every open view in the core and the panes' shares of the others under `shares`; an image from before (live view as the core's own fields, the rest under `views`) loads them into views[] once; the ~2,400 references rewritten mechanically from the compiler's errors (p.active is p.view().active, `const v = p.view()` hoisted in 823 places), Kind.zig (4 lines) and addon_test.zig (3) among them, macos.zig by pattern (not compiled here); tape's watched host writes read the live view's frontend and active; perf.zig's accessor lines change its harness hash, the debug baselines re-keyed to it (their numbers the base's); no golden changes; gates: core-test 956/956 both, unit-test 1487/1488 tui and 1530/1531 gui, behaviour-test 248/248 both, pane-test 171/172, snap all 105 ok, process-test 0 of 37 failed tui, replicas ok
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
|
| |
|
|
|
|
|
| |
architecture §16.9 has it (pa_window 24 bytes and no art bytes, pa_placement 80 with at, from and dirty_y/dirty_h and no stride, pa_word 32 with no size and PA_WORD, tick(st, now_ns, flags) with PA_FOCUSED, PA_OK and the conventions, the dump and load rules, pa_put with attrs, pa_copy and PA_POD, whose fragment gcc -Werror refused as written and is ternaries here) and fake.c and image's table speak it; tick is called at its deadline and once after open, a taken key or click, a word, a job's answer, a load and a change of PA_FOCUSED (the pane has the keyboard in some view, its window focused), nextWake folds the deadlines, and a Debug build hashes the kind's allocations across paint and panics naming it; paint runs once a frame, the slot keeps its prefix and whether it placed a picture, the tag is that prefix, the core's art words when it did and the path, and a frame that changes it owes a step to lay the tag out again; Addon load and reload are an addon effect the host performs (memfd, BLAKE3, dlopen, the head checks and init, the bytes a process holds shared and counted) answered by an addon event carrying what the library brings as data, whose update does the collisions, the reload's validate-then-swap at a new generation and the waiting panes deterministically, while unload is the update's alone and lets the library go by effect; a kind's job is Effect.job run on the host's workers (host_io.KindJob: tui and gui on Io.concurrent, macOS and tests on the loop) answered by Event.job, stale generations dropped and a library closed after its last job; each core has its own Registry (16 kinds, their words, leaders and extensions, digested and dumped; their code a local side array a replica has none of), so a test session no longer unloads the world and Restore and -l keep the process's kinds; every call into a kind whose answer the core acts on is taped as an Outcome (its answer, refusal, message, Look, text pane) and the prefix a step reads is taped too, so a replica's core, which runs no kind's code, steps alike; a pane of a kind on a replica is a Shadow drawn from the authority's frames, painted for the peer's view and sent by Screens as row hashes with the rows that changed and each raster once a key (screen_state.Shadow), its keys going to the authority; the tail, reload and the Addon word and event are Addon.zig's, the slot's record dump.zig's, a tail word's run exec.zig's; --addon-test gains focus, blur, dump, body, ctl and the window line, its jobs done as a host does them; fake.out gains its window line, loses a body row once the tag carries the art words and lands FakeJob at once; tests rewritten: the image dump test (the replica-style load keeps the record), the Petscii/Palette/Ascii tag test in ctl.zig and test/panes.zig's image tag test (a picture decoded and shown first), fs_namespace's decode test (the job performed as a host would), Kind.zig's registry test and Addon.zig's crash-mark test, the addons suite's ABI numbers; new: a job under way across a reload (T1), a Shadow's frame codec, and a replica with no code for a kind showing its pane and passing its keys; gates: core-test 958/958 and behaviour-test 249/249 on both, unit-test 1490/1491 tui and 1533/1534 gui, pane-test 171/172, snap all 105 ok, process-test 0 of 37 failed on both, addon-test ok, hxdiff 19 and hxparity 6 mismatches by the base's names, hxgolf 0, replicas, gui-golden, monkey-gui, daemon.py main and window-clients, soak local pass, perf medium within 1% of the base
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
|
| |
|
|
|
|
|
| |
rows, an undo entry's bytes) is taken as its length, never its bytes, a text's revision saying what it holds, where it was hashed once and kept as a word by its address (walk.zig's blob words, from when no text was written in place): since the journal a session's edit writes its text in place, and since the change before a replica's does too, so the word went at every edit and the next unit's digest hashed the whole text again, 100 MB of it in a big file, on the session's step and the replica's; the words go with what kept them (walk.forget, wordOf, forgetAll, unhashed, makeWords, the bench's blob_words_on, memory.freeRaw's and unpin's forgets, File.edit's and grow's), and so does the join's hashing thread (replica.Peers' held, texts, hasher and rounds, joins, hashTexts, unhashedDone, the `hashed` note), which made the words of the texts a session never digested before a join's digest took them, its pins with it: a client is joined at the boundary its connection comes at; the semantics change, so the wire's version is 20 (a client of the build before is refused at its hello, as any other build's, and test/daemon.py's own hello says 20); a blob under 4 KiB is hashed whole as before, so a small text's bytes stay in the digest, and the replicas harness (test/replicas.zig) compares every file's text byte for byte at quiescence beside the digest; tests: walk.zig's 'a blob is hashed once and folded in as a word; one written in place is not its word, which a Debug build checks; a freed one's word goes' becomes 'a blob of 4 KiB and up is digested by its length, what it holds told by its keeper's revision; a smaller one by its bytes'
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
|
| |
|
|
|
|
|
| |
floods (PARDES_SCREEN_STATE=<dir>, which nothing sets) are `zig build replicas -- --screen-state <dir>`, replica.zig's big-file digest bench (PARDES_DIGEST_BENCH) is the last of rollback-bench, and Store's timings (STORE_TIMES) go, no bench step taking a store
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
|
| |
|
|
|
|
|
| |
--tui with --tty kept as its old spelling, the process tests' backend .tui and --backend=tui, docs/typ/builtins-tui.json, the Settings report's platform), and the builtin Tty is Term (its registry entry, SPC n t, every default tag, the docs, tests and snap scripts), Tty still run as Term through exec.acme_words, Tty+fish too on a tag and a ctl; tty as a pane's raw mode and as the OS's terminal unchanged
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
|
| |
|
|
|
|
| |
step's unit, not inside it (resize_to, made by Peers.unit, null from `create` on): a client typing at another size than the last typist read history rows its replay held as holes, and the rows the session sent for them came in the new grid's numbering, which the replica cannot take, so its replay read blank text (the soak's local80 cell: jumps, njumps, jcur, pane 1.body at a b; fixer 2's H18, pane N.ovl); test/replicas.zig sizes them after its fill too; an in-process test reads the history in the step that resizes, with the session's order of flush, fill and unit, and another makes a terminal from 0xAA memory and finds no size waiting
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
| |
|
|
|
|
| |
and Recent pickers and jump, now and then: the store's and Recent's rows, a closed pane's jumps, from replicas that rolled back over them
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
| |
|
|
|
|
| |
printable character at the cursor, Backspace, Left and Right, in tty mode on the primary screen), from an overlay of its own that no digest, dump or rollback copy sees, painted over the terminal's body at the end of renderBody (the tty's and the GUI's); judged at each delta once the authority has the key and 50 ms have passed, every prediction taken back on a mismatch; drawn at once at a prompt pardes's marks vouch for, elsewhere only once an echo was confirmed, never after a drawn misprediction until one is, so a password prompt or a program that ignores input shows nothing typed; the keys typed behind one not predicted (Tab, Enter) predicted over the screen that shows its doing; zig build replicas -- --echo types a corpus into a played shell at 30/80/207 ms: a key to its character p50 239 -> 8.9 ms and p90 248 -> 15.8 ms at 207, no wrong character drawn; docs/next-daemon.md: the decisions, the hook for the remote client, the numbers, the gates and what is not done
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
| |
keeps over it taped, the history a step reads sent before its unit, rows sent a blob once a message, and the rows cache out of the rollback copy
terminal.zig, tape.zig: ingestWithPositions tracks its rows only where an ingest happens; what the ingest told them (each tracked row's row after it and whether it was erased, a clear, whether the pages lived, a pin not made) is a taped value (Site.positions), which a screen-state replica, ingesting nothing, takes; the logic after it runs alike on both. On the stack, so the local path allocates nothing new. buildRows marks a terminal read (State.read) outside the render (terminal.drawing).
screen_state.zig, replica.zig: the authority keeps each peer's holes as its replica does (Peer.holes, from the deltas it sent), and Screens.fill sends a peer the rows it lacks of every terminal a step read, after the step and before its unit, so a replay reads what the step read (a text-mode cursor moved into history the flood scrolled past diverged). encodeRows sends each distinct blob once a message and the rows as addresses (a fill after yes is a third). The rows cache is dropped when a copy is taken and emptied when it is put back, not saved.
test/replicas.zig: Ctrl-b in the mix (a shell out of tty mode: text, cursors and selections over it), and a shell's session (prompts, titles, directories, clears) between the TUI and the flood. 4 seeds x 120 inputs x 3 clients at 30, 80, 207 ms: no divergence, every replica and terminal the authority's, 4,653 deltas checked.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
| |
|
|
|
|
|
|
|
|
| |
trimmed by the thousand (a yes delta applies in 97 us, was 2.4 ms), holes blanked when they come into view, and input into a terminal left to the authority
screen_state.zig: when a delta's scroll is more than the replica's screen and history hold, nothing it holds is in the new state; its rows stay where they are, renumbered, the active ones and the tail written from the delta, the rest holes, and only as many more rows grown as the authority's history is longer. A hole has whatever that row held before, so `blank` clears the holes in view before they are asked for. The bench's floods run under the C allocator, as a replica does.
replica.zig: a replica's input whose live pane is a terminal is not speculated (its step reads the emulator, the last delta's here and the pty's newer one there): it waits for the authority, as a key in tty mode already did at its pty write. ReleaseFast, 200x60 yes at 16 MB/s: apply p50 97 us, p99 284 us; encode a frame 41 us at one peer and at three; 64 KB/s a peer.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
several in flight against the version last sent, its history the authority's to the row, and the authority's own steps batched into a unit a frame
screen_state.zig: each state the authority sends is a version of its terminal (Track), made once for every peer; a peer's delta is against the version it was last sent (the stream is ordered and loses nothing), up to 32 in flight, a peer at the bound waiting for an acknowledgement and then getting the newest; peers at one base are sent the same bytes. A replica refuses a delta against another base and is started afresh. It keeps the authority's history extent to the row (no byte limit of its own, trimmed to the authority's count), takes its viewport and its cursor's semantic content, and grows a flood's rows in bulk. The soak now keeps several in flight, lands them late in order, loses one now and then, and checks the history extent and the viewport too.
replica.zig: Screens (the authority's tracks and peers) and Replica.screen/screenRequest, messages of u8 pane, u32 serial, u8 kind (a delta, rows; an ack, a row request, a reset). Deltas are applied with the replica's epoch paused, so an emulator outlives a rollback. Authority.output/pending: its own steps (a pty's output) go into one unit, sent before any other input it steps and at a frame, with one digest; under screen state a digest is one word.
tape.zig, terminal.zig: under screen state an output is recorded without its bytes, pty writes are off the tape on both sides, no digest covers an emulator, and a replica's core never ingests or resizes one (a resize's own core work, the pointer selections cleared, still runs). A terminal counts its writes (State.changes); its MarkedHandler cycle rides in each delta.
test/replicas.zig: the session's shell runs a TUI, then 8 MB of yes read 4 KB at a time, while three clients type; deltas go out before each input the authority steps and once a frame, acknowledgements and row requests come back over the links, every applied delta is checked against the authority's screen at that version and every terminal at quiescence.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
|
|
walker copy put back where it was, replicas that apply their own input at once and take the authority's tape as the truth, and a harness of typists over delayed links
src/rollback.zig: the replica core's allocator (`Epoch`) knows every allocation, and while a copy is held puts off frees of what was there before and refuses to grow it in place, so what the core held stays where it was. The copy (`Copy`) is the digest's walk saving the bytes of every allocation it reaches, field by field (the derived caches and this process's own too: a safe build's free writes over what it frees, and the frame arena points at what the epoch then frees), but the queues a step drains and the frame statistics; a file's text, its line index and its undo snapshots are pinned instead (made anew at each edit, never written; now freed raw everywhere), checked unchanged in a safe build. Put back, each allocation's bytes go back over it, the epoch's allocations are freed and its put-off frees forgotten; let go, they are made. Tree-sitter's trees, the C heap's, are held the same way (syntax.held_trees).
src/replica.zig: an `Authority` records its core's every input to a tape in memory and hands each client's input back as a unit (origin, sequence number, its tape); a `Replica` applies its own input at once, as the authority will (update, then pump), recorded by its own tape (tape.zig's speculation mode), and for each unit in order drops its pending head when the authority recorded the same (inputs, host calls and answers: the fast path), else puts the copy back, replays the units since and this one (every digest checked) and applies its pending inputs again. A host call or a syscall stops a speculation: it is answered harmlessly, undone when it returns, and the input waits for the authority's answer; a clock or environment read is a guess the comparison checks; an emulator is the authority's (terminal.zig's writers stop a speculation; a safe build asserts none changed). The view's events (a resize, the pointer) wait. Messages are bytes: no transport yet.
Each client has a view of its own (src/views.zig): the authority goes into it before the client's input (`toView`, taped), the replica's speculation the same, and the copy walks the views not live too (`parked`, each pane's `pv`).
test/replicas.zig (`zig build replicas`): one authority and N replicas, a view each (`--shared` puts them in one), each client typing monkey.zig's mix at 50-200 ms over its own link (half the RTT plus 10% jitter, in order), the authority's pty echoing 2 ms later; gates: no replica diverges, every replica's digest is the authority's at quiescence; reported per RTT: inputs applied at once, stopped, waiting, fast paths, rollbacks (mispredicted, foreign), and what a speculative apply, a copy and a rollback cost. 3 clients x 100 inputs at 30, 80 and 207 ms: no divergence over 12 Debug seeds and 20 ReleaseFast ones with a filled heap; ReleaseFast p99 a speculative apply under 0.2 ms, a rollback under 1.6 ms. `zig build rollback-bench` times the real copy now (take, release, an edit, restore).
What the replicas found, fixed: a release build reads what the heap had where Debug's fill made two cores agree: `col_panes` and `restore_buf` digested undefined (zeroed now), `terminal.State.reply_dropped` never set at create, `Text.sels` and `PendingPipe.ranges` digested past their length (a walker `len` cuts a slice too). PARDES_REPLICA_FILL fills a replica's allocations to show what is left.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
|