Iris, on the two findings from the incremental-text investigation:
"let's switch to new lines for the test, and also let's keep the single
line around for stress + could be something to try to optimize later."
The streamed tail now takes a blank line every 4-12 deltas, so it is 53
markdown blocks with a longest of 502 characters instead of one block of
14,888 -- against a measured p50 of 147 and a largest-ever 1,580 over
7,706 blocks of real assistant messages. Layer 1's streaming frame went
from p50 3.86ms / p90 8.65ms / worst 10.95ms to p50 2.20 / p90 5.90 /
worst 8.78.
The run-on message is kept as the first two backlog events, 14,824
characters in one block, just under text_cap's 16 KiB so it draws in
full. The *streaming* pathology stays in frame_profile.rs rather than the
fixture: it needs a growing block, and iterating on it there costs a
second instead of a two-minute phone run.
Adding it is purely additive -- the random state is saved and restored
around those two events, so every other backlog event is byte-identical.
That is not tidiness: the first attempt shifted the backlog and broke
`a_long_press_and_drag_selects_text`, which replays a real recording at
(300, 1000) and needs the content it was recorded against to still be
there. BACKLOG_COUNT is 3202 now, in generate.py, fixture.rs and
BenchFixture.kt, which split the file by line index.
And the answer to Iris's question, which the code already had: the newest
message does *not* cap. `build_row`'s `cap` is false for the live tail
because a row that grew while capped would appear to stop growing, and a
reply growing past the cap is never caught either since it grows through
apply_delta. So a streamed block's shaping cost has no ceiling -- ~29ms
per delta at 50k characters, ~58ms at 100k.
Recorded but not chased: the emulator's `stream: build p50` did not move
(10.4 -> 10.5ms) while layer 1's frame nearly halved, so most of a
streaming frame on a GPU path is the whole-arena primitive re-upload
layer 1 never performs -- 11,568 primitives rewritten per delta, with the
fling phase as the control at 0.4ms for the same primitives moved
through move_offsets.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Iris: "remove both decisions and iris.md. I've decided to instead make
decisions when planning with agents rather than after they do things, and
they're both too long for me to wanna read, + don't cover all the
decisions I'll wanna make about the code anyways. I'll just naturally run
into things for now. Todo is important though."
So docs/DECISIONS.md (850 lines) and docs/IRIS.md (1,986) are gone, and
AGENTS.md now says not to start another: raise a choice while planning it
with her, otherwise decide it and put the reasoning at the code it
governs. The TODO lists stay. docs/SUBAGENTS_DECISIONS.md went with them
-- same artefact, same reasoning, and she did not name it, so its six
decisions were folded into docs/SUBAGENTS.md rather than deleted.
Deleting the logs left ~30 citations dangling in code comments and docs.
Each states its reason inline and cited the file only for provenance, so
they now read "decided 2026-09-07" or name the module doc that carries
the reasoning.
The root had six things that were not a program or a document. Moved,
per "I only meant top level sh files":
run-tests.sh, test-wg-tunnel.sh, wg-setup-host.sh -> scripts/
rigs/ -> scripts/rigs/
xtask/ -> scripts/xtask/
A project's own scripts stayed with the project: app/*.sh, app-rust/*.sh,
iris/*.sh and server/enroll-link.sh did not move.
`target/` at the root is deleted and cannot come back: there was never a
workspace there, and the 29 MB was only xtask's scratch space, now in
scripts/xtask/target/. `cargo xtask apk` still runs from the repo root
and now publishes to scripts/build/outputs/apk/<mode>/ -- one directory
deep, because that is what Dev Updater's `*/build/outputs/apk/*/*.apk`
discovery pattern needs, and scripts/xtask/build would have been two.
Verified: ./scripts/run-tests.sh and `cd iris && cargo test` green, clippy
and fmt clean everywhere, `cargo xtask apk debug --abi x86_64` builds and
signs an APK carrying lib/x86_64/libai_app.so at the new publish path, and
the repo root is now eleven entries with no build output among them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Iris: "the organization of the rust rewrite is a mess right now... there
shouldn't be anything related to the app inside of iris. Iris is supposed
to be the UI framework alone." And, on the crate count: "I'm confused why
the app only code needs more than one crate though."
Nine cargo workspaces become three, and the port's project code -- which
sat in five places, four of them inside the framework -- becomes one crate,
`ai-app`, in `app-rust/`:
client-core -> app-rust/src/client
iris/transcript-ui -> app-rust/src/ui
iris/transcript-fixture -> app-rust/src/ui/fixture.rs + tests/ + touch/
iris/desktop-app -> app-rust/src/desktop + src/bin_desktop.rs
iris/android-app -> app-rust/src/android + android-project/
android-shell -> app-rust/src/shell
iris/ keeps core, macro, the iris crate, tabs-ui and rig-input, and now
mentions no session, transcript, setup or server anywhere.
Only two of the old splits had a reason that survived reading. event-model
stays a crate at the repo root because server/ depends on it too, so a
crate is what makes the backend and the app agree by construction. The two
Android .so names looked like a hard constraint -- a package produces one
library artifact -- until P2 turned out to already plan merging those two
Android apps into one; both faces now come out of libai_app.so, picked
apart by features so `--no-default-features --features shell` keeps wgpu,
parley and iris out of the Compose app's APK. docs/RUST.md's "One app
crate" has the rest, including what each remaining feature is for.
DECISIONS.md and SUBAGENTS.md move into docs/ with everything else.
Verified: ./run-tests.sh and `cd iris && cargo test` green, clippy and fmt
clean in all five workspaces, `cargo ndk -t x86_64` links libai_app.so,
build-apk.sh produces an APK that installs and launches on this checkout's
emulator (Gl ... virgl, as expected), and the phone-sized headless
screenshot renders the transcript unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four things Iris asked for on 2026-09-08.
**A LazySpan no longer cares about masks.** It asserted that something
around it had called `.masked()` and refused to draw otherwise, which is
why a plain full-screen list -- the benchmark, any simple app -- panicked.
It cared only because it draws a row straddling an edge in full and relied
on somebody else to cut off the overhang; it clips itself to the box it
was offered now. Strictly stronger than the assert, which a mask *larger*
than the list's box satisfied while letting the overhang through anyway --
the fault it was written for. The transcript's `.masked()` wrapper goes
with it, and `Painter::is_masked` with that.
**Everything on the transcript screen is capped.** One rule in one place,
`client_core::text_cap`, mirrored as `TextCap.kt` with the same numbers so
a bench comparing the apps compares renderers rather than policies:
a tool call's input 80 lines or 4 KiB -> "Show all N lines"
a tool call's output 80 lines or 4 KiB -> (already was, in iris)
a message 200 lines or 16 KiB -> "Show all N lines"
The input is what the edit-card report needed: an Edit's old_string and
new_string arrive whole and are routinely the biggest text on screen.
Messages are capped in both apps, user and agent alike.
Three rules that took a screenshot to get right. A message is cut on a
block boundary, never mid-block -- cut to its own opening line a fence
renders as an empty panel, which reads as a fault rather than as a cap --
except a message that is one enormous block, which is truncated, since
dropping it would leave the row blank. A reply still streaming is never
capped. And the input's two blocks share one "Show all", while input and
output have their own.
**Compose stops wrapping raw text**, per Iris's call: a tool's leftover
input fields and its output pan sideways like the command already did.
`on_tap` and hold-the-edge move to `transcript-ui/src/tap.rs`, since a
message's "Show all" needs exactly what a tool card's tap already had.
The APK is cross-compiled here and run against the server on the host, so
everything build.rs baked in (AI_APP_TRANSCRIPT_HOST/_PORT/_TOKEN and this
machine's CA) was good for exactly the pair that built it -- and a token in
a delivered artifact besides. MainActivity registers aiapp://enroll, hands
the URI and the app's private files directory to Rust, and
client_core::config stores it 0600; transcript_client reads it afresh per
transport, so opening a new link repoints a running app.
Diagnostics says which of three things is true, because they want different
actions: 'enrolled: host:port', 'not enrolled -- open the enrol link from
Dev Updater', and 'enrolment unreadable: ...' for the case nothing could be
found out. The last is why status() has an Unknown arm at all.
ui-sandbox.sh's printed enrol command now carries the CA, which is what
makes it work for an app with no baked copy.
Verified on this checkout's emulator: fresh install reads 'not enrolled',
the intent enrols (log: 'enrolled with 10.0.2.2:8519', enrollment.json
-rw-------), Diagnostics then reads 'enrolled: 10.0.2.2:8519', and the CA
reconstructed from that link is byte-identical to the machine's ca.pem and
validates the server over curl. Android offered the chooser between this
app and the Compose one, which is the intended behaviour.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Iris's ask after using the Compose bench build on her phone: the old
scroll phase used animateScrollBy, which can only ever cover the fixed
distance/time it's given, so it never flings the way a real fast swipe
does. BenchRun.run now has four phases: fling (8 flings out + 8 back
through the list's own FlingBehavior at 12,000px/s), stream (unchanged),
type (600 fixed characters into the real composer TextFieldValue, then
deleted, to exercise wrapping and the transcript being pushed upward),
and keyboard (five show/hide cycles via WindowInsetsControllerCompat,
each confirmed by isImeVisible rather than assumed).
FrameStats.markPhase/phaseLines slice the same FrameMetrics recording
by phase rather than running a second recorder; debugReport gains a
phaseFrames section ahead of the existing whole-run frames/accounting/
work sections, which are otherwise unchanged.
Also fixes a pre-existing, unrelated break in MainActivity.kt's
benchSessionSummary() -- missing several SessionSummary constructor
arguments from an earlier change -- since it blocked compileBenchKotlin
outright.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
BenchFixture.kt/BenchNetwork.kt fake the backend for the bench build: a
URLStreamHandlerFactory installed only under BuildConfig.FIXTURE_MODE
answers TranscriptSource/EventStream's requests from an in-memory copy of
the bundled fixture instead of opening a socket, so the fold, the paging
and uniqueItems under test are the screen's real ones rather than a
shortcut built for this. MainActivity opens straight onto that session
when FIXTURE_MODE is set, with no enrollment and no permission prompts.
BenchRun.kt drives the same scroll loop and streaming phase
transcript-bench.sh/stream-bench.sh drive over ui-trace, but in-process
(24 swipes through the real LazyListState, then 400 fixture events
appended at 20/s through the real live-fold path), and adds process CPU
time, peak RSS and battery current to the render report -- "unavailable"
rather than a fabricated number where the device can't answer.
"Run benchmark" sits beside the existing "Copy" in session settings,
found by that exact label the way every other control here is
(SessionSettingsDialog's onRunBenchmark, null on every build but bench).
debugReport gained an optional `extra` section for this; empty and
invisible on every other build's report.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Own application id (.bench suffix) and label ("AI Sessions bench" via a
build-type resValue over the new @string/app_name), release
optimisations, signed with the same key build-apk.sh already generates,
FIXTURE_MODE=true wired through BuildConfig. Its asset source set points
straight at app/bench-fixture/assets rather than a copy under androidApp,
so there is one file to keep in sync with the generator, not two.
build-apk.sh bench builds it; the CA-pinning step is untouched and still
requires a real ca.pem to exist, even though this build never connects --
simplest to let it pin whatever is there rather than special-casing it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Deterministic (seeded), in the app's own event model rather than a real
transcript: 3,601 events split into a 3,200-event opening backlog and a
400-event tail both bench harnesses replay as the streaming phase, with
headings, inline markdown, fenced code in six languages, a table, tool
calls with kilobyte-scale input/output, and two embedded PNGs.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Splits each frame sample at queue.submit into redraw-to-submit (iris's own
CPU work) and submit-to-after-present (driver/GPU wait), so RUST.md's I5
"where does iris's frame time go" question can be answered with a number
per half instead of a single total. Adds a force-gles Cargo feature that
switches the Android wgpu::Instance from Backends::PRIMARY to Backends::GL
at compile time (no runtime env-var path exists into an already-launched
Android process on this machine), for isolating SwiftShader-Vulkan vs.
GLES/virgl as the software-mode gap's cause. app/iris-scroll.sh extracts
transcript-bench.sh's exact 24-swipe/6-cycle gesture loop for iris's own
demo app, which transcript-bench.sh cannot drive directly since it opens a
session through the Compose app's own UI.
Verification (this pass, on a disk-pressure-limited host running low on
space): cargo fmt --all clean, no diff. cargo clippy --workspace
--all-targets: no warnings from this diff (pre-existing future-incompat
notices from wgpu/winit/naga only). cargo test --workspace and cargo ndk
for iris-android-app --features transcript-screen were verified clean by
the previous pass on this identical diff (fmt/clippy/test/ndk all clean,
per that pass's own report); not re-run here because the host's disk was
93% full and a concurrent ai-server rebuild (stable toolchain moved to
1.98.1, rebuilding aws-lc-sys from scratch) had driven I/O pressure to
~60%, so a repeat cargo test --workspace sat 50+ minutes doing no useful
work and was stopped rather than left to make the disk situation worse.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The Agent tool runs subagents in the background, so the parent's result
arrives at launch while the subagent works on for minutes; finishing on it
read a running agent as finished with a transcript cut off at launch. A
subagent now ends on its own message_delta end_turn, and a later line for a
finished one reopens it, since a background agent can be messaged again.
The card's expander row was only the chevron's height, so a tap for it
landed on the first subcard; it is the platform's 48dp minimum now.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A subagent is a second transcript owned by a session, in the same event
model, with no process and no controls. The claude translator routes lines
carrying parent_tool_use_id to a per-subagent translator and transcript
under <session>/subagents/<tool_use_id>; three routes expose the list, a
transcript page and the SSE stream. Echo grows /subagent [n] as the rig.
On the phone a card with subagents ends in a chevron expander, collapsed by
default, opening to outlined subcards styled like dev-updater's components;
a subcard opens SessionScreen in read-only form, addressed through
TranscriptAddress so paging, cache and stream are shared.
Design in SUBAGENTS.md; choices awaiting review in DECISIONS.md.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New xtask/ crate (no deps) runs cargo ndk -> javac -> d8 -> aapt2 ->
zipalign -> apksigner directly, signed with the same key build-apk.sh
uses. Both pass conditions proved on the ai-app-2 emulator: the xtask
APK installs over the Gradle-built shellApp, and the notification
service starts and posts a real notification while backgrounded.
Adds one printRuntimeClasspathJars task to shellApp/build.gradle.kts
(and a matching signingConfig) -- the one disclosed Gradle call the
xtask still makes, to resolve the AndroidX/:link dependency graph.
That call's Kotlin compilation of :link as a side effect also answers
E3's open kotlinc question, so no Java port of ServerStore was needed.
Wires a second Apk component into .dev-updater.ron beside the existing
one. Full writeup in RUST.md's E5 box.
Co-Authored-By: Claude Sonnet <noreply@anthropic.com>
Two Java classes (MainActivity, NotificationService) hand their lifecycle
to a new android-shell crate built on client-core; client-core gains
notifications.rs (the /notifications SSE parse and attention_line, ported
from Notifications.kt). Packaged as a new app/shellApp Gradle module
rather than a rewrite of app/androidApp in place, so that module's working
Compose UI is untouched.
Both pass conditions held on the emulator: a notification arrived in
Android's drawer with the app closed, and a shared text share landed as a
real message in a sandbox session's transcript. Found and fixed three
real bugs along the way (a silently-wrong JNI signature from a generic
JObject parameter, a class-by-name lookup failing on this crate's own
background thread for lack of an app ClassLoader, and onStartCommand
opening two /notifications connections per enrollment -- the last a
latent bug in Notifications.kt itself). Full account, exact commands and
what was deliberately cut are in RUST.md's E3 box.
Co-Authored-By: Claude Sonnet <noreply@anthropic.com>
Off by default and per session: it spends quota the moment quota exists,
with nobody watching, which is not a thing a default may decide. Switched
on from the session settings dialog, with the message it sends editable
("continue" unless something else is typed).
Running out of quota becomes a state rather than an error. The Claude
driver recognises its dialect's sentence -- `Claude AI usage limit
reached|1788546972` -- and reports `LimitReached` with the reset time it
gave; nothing above a driver matches on a string. The transcript draws it
as a divider, like a clear or a compaction.
The schedule is a plan to *ask*, never a plan to send. Both reset times
available are untrustworthy in the direction that matters -- the dialect's
is written when the turn fails, the endpoint's moves when the window does
-- so the wait ends in a question to the usage meter, and only `ok` with
no window at 100% sends anything. A window still spent reschedules to its
own reset time, which is what makes a limit that lifts late wait longer
and one that lifts early resume sooner. A meter that cannot be asked is a
longer wait too, never a send. A day after the limit was hit the wait
gives up and says so in the transcript, so a machine that can never be
asked is not retried for ever.
The schedule is persisted on the session: a five-hour window outlasts a
backend restart, and a wait forgotten across one never comes back.
Driven end to end with echo, never a real account: `/limit [minutes]`
reports the same event a real driver does and `/usage` sets what the meter
answers, deliberately separate so the two can disagree. The wait moved
from the dialect's two minutes to the meter's seven when the meter changed
its mind, and the message went out on the first check after the meter came
back under the limit.
Also makes the settings dialog scrollable, which these two controls made
necessary: at a 1.5x system font it clipped the last of them with nothing
on screen to say so.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`Config::default_effort` is what a session starts at when nothing chose one,
applied in `spawn_session` rather than filled in by the spawn screen so it
holds for an import and a bare API call too. It is set by the spawn screen's
own picker, whose label says so: one control, where new sessions are made,
rather than a settings page for a single value. Not on a provider, because
providers are discovered and the next rediscovery would erase it; not on the
phone, because a second device would then spawn at a level nobody there
chose. `GET`/`POST /defaults` carry it as a struct, so the permission mode --
still hardcoded to `auto` on the spawn screen -- can move there later without
a second route.
Only drivers that read a level are given one: an echo session was storing a
`--effort` it never passes to anything, which is a config file answering a
question about itself wrongly.
Separately, `AGENTS.md` is 35 KB sent with every request in this repo, and 12
KB of it was rigs and reference measurements that only matter once you are
running one. Those are the `ai-app-rigs` skill now -- the same text, still the
only copy, read when the work touches it. 35,198 -> 20,813 chars.
Verified on the emulator against the sandbox: the spawn screen pre-fills from
the server, picking `low` spawned a session at `low` and left `/defaults` set
to it, and an echo session spawned afterwards took no level at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Output is about an eighth of what a session costs and thinking is nearly
all of it -- prose is ~1.5% of output tokens, measured over 27,015 requests
of this account's own transcripts -- so the level is the largest saving
available short of shortening the conversation itself.
Shaped like the working directory rather than like the model: the CLI's
only two setting control requests are `set_model` and `set_permission_mode`
(checked against the 2.1.258 binary), so `--effort` is read when the process
launches and cannot be asked of a running one. `set_session_effort` records
the level and stops the process; the next message or Start launches one that
has it. That is also why the picker is in the session settings dialog beside
Move, and not on the bar beside the model and the mode, which take effect
mid-turn.
`None` is a level in its own right -- the CLI's own default -- so the picker
can return to it, and a blank is normalized to it at the boundary rather
than stored as a level the CLI would reject.
Offered only where it means something: `DriverKind::takes_effort` reports
the capability and the phone leaves the row out entirely, rather than the
session-type branch this app does not have anywhere else. A llama session
would otherwise get a control whose only effect is stopping its process.
Verified on the emulator against the sandbox's fake CLI: the picker sets it,
the server reports it, and an echo session's dialog is unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The rate-limit bar answered a question about an account, and picked the
answer by machine. One machine runs echo, the Claude CLI and a local
model side by side, so every echo session on it drew the CLI's five-hour
window: a quota that session cannot spend and could never run down. A
session now names its meter (`usageProvider`, from
`DriverKind::usage_provider`, which `usage::providers_for` reads too so
the two lists cannot disagree), and the phone matches on machine *and*
provider. Nothing meters echo or llama, and nothing at all is drawn --
including while the first fetch is out, since "checking" under a session
that turns out to meter nothing is a row the screen then withdraws.
Echo gets a meter it can be *told* about instead: `/usage 42`,
`/usage 95 20`, `/usage 42 never`, `/usage notloggedin`,
`/usage unreachable`, `/usage failed`, `/usage off`. Those states cost
real quota to arrange, which is why none of them had been looked at.
And llama.cpp runs wherever a setup says, which was the last of phase 5.
`Transport::reserve_port` is the second half of what a transport is --
"run this" plus "reach this port" -- returning the port the server binds
there and the port that reaches it here, and `Launch::reaching` puts the
`-L` tunnel on the connection that already carries the command. Three
things that came out of building it:
- A forwarded launch gets a pty and every other one keeps `-T`. Killing
the ssh client ends a CLI by closing the stdin it reads; llama-server
never reads its stdin, so the same kill left it running on the far
machine with the model loaded -- one orphan per stopped session.
- The model is looked for on the machine that will serve it, at that
machine's own models directory, so `GET /setups/{id}/models` is what
the spawn screen offers rather than the backend's own downloads.
- The readiness poll watches the process, not only the port: a model
that will not load exits in a second and would otherwise have been
reported as "gave up after 300s". The failure carries the log's tail.
Exercised end to end against this VM over ssh to itself: spawn, load,
answer, outlive a backend restart, be adopted, answer again, and stop --
with both the ssh client and the far llama-server gone afterwards. The
local path, the Claude bar and the spawn screen checked on the emulator.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The same pass the server had, on the Kotlin side: comments restating what
the code says are gone, and the ones recording a measurement, a constraint
or an incident are kept but cut to a few lines each. 6540 comment lines to
5674, and 920 lines off the app.
Two doc comments had drifted onto the item above the one they describe --
`contextAfter`'s onto `sessionWorking` in Events.kt, and `UsageMonitor`'s
equivalent on the server was fixed in the previous commit. Each is back on
its own item, which is the only non-comment line this diff moves.
The comments are reflowed to the column limit at their own indentation:
several were written wide, and ktfmt re-wrapped them into lines holding a
single orphan word. `/tmp` script, not kept -- ktfmt is idempotent over the
result, which is the check.
Left alone deliberately: this codebase's remaining comment density is high
because the comments carry things the code cannot say -- what a null means,
what a number was measured against, which bug a guard exists for. Of the
238 one-line doc comments in the app, five were pure restatement of the
name and were removed; the rest each say something the signature does not.
ktfmtFormat, compileDebugKotlin, lintDebug and testDebugUnitTest pass;
cargo test (127), clippy --all-targets and fmt still clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reopening a session downloaded the conversation again, every time, over
the tunnel. It now draws from a copy of what the server has already sent
and asks for one event to check that copy is still current.
Per session, under cacheDir, the server's own event lines in chunks named
for the range they cover -- so a coalesced page, whose lines do not say
what they cover, still records it. Only the contiguous run ending at the
newest chunk is served; a gap is closed by paging through it, bounded by
`after` on /transcript so the page stops where the phone's copy starts
and can therefore be kept. Nothing is derived and stored: rows are a
rendering, and a cache of them would need throwing away on every change
to the fold.
Nothing here is load-bearing. Missing, evicted, damaged or unwritable all
degrade to the cold open this screen did before, and the check before the
stream resumes -- one request, one event -- is what stops a replaced or
truncated file being spliced onto a copy of a different conversation.
What that check cannot see, a line changed mid-file with the tail intact,
is what Reload in session settings is for.
Measured on the emulator against ui-sandbox, on a 505-event session:
reopening it costs one request for one event, including scrolling the
whole conversation back; a cold open is two requests and 100 events. A
reset after falling 300 behind fetched the gap as four coalesced rows
rather than re-fetching 104 events and discarding them. Every chunk was
checked line by line against what the server says for the range its name
claims, across the reset and the gap-fill.
transcript-bench.sh, same viewport content and gestures, before and
after: p50 16.9ms both, p90 25.6 -> 23.2ms, p99 33.5 -> 36.7ms, and the
transcript's own draw accounting 0.33ms -> 0.32ms with place 0.31ms
either way. Within the emulator's noise, which is what a cache must be:
it changes what is fetched, not what is drawn.
Building it also found that the server handed out the same transcript
line two different ways. serde_json's default float parser is not
correctly rounded, so a ts written as ...0757 came back from /transcript
as ...0755 while the SSE stream sent the original -- invisible on screen,
since a ts is drawn as a relative time, and visible here only because the
cache compares a line it holds against the server's answer. Fixed with
float_roundtrip, with a test that fails the moment it is dropped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A table is recognised by its delimiter row, the only line of one that cannot
be anything else, and its header is the line above -- the single place the
scanner looks ahead. Colouring every `|` instead would have marked the pipes
of a shell command written in a paragraph.
Addresses come in two shapes: `<...>` needs a scheme's colon or an at sign
inside it and no whitespace, which leaves `<div>` alone; a bare `scheme://`
needs no closer, so where it ends is the decision -- the sentence's trailing
punctuation is given back, and so is a closing bracket unless one opened
inside the URL.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The token scanner asks what a character is; markdown's meaning is where it
sits, so a `#` opens a heading at the start of a line and is an ordinary
character three words in. `MarkdownSyntax.kt` reads structure a line at a
time and then each line's prose left to right, and `spansOf` is the one
entry point that hides which of the two scanners a language got.
Conservative wherever a guess would be invisible: emphasis needs a closer on
the same line with no space beside either marker, so the `*p = *q` of a C
fragment opens nothing; an underscore may not start or end inside a word;
and an indented code block is left plain, since four spaces after a blank
line and four after a bullet are the same line.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The stretch is rendered on the box around the list, so everything in that
box bent with it -- the numbers included. `Modifier.overscroll` wraps the
effect's single node, so one effect cannot be rendered in two places and
the numbers cannot be excluded while they live in the rows. They had to
leave.
They now sit beside the scrolling box: the rows leave a spacer where the
numbers go and a `SubcomposeLayout` draws them there. That is the one
arrangement that keeps them level with their lines, which is the thing a
numbered listing may never get wrong. Which numbers exist *and* where each
one goes both come from the list's own `layoutInfo`, read in the measure
block -- and subcomposition happens during measurement, so it composes from
the answer the list has just produced rather than from one it read a frame
ago. A column translated by the scroll position could not do that: the
translation would be a layout read and current while the set of numbers was
a composition behind it, and during a fling the numbers would slide against
their lines.
Checked by sampling the screen at about 1kHz through a fling: 23,520 row
observations over 552 frames, every one with its number at exactly its own
top, no drift at any point. Also that the gutter holds its x while the text
scrolls sideways, and that a short file and an empty one still draw.
A consequence worth having: the numbers are no longer inside the
`SelectionContainer`, so selecting part of a file and copying gives the code
rather than the code with a number in front of every line. That follows from
where they are now rather than from anything asked for, and a copy was not
exercised.
The stretch itself still cannot be seen from this VM, so whether the numbers
now stay flat while the text bends is the thing to look at on the phone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`Modifier.horizontalScroll` builds its own overscroll effect per node, so
with a node per row only the line under the finger bent when the file was
dragged past either end and the rest sat still beside it -- the same
complaint as the offsets, one layer out.
There is an overload that takes the effect instead of making one, and it
leaves the rendering to the caller. So the viewer makes one effect, hands
it to every row, and renders it once on a box around the list: the file
bends as the block it scrolls as. This only works now that every row is
the same width -- rows that disagreed about where the end was would
disagree about when to stretch.
**Not seen working.** Measured on the emulator here: over-dragging well
past the end and capturing mid-gesture produces a frame with no stretch in
it at all, and the list's own vertical overscroll does not appear either,
so this VM cannot show the effect for any scrollable. Noted in
~/.claude/MACHINE.md so the next session does not spend the same half hour
on it. What was checked here is everything either side: the scroll still
reaches both ends, the position survives scrolling vertically, and nothing
else moved. The stretch itself wants a look on the phone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sharing one `ScrollState` across the viewer's rows was not enough to make
them move together. `Modifier.horizontalScroll` is a node per row, and each
one coerces the shared offset into *its own* range -- its content width
less its viewport -- so a short line's range is zero and it stayed put
while the long line beside it moved. Each row also writes `maxValue` on the
shared state as it measures, so how far the file could be dragged at all
was decided by whichever row happened to measure last, and changed as the
list scrolled.
Both go away once every row is the same width. `FileLines` now carries the
longest line in columns, and the viewer turns that into one content width
from a single character's advance -- arithmetic rather than twenty thousand
measurements, because the face is monospace -- and gives it to every row. A
tab counts as eight columns and deliberately upwards: over-estimating
leaves a little empty space past the longest line, under-estimating puts
the end of that line out of reach. The width is capped well under what
`Constraints` can carry, so a minified file is a scroll that stops early
rather than a crash.
Reported by Iris on 2026-09-04. Checked on the emulator against the
generated 1 MB file, whose lines run from one character to sixty-eight:
the file now moves as a block, the offset survives scrolling vertically and
newly composed rows arrive at it, and the far end of the longest line is
reachable. Also checked on the two cases the change had no reason to touch
-- a file narrower than the screen, which still does not scroll at all, and
an empty one, whose zero content width draws its one numbered line.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The three file sizes the viewer's and the editor's limits were measured
against are built by ui-sandbox.sh now, beside the rest of the explorer's
fixture tree, so the figures in EXPLORER.md can be taken again instead of
each session inventing its own generated file and getting a slightly
different one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three numbers, taken on the emulator through the app's own render report
and written into EXPLORER.md; the fixture tree the sandbox now builds is
what they were taken against.
The viewer's scan was on the main thread. Decision 8 said off it, and the
first version did it in a `remember` inside the composition, which is not
that -- 460ms of frozen screen on a 1 MiB file, long enough that the
accessibility tree cannot be read, which is exactly what "the app has
stopped" looks like from outside. It runs on Dispatchers.Default now, with
a spinner where the file will be. Reading a megabyte is otherwise fine:
the viewer is a row per line, and it opens and scrolls 28,660 of them.
Edit mode needed a cap, and not the one the plan expected. The cost that
matters is not the highlighting -- 40ms a keystroke at 128 kB, which is
survivable -- it is Compose laying out one enormous text in the field:
2,027ms per frame at 128 kB, with typed characters dropped, and no
response at all at 1 MiB. Switching highlighting off would have saved
nothing, since every arrangement of a single text field pays it. So
EDIT_LIMIT is 32 kB, the largest size actually measured as usable, and
above it the pencil is disabled with the reason in words beside it: a
disabled control teaches what the thing can do but cannot say why it is
off, and a reader who cannot edit a file they can plainly read would
otherwise conclude the app is broken.
`FileLines.of` is timed like everything else here, so the figure lands in
the render report rather than needing a harness to ask for it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The drag-right-to-go-back gesture was removed on the other side while the
file explorer was being built on this one, so the explorer's own use of it
goes with it: back inside the explorer is the arrow and the platform's own
gesture, which is what the rest of the app now has too.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The other half of EXPLORER.md: a folder button on the session header opens
the machine's filesystem, starting where the session works.
It draws **over** the session in the same `Box`, so the session under it
stays composed -- its event stream keeps flowing, its draft and scroll
position stay where they were, and coming back from a file costs nothing.
Back steps one level inside it (editor, viewer, directory, parent) and only
closes from where it opened; the platform gesture, the button and the swipe
all go through the one function, so they cannot mean different things.
The viewer is a `LazyColumn` of lines rather than one `Text`, because text
layout is linear in the text and a twenty-thousand-line file in a single
`Text` measures all of it to draw a screenful. Lines do not wrap and share
one horizontal scroll, so a logical line is a visual line and the gutter
cannot come to number the wrong text; the gutter's width is measured from
the digit count of the line count in the style it is drawn in. The editor
is a `BasicTextField` with a `VisualTransformation` carrying the scanner's
spans, which is the one Compose API that colours a field's own text rather
than replacing the field.
`fileLanguage` reads the same table `fenceLanguage` does, so a language
added for fences is a language added for files.
A file that changed on the machine while it was open here refuses to be
overwritten and asks, with what each of the three answers costs. That is
the ordinary case, not the exotic one: an agent editing the file somebody
is reading is what this whole feature is for.
The speedometer moves off the header into the session settings dialog,
where the session's other about-the-session controls are, and the folder
takes a place between the usage chart and the cog -- widest scope to
narrowest, cog at the end, as Iris asked. Both benchmark scripts move onto
`ui-trace`'s new tap-by-label action in the same change, so the render
report is never unavailable and never pressed at a coordinate that has
stopped meaning anything; `app/bench-lib.sh` is what they share, and
`grep -n "tap [0-9]" app/*.sh` is the check.
Exercised on the emulator against the sandbox's new fixture tree, with a
screenshot or a ui-trace for each: the listing (dotfiles, directories
first, a symlink to a directory sorted with them, a name with a tab in it),
a highlighted file, binary, too big, a permission error, editing and
saving, the 409 and its Overwrite, back with unsaved edits, creating a name
that exists, creating one that does not and landing in the editor, an empty
directory, and `..` above the directory the session opened in.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The screen no longer follows a horizontal drag. Back is the arrow at the
top left and the platform's own edge gesture, both unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The chip was the renderer's span background, and a span's background is
part of the text's own drawing: the text node paints the selection first
and the glyphs over it, so an opaque chip covered the selection and
selecting a sentence highlighted every word of it except the ones in
backticks. The previous fix let the selection show through by taking the
chip to 60% alpha, which is a compromise on both sides -- the chip is a
weaker step down from the page, and selected it reached #3C344F where the
words around it reached #776394.
There is a place that is under both, and a fenced block was already in it:
a modifier on the text rather than a style inside it. So `appendCodeChip`
takes the code span from the renderer's inline builder, keeps its style and
its space of padding either side but drops the background, and marks the
range; `LinkedText` draws those ranges in a `drawBehind`. The chip is back
to the full `rawSurface` fill (measured #11111B against a #1E1E2E page) and
a selection over it now lands at #776394, the same as the rest of the
sentence -- the fenced block's numbers exactly.
The geometry is one box per line, from the bounding boxes of the run's
first and last characters, taken as far as the line's `visibleEnd`. Not
`getPathForRange`: that is the shape of a *selection*, which runs to the
right edge of every line but the last, and a code span that wrapped left a
full-width empty chip behind on the line above -- twice in one fixture.
`visibleEnd` is the same rule the selection rectangle obeys, so the chip
stops where the selection stops instead of sticking its padding space out
past the end of a selected line.
Checked on the emulator against a fixture with chips in a heading, three
kinds of list item, a quote, a table cell and a link label, unselected and
under Select All, and a link with a chip in its label still opens. Cost,
against the same build without the change, streaming sixty paragraphs of
three chips each: measure 755ms against 776ms, record 327ms against 321ms,
transcript draw 0.22ms in both.
The first half of EXPLORER.md: server/src/files.rs, which lists a
directory, reads a file, writes one, and creates a file or a directory on
whichever machine a setup names.
Each operation is one small POSIX script run through `Transport`, the way
the import listing and the usage fetch already ask a machine a question,
so the local and the ssh case are one implementation rather than two that
drift. The path crosses as a positional argument and never as script
text; `PATH_PRELUDE` is the one line that gives a leading `~` its
meaning, because a shell expands a tilde in text and not in an argument,
and it is the far machine's home that has to answer.
A read has four answers -- text, binary, tooBig, or the machine's own
error -- because a binary file drawn as text and a big one cut off
silently are both wrong in ways the reader cannot see. A write carries
the sha256 the read reported and is refused with a 409 when the file has
moved on, which is what happens whenever an agent is editing the file
somebody is reading.
`Transport::capture_with_input` is the one description of "run this
there, with this on stdin", and `ship_attachment` moves onto it rather
than assembling a second ssh invocation of its own. It is also the only
capture that hands back the exit status, which is how the write says
"this is not the file you read" without that answer looking like a
failure.
Exercised on both transports against the sandbox -- ssh to this VM with a
throwaway key, since the quoting and the stdin path are what that proves
-- including a filename with an apostrophe, one with a tab, an unreadable
file, a binary one, one over the limit, and the 409.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
iris
