os: Frequency, processor counts, times and usage (OS slice 3, step 1)

orion commented

2026-10-25 23:48:54 +00:00

Collaborator

Goal

First step of #127 (OS slice 3, processors): Frequency, the processor counts, the processor times and Processor.usage, on every supported system. describe with the x86/ decoders and the frequencies are the two later steps of the epic.

Source

docs/design/OS.md section 4 (Frequency), section 5.2 (Processor, ProcessorTimes, ProcessorUsage), section 7 (natives per system), section 10 slice 3.

Scope

Frequency in std/os (ofHertz, ofMegahertz, hertz, megahertz, gigahertz, show as 3.6 GHz / 800 MHz, Show, Compare).
Processor.logicalCount(), physicalCount(), times(), timesOfEach(), usage(interval), ProcessorTimes (with Subtract and total()), ProcessorUsage (with between).
Per-system halves: Linux (/proc/stat and sysconf/sysfs topology, pure parsers), Windows (GetLogicalProcessorInformationEx, GetSystemTimes, NtQuerySystemInformation looked up at the call), macOS (host_statistics, host_processor_info, hw.physicalcpu), POSIX/BSD where they share; FreeBSD and the browser answer Unsupported as in slice 2.
usage is TorbScript over times and sleep, as written in OS.md 5.2.

Acceptance criteria

Frequency conversions and show are tested (800 MHz, 3.6 GHz, rounding, ordering).
Processor.logicalCount() and physicalCount() answer Ok on Linux, Windows and macOS, with physicalCount() <= logicalCount() and logicalCount() >= 1.
times() and timesOfEach() answer cumulative times; two readings are monotonic; timesOfEach().length == logicalCount(); ProcessorTimes.total() is the sum of the five fields.
The /proc/stat parser (one aggregate line, one line per processor, iowait, nice, irq, steal folded into other, clock ticks to Duration via CLK_TCK) is pure and tested on every machine with captured inputs, including a short and a malformed line (Err, no panic).
ProcessorUsage.between is tested with fixed differences: fractions are in 0..1, busy == user + system within rounding, and a zero-length difference answers 0, not NaN.
usage(interval) is a Task that sleeps interval between two readings; a test runs it with a short interval and checks the ranges.
FreeBSD and the browser answer OsError.Unsupported; torb check --every-target is clean.
docs/standard-library/os.md documents the new API; docs check clean; skills/ regenerated.
tools/gates.sh a green in forge CI.

Seed impact

None expected: new std names and new natives in the manifest, no syntax change, no new driver flag, and the compiler looks up no name of this step by string and does not import Processor. If CI proves otherwise, the PR uses the two-commit seed rule and says so.

Size

One PR, two Conventional Commits at most: feat(os): Frequency and processor counts and feat(os): processor times and usage, or one commit if small.

Dependencies

#126 is closed (landed). Part of the epic #127; the next steps (describe with x86/, then frequencies and the slice's done mark) are filed when this one lands.

## Goal First step of #127 (OS slice 3, processors): `Frequency`, the processor counts, the processor times and `Processor.usage`, on every supported system. `describe` with the `x86/` decoders and the frequencies are the two later steps of the epic. ## Source - `docs/design/OS.md` section 4 (`Frequency`), section 5.2 (`Processor`, `ProcessorTimes`, `ProcessorUsage`), section 7 (natives per system), section 10 slice 3. ## Scope - `Frequency` in `std/os` (`ofHertz`, `ofMegahertz`, `hertz`, `megahertz`, `gigahertz`, `show` as `3.6 GHz` / `800 MHz`, `Show`, `Compare`). - `Processor.logicalCount()`, `physicalCount()`, `times()`, `timesOfEach()`, `usage(interval)`, `ProcessorTimes` (with `Subtract` and `total()`), `ProcessorUsage` (with `between`). - Per-system halves: Linux (`/proc/stat` and `sysconf`/sysfs topology, pure parsers), Windows (`GetLogicalProcessorInformationEx`, `GetSystemTimes`, `NtQuerySystemInformation` looked up at the call), macOS (`host_statistics`, `host_processor_info`, `hw.physicalcpu`), POSIX/BSD where they share; FreeBSD and the browser answer `Unsupported` as in slice 2. - `usage` is TorbScript over `times` and `sleep`, as written in OS.md 5.2. ## Acceptance criteria - [ ] `Frequency` conversions and `show` are tested (`800 MHz`, `3.6 GHz`, rounding, ordering). - [ ] `Processor.logicalCount()` and `physicalCount()` answer `Ok` on Linux, Windows and macOS, with `physicalCount() <= logicalCount()` and `logicalCount() >= 1`. - [ ] `times()` and `timesOfEach()` answer cumulative times; two readings are monotonic; `timesOfEach().length == logicalCount()`; `ProcessorTimes.total()` is the sum of the five fields. - [ ] The `/proc/stat` parser (one aggregate line, one line per processor, `iowait`, `nice`, `irq`, `steal` folded into `other`, clock ticks to `Duration` via `CLK_TCK`) is pure and tested on every machine with captured inputs, including a short and a malformed line (`Err`, no panic). - [ ] `ProcessorUsage.between` is tested with fixed differences: fractions are in 0..1, `busy == user + system` within rounding, and a zero-length difference answers 0, not NaN. - [ ] `usage(interval)` is a `Task` that sleeps `interval` between two readings; a test runs it with a short interval and checks the ranges. - [ ] FreeBSD and the browser answer `OsError.Unsupported`; `torb check --every-target` is clean. - [ ] `docs/standard-library/os.md` documents the new API; `docs check` clean; `skills/` regenerated. - [ ] `tools/gates.sh a` green in forge CI. ## Seed impact None expected: new std names and new natives in the manifest, no syntax change, no new driver flag, and the compiler looks up no name of this step by string and does not import `Processor`. If CI proves otherwise, the PR uses the two-commit seed rule and says so. ## Size One PR, two Conventional Commits at most: `feat(os): Frequency and processor counts` and `feat(os): processor times and usage`, or one commit if small. ## Dependencies #126 is closed (landed). Part of the epic #127; the next steps (`describe` with `x86/`, then frequencies and the slice's done mark) are filed when this one lands.

orion added the

ready

loop

labels

2026-10-25 23:48:54 +00:00

orion referenced this issue

2026-10-25 23:48:55 +00:00

os: processors (OS slice 3, epic) #127

orion referenced this issue

2026-10-25 23:49:00 +00:00

os: processors (OS slice 3, epic) #127

orion commented

2026-10-26 00:08:54 +00:00

Author

Collaborator

[loop:torbscript] claim implement by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z

orion commented

2026-10-26 00:09:53 +00:00

Author

Collaborator

[loop:torbscript] implement: checkpoint, no code pushed yet (the worker's reading budget ran out on orientation), next: implement continues in a fresh worker. Approx. tokens: 80k (reading only).

Findings for the next worker (so it can write code at once):

Model to copy: commit 319caaf8 (feat(os): add Memory and Swap). Touches: std/os/src/memory.trb (public type + match OperatingSystem.current arms, .FreeBsd/.Browser => Fail OsError.Unsupported(...)), std/os/src/{linux,macos,windows}/memory.trb, {macos,windows}/native.trb, std/os/src/lib.trb (public use + doc list), std/os/tests/*.test.trb, compiler/src/backend/c/natives.trb (manifest rows via systemRuntimeOf(...), per-system guards), runtime/include/torb_os.h, runtime/os/{macos,windows}.c, docs/standard-library/os.md, docs/design/OS.md (slice mark), regenerated skills/.
Linux needs no new native: /proc/stat through readSystemFile (std/os/src/linux/system.trb) with a pure public parser (see parseMemory); CLK_TCK via Posix.configuration("CLK_TCK"), logical count via configuration("NPROCESSORS_ONLN") or the cpuN lines, physical count from /sys/devices/system/cpu/cpuN/topology/{core_id,physical_package_id} (needs listSystemDirectory, not in the Linux natives yet; alternative: parse /proc/cpuinfo physical id/core id, which readSystemFile can read).
macOS: counts via Bsd.sysctlInteger("hw.physicalcpu"/"hw.logicalcpu") (std/os/src/bsd/system.trb), times via new MacOs.processorLoad / processorLoadOfEach (OS.md section 7).
Windows: new Windows.processorTopology/systemTimes/processorTimesOfEach (OS.md section 7), NtQuerySystemInformation looked up at the call as in windows.c page files; 100 ns ticks to Duration.
Design text: OS.md section 4 (Frequency), 5.2 (signatures; usage body is given verbatim), 7 (natives).
Generated files: runtime/machine_natives.c (312 natives, numbered torb_machine_native_N in the sorted order of the C names, count at the end) and runtime/include/torb_natives.h come from torb natives --header (run from compiler/). The sandbox worker has no cc and no torb, so these must be regenerated either by a careful script (insert the new entries at their sorted position and renumber; see the diff of 319caaf8 for the exact shape) or by a CI round trip. Decide this first.
Branch loop/488-os-processors was created from main in a local worktree with no changes; nothing to push.

[loop:torbscript] implement: checkpoint, no code pushed yet (the worker's reading budget ran out on orientation), next: implement continues in a fresh worker. Approx. tokens: 80k (reading only). Findings for the next worker (so it can write code at once): - Model to copy: commit 319caaf8 (`feat(os): add Memory and Swap`). Touches: `std/os/src/memory.trb` (public type + `match OperatingSystem.current` arms, `.FreeBsd`/`.Browser` => `Fail OsError.Unsupported(...)`), `std/os/src/{linux,macos,windows}/memory.trb`, `{macos,windows}/native.trb`, `std/os/src/lib.trb` (public use + doc list), `std/os/tests/*.test.trb`, `compiler/src/backend/c/natives.trb` (manifest rows via `systemRuntimeOf(...)`, per-system guards), `runtime/include/torb_os.h`, `runtime/os/{macos,windows}.c`, `docs/standard-library/os.md`, `docs/design/OS.md` (slice mark), regenerated `skills/`. - Linux needs no new native: `/proc/stat` through `readSystemFile` (std/os/src/linux/system.trb) with a pure public parser (see `parseMemory`); `CLK_TCK` via `Posix.configuration("CLK_TCK")`, logical count via `configuration("NPROCESSORS_ONLN")` or the `cpuN` lines, physical count from `/sys/devices/system/cpu/cpuN/topology/{core_id,physical_package_id}` (needs `listSystemDirectory`, not in the Linux natives yet; alternative: parse `/proc/cpuinfo` `physical id`/`core id`, which `readSystemFile` can read). - macOS: counts via `Bsd.sysctlInteger("hw.physicalcpu"/"hw.logicalcpu")` (`std/os/src/bsd/system.trb`), times via new `MacOs.processorLoad` / `processorLoadOfEach` (OS.md section 7). - Windows: new `Windows.processorTopology`/`systemTimes`/`processorTimesOfEach` (OS.md section 7), `NtQuerySystemInformation` looked up at the call as in `windows.c` page files; 100 ns ticks to `Duration`. - Design text: OS.md section 4 (`Frequency`), 5.2 (signatures; `usage` body is given verbatim), 7 (natives). - Generated files: `runtime/machine_natives.c` (312 natives, numbered `torb_machine_native_N` in the sorted order of the C names, count at the end) and `runtime/include/torb_natives.h` come from `torb natives --header` (run from `compiler/`). The sandbox worker has no `cc` and no `torb`, so these must be regenerated either by a careful script (insert the new entries at their sorted position and renumber; see the diff of 319caaf8 for the exact shape) or by a CI round trip. Decide this first. - Branch `loop/488-os-processors` was created from main in a local worktree with no changes; nothing to push.

orion commented

2026-10-26 00:11:00 +00:00

Author

Collaborator

[loop:torbscript] claim implement by session:loop-torbscript-20261026 until 2026-10-26T06:12:00Z

orion commented

2026-10-26 00:13:19 +00:00

Author

Collaborator

[loop:torbscript] implement: checkpoint (cont 1 ended at the context guard; no code written yet, only design decisions fixed)

Branch: loop/488-os-processors (from origin/main c72cabec). Worktree: /home/node/workspaces/orion/code/wt/torbscript-language-488 (clean).

Decisions (read from 319caaf8, std/time, std/number):

Natives cannot return List out-params cheaply (lists are a "coerced collection" in the manifest). So per-processor natives answer ONE var text: String, one line per processor; TorbScript parses it with a pure public parser. macOS: MacOs.processorLoad(var user, var system, var idle, var nice, var failure) (ticks) and MacOs.processorLoadOfEach(var text, var failure) (lines user system idle nice). Windows: Windows.processorCounts(var packages, var cores, var logical, var failure) (GetLogicalProcessorInformationEx), Windows.systemTimes(var idle, var kernel, var user, var failure) (100 ns), Windows.processorTimesOfEach(var text, var failure) (lines idle kernel user, from NtQuerySystemInformation class 8, struct = 5 LARGE_INTEGER + ULONG, looked up at the call like page files). Windows: system = kernel - idle, waiting/other = 0 (same for aggregate and each, so they agree). Update OS.md section 7 rows accordingly (deviation from "parallel lists").
Duration has only Int64.seconds()/milliseconds() constructors. Helper durationOfTicks(ticks, perSecond): build ns via Timestamp.fromUnixNanoseconds(n) - Timestamp.fromUnixNanoseconds(0) (Timestamp subtract gives a Duration), n = ticks / hz * 1e9 + ticks % hz * 1e9 / hz (no overflow).
Frequency.show: whole MHz rounded; if rounded MHz >= 1000 then GHz in tenths rounded half up ((v + 50_000_000) / 100_000_000), below 1 MHz "{v} Hz".
Linux: parseTimes(text, ticksPerSecond) in std/os/src/linux/processor.trb, cpu aggregate + cpuN lines; fewer than 4 numbers => OsError.Malformed("/proc/stat", ...); other = nice + irq + softirq + steal (guest/guest_nice already inside user/nice); waiting = iowait. CLK_TCK via Posix.configuration("CLK_TCK"); logical = configuration("NPROCESSORS_ONLN"); physical = distinct (physical id,core id) of /proc/cpuinfo via readSystemFile, falls back to logical when absent (Arm), clamped to <= logical.
ProcessorUsage.between: busy = (user + system)/total (per issue AC busy == user + system), fractions clamped to 0..1, zero total => 0.
usage as OS.md 5.2 (public fn ... : Task<Result<...>> pattern: std/network/src/tcp.trb:25; sleep(seconds: Float64): Task<Void> from std/time, .await()).
Test pattern: pure parser tests in std/os/tests/processor.test.trb (see memory.test.trb), host-shape tests in host.test.trb.

Next step: write std/os/src/processor.trb (+ linux/macos/windows halves, natives, lib.trb), tests, compiler/src/backend/c/natives.trb rows, torb_os.h, macos.c/windows.c, regenerate machine_natives.c/torb_natives.h by script (outside repo), docs (os.md, OS.md slice mark/section 7), skills/ regen; push, PR, ci-wait.

[loop:torbscript] implement: checkpoint (cont 1 ended at the context guard; no code written yet, only design decisions fixed) Branch: `loop/488-os-processors` (from origin/main c72cabec). Worktree: `/home/node/workspaces/orion/code/wt/torbscript-language-488` (clean). Decisions (read from 319caaf8, `std/time`, `std/number`): - Natives cannot return `List` out-params cheaply (lists are a "coerced collection" in the manifest). So per-processor natives answer ONE `var text: String`, one line per processor; TorbScript parses it with a pure public parser. macOS: `MacOs.processorLoad(var user, var system, var idle, var nice, var failure)` (ticks) and `MacOs.processorLoadOfEach(var text, var failure)` (lines `user system idle nice`). Windows: `Windows.processorCounts(var packages, var cores, var logical, var failure)` (GetLogicalProcessorInformationEx), `Windows.systemTimes(var idle, var kernel, var user, var failure)` (100 ns), `Windows.processorTimesOfEach(var text, var failure)` (lines `idle kernel user`, from NtQuerySystemInformation class 8, struct = 5 LARGE_INTEGER + ULONG, looked up at the call like page files). Windows: system = kernel - idle, waiting/other = 0 (same for aggregate and each, so they agree). Update OS.md section 7 rows accordingly (deviation from "parallel lists"). - `Duration` has only `Int64.seconds()/milliseconds()` constructors. Helper `durationOfTicks(ticks, perSecond)`: build ns via `Timestamp.fromUnixNanoseconds(n) - Timestamp.fromUnixNanoseconds(0)` (Timestamp subtract gives a Duration), n = `ticks / hz * 1e9 + ticks % hz * 1e9 / hz` (no overflow). - `Frequency.show`: whole MHz rounded; if rounded MHz >= 1000 then GHz in tenths rounded half up (`(v + 50_000_000) / 100_000_000`), below 1 MHz `"{v} Hz"`. - Linux: `parseTimes(text, ticksPerSecond)` in `std/os/src/linux/processor.trb`, `cpu` aggregate + `cpuN` lines; fewer than 4 numbers => `OsError.Malformed("/proc/stat", ...)`; other = nice + irq + softirq + steal (guest/guest_nice already inside user/nice); waiting = iowait. `CLK_TCK` via `Posix.configuration("CLK_TCK")`; logical = `configuration("NPROCESSORS_ONLN")`; physical = distinct (`physical id`,`core id`) of `/proc/cpuinfo` via `readSystemFile`, falls back to logical when absent (Arm), clamped to <= logical. - `ProcessorUsage.between`: busy = (user + system)/total (per issue AC `busy == user + system`), fractions clamped to 0..1, zero total => 0. - `usage` as OS.md 5.2 (`public fn ... : Task<Result<...>>` pattern: `std/network/src/tcp.trb:25`; `sleep(seconds: Float64): Task<Void>` from std/time, `.await()`). - Test pattern: pure parser tests in `std/os/tests/processor.test.trb` (see memory.test.trb), host-shape tests in `host.test.trb`. Next step: write `std/os/src/processor.trb` (+ linux/macos/windows halves, natives, lib.trb), tests, `compiler/src/backend/c/natives.trb` rows, `torb_os.h`, `macos.c`/`windows.c`, regenerate `machine_natives.c`/`torb_natives.h` by script (outside repo), docs (`os.md`, OS.md slice mark/section 7), `skills/` regen; push, PR, ci-wait.

orion commented

2026-10-26 00:13:49 +00:00

Author

Collaborator

[loop:torbscript] claim implement (continuation 2) by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z

~~orion referenced this issue 2026-10-26 00:16:29 +00:00~~

feat(os): Frequency and processor counts, times and usage #491

orion commented

2026-10-26 00:16:39 +00:00

Author

Collaborator

[loop:torbscript] implement: checkpoint (continuation 2 reached the context guard after step 1), next: implement continues in a fresh worker. PR: #491 (no 'Closes' yet). Approx. tokens: 75k.

Branch loop/488-os-processors, worktree /home/node/workspaces/orion/code/wt/torbscript-language-488. Push needs git -c credential.helper= -c "http.extraHeader=Authorization: token $(cat ~/.secrets/torbdev-token)" push.

DONE (commit 1, pushed, unverified: no local torb, CI not run yet):

std/os/src/processor.trb: Frequency, Processor (counts/times/timesOfEach/usage), ProcessorTimes (Subtract, total), ProcessorUsage.between, public durationOfTicks.
std/os/src/linux/processor.trb: logical/physical count, parseTimes(text, ticks) -> (total, each), parseCoreCount.
std/os/src/macos/processor.trb (loadOf, parseLoads), std/os/src/windows/processor.trb (timesOf, parseTimes).
native declarations in macos/native.trb (processorLoad, processorLoadOfEach) and windows/native.trb (processorCounts, systemTimes, processorTimesOfEach); lib.trb exports + doc list.

NOT DONE:
2. tests std/os/tests/processor.test.trb (Frequency show/order, parseTimes incl. short and malformed line, parseCoreCount, macOS/Windows parsers, ProcessorUsage.between incl. zero difference, usage with a short interval) and host-shape tests in host.test.trb (counts, monotonic times, timesOfEach length == logicalCount, FreeBSD/browser Unsupported by torb check --every-target).
3. manifest rows in compiler/src/backend/c/natives.trb (copy shape of virtualMemory/pageFiles rows), runtime/include/torb_os.h, runtime/os/macos.c (host_statistics HOST_CPU_LOAD_INFO; host_processor_info lines user system idle nice), runtime/os/windows.c (GetLogicalProcessorInformationEx counts; GetSystemTimes; NtQuerySystemInformation class 8 lines idle kernel user), then regenerate runtime/machine_natives.c + runtime/include/torb_natives.h by script (no cc/torb in the sandbox).
4. docs: docs/standard-library/os.md, docs/design/OS.md section 7 rows (replace 'parallel lists' with the one-text design) and slice mark, regenerate skills/.
5. set 'Closes #488' in the PR, ci-wait.sh, fix red CI, request Argus.

Unsure syntax in commit 1 that CI may flag: self.other inside ProcessorTimes.subtract (param shadows the field), Ok((total: found, each: each)) tuple result, socket-named vars, Ok ProcessorTimes(...) multi-line calls, name.withoutPrefix("cpu").

[loop:torbscript] implement: checkpoint (continuation 2 reached the context guard after step 1), next: implement continues in a fresh worker. PR: https://git.torb.dev/torbscript/language/pulls/491 (no 'Closes' yet). Approx. tokens: 75k. Branch `loop/488-os-processors`, worktree `/home/node/workspaces/orion/code/wt/torbscript-language-488`. Push needs `git -c credential.helper= -c "http.extraHeader=Authorization: token $(cat ~/.secrets/torbdev-token)" push`. DONE (commit 1, pushed, unverified: no local `torb`, CI not run yet): - `std/os/src/processor.trb`: Frequency, Processor (counts/times/timesOfEach/usage), ProcessorTimes (Subtract, total), ProcessorUsage.between, public `durationOfTicks`. - `std/os/src/linux/processor.trb`: logical/physical count, `parseTimes(text, ticks)` -> `(total, each)`, `parseCoreCount`. - `std/os/src/macos/processor.trb` (`loadOf`, `parseLoads`), `std/os/src/windows/processor.trb` (`timesOf`, `parseTimes`). - native declarations in `macos/native.trb` (`processorLoad`, `processorLoadOfEach`) and `windows/native.trb` (`processorCounts`, `systemTimes`, `processorTimesOfEach`); `lib.trb` exports + doc list. NOT DONE: 2. tests `std/os/tests/processor.test.trb` (Frequency show/order, parseTimes incl. short and malformed line, parseCoreCount, macOS/Windows parsers, ProcessorUsage.between incl. zero difference, usage with a short interval) and host-shape tests in `host.test.trb` (counts, monotonic times, timesOfEach length == logicalCount, FreeBSD/browser Unsupported by `torb check --every-target`). 3. manifest rows in `compiler/src/backend/c/natives.trb` (copy shape of `virtualMemory`/`pageFiles` rows), `runtime/include/torb_os.h`, `runtime/os/macos.c` (host_statistics HOST_CPU_LOAD_INFO; host_processor_info lines `user system idle nice`), `runtime/os/windows.c` (GetLogicalProcessorInformationEx counts; GetSystemTimes; NtQuerySystemInformation class 8 lines `idle kernel user`), then regenerate `runtime/machine_natives.c` + `runtime/include/torb_natives.h` by script (no `cc`/`torb` in the sandbox). 4. docs: `docs/standard-library/os.md`, `docs/design/OS.md` section 7 rows (replace 'parallel lists' with the one-text design) and slice mark, regenerate `skills/`. 5. set 'Closes #488' in the PR, ci-wait.sh, fix red CI, request Argus. Unsure syntax in commit 1 that CI may flag: `self.other` inside `ProcessorTimes.subtract` (param shadows the field), `Ok((total: found, each: each))` tuple result, `socket`-named vars, `Ok ProcessorTimes(...)` multi-line calls, `name.withoutPrefix("cpu")`.

orion commented

2026-10-26 00:17:31 +00:00

Author

Collaborator

[loop:torbscript] claim implement (continuation 3, last) by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z

orion commented

2026-10-26 00:21:27 +00:00

Author

Collaborator

[loop:torbscript] implement: continuation 3 pushed tests and natives (commits 5ee05404, 1ef1e8e4), still not compiled or CI-checked. next: waiting (continuation limit reached, remaining: see below). PR: #491 (no 'Closes' yet). Approx. tokens: 85k.

DONE this run:

std/os/tests/processor.test.trb: Frequency show/units/order, durationOfTicks (remainder, no overflow), /proc/stat (recorded 6.8 input, 4-field kernel, short line, non-number, missing cpu line), /proc/cpuinfo core count (x86 SMT, two sockets, Arm without ids), macOS loadOf/parseLoads, Windows timesOf/parseTimes, ProcessorUsage.between (fractions, subtraction, zero total, backwards counter).
host.test.trb: group "the processors" (counts, monotonic times, timesOfEach length == logicalCount, usage of 50 ms in 0..1).
Manifest rows in compiler/src/backend/c/natives.trb (MacOs.processorLoad/processorLoadOfEach, Windows.processorCounts/systemTimes/processorTimesOfEach), torb_os.h prototypes, C in runtime/os/macos.c and runtime/os/windows.c, torb_natives.h + machine_natives.c regenerated by script (sorted by C name, renumbered, count 317).

REMAINING (precise):

Run CI on the branch (ci-wait.sh); expect fixes in commit-1 syntax (see #6668: self.other in ProcessorTimes.subtract, tuple result Ok((total:, each:)), withoutPrefix) and in the generated files if torb natives --header output differs from the script (regenerate with the real tool in CI output if the gate diffs).
FreeBSD/browser Unsupported: covered by the match arms; confirm torb check --every-target is green in CI.
Docs: docs/standard-library/os.md (Processor, ProcessorTimes, ProcessorUsage, Frequency), docs/design/OS.md section 7 rows (one-text design instead of 'parallel lists') + slice-3 step-1 mark; regenerate skills/.
PR body 'Closes #488', CI green, request Argus review.

Worktree removed after push; recreate from loop/488-os-processors.

[loop:torbscript] implement: continuation 3 pushed tests and natives (commits 5ee05404, 1ef1e8e4), still not compiled or CI-checked. next: waiting (continuation limit reached, remaining: see below). PR: https://git.torb.dev/torbscript/language/pulls/491 (no 'Closes' yet). Approx. tokens: 85k. DONE this run: - `std/os/tests/processor.test.trb`: Frequency show/units/order, durationOfTicks (remainder, no overflow), `/proc/stat` (recorded 6.8 input, 4-field kernel, short line, non-number, missing cpu line), `/proc/cpuinfo` core count (x86 SMT, two sockets, Arm without ids), macOS `loadOf`/`parseLoads`, Windows `timesOf`/`parseTimes`, `ProcessorUsage.between` (fractions, subtraction, zero total, backwards counter). - `host.test.trb`: group "the processors" (counts, monotonic times, timesOfEach length == logicalCount, usage of 50 ms in 0..1). - Manifest rows in `compiler/src/backend/c/natives.trb` (MacOs.processorLoad/processorLoadOfEach, Windows.processorCounts/systemTimes/processorTimesOfEach), `torb_os.h` prototypes, C in `runtime/os/macos.c` and `runtime/os/windows.c`, `torb_natives.h` + `machine_natives.c` regenerated by script (sorted by C name, renumbered, count 317). REMAINING (precise): 1. Run CI on the branch (ci-wait.sh); expect fixes in commit-1 syntax (see #6668: `self.other` in `ProcessorTimes.subtract`, tuple result `Ok((total:, each:))`, `withoutPrefix`) and in the generated files if `torb natives --header` output differs from the script (regenerate with the real tool in CI output if the gate diffs). 2. FreeBSD/browser `Unsupported`: covered by the match arms; confirm `torb check --every-target` is green in CI. 3. Docs: `docs/standard-library/os.md` (Processor, ProcessorTimes, ProcessorUsage, Frequency), `docs/design/OS.md` section 7 rows (one-text design instead of 'parallel lists') + slice-3 step-1 mark; regenerate `skills/`. 4. PR body 'Closes #488', CI green, request Argus review. Worktree removed after push; recreate from `loop/488-os-processors`.

orion commented

2026-10-26 00:22:27 +00:00

Author

Collaborator

[loop:torbscript] claim fix by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z (finish PR 491: compile, CI)

orion commented

2026-10-26 00:54:08 +00:00

Author

Collaborator

[loop:torbscript] fix: docs commit 13855d31 pushed (os.md, OS.md rows), worker ran out of time waiting on CI (runner queue busy, bootstrap for 13855d31 only started 00:49Z). next: fix (continuation). Remaining: read CI result on 13855d31 and fix compile/gate findings (ProcessorTimes.subtract self.other, tuple result Ok((total:, each:)), withoutPrefix, regenerate natives files if gate diffs), regenerate skills/, PR body Closes #488, CI green, request Argus review.

[loop:torbscript] fix: docs commit 13855d31 pushed (os.md, OS.md rows), worker ran out of time waiting on CI (runner queue busy, bootstrap for 13855d31 only started 00:49Z). next: fix (continuation). Remaining: read CI result on 13855d31 and fix compile/gate findings (ProcessorTimes.subtract `self.other`, tuple result `Ok((total:, each:))`, withoutPrefix, regenerate natives files if gate diffs), regenerate `skills/`, PR body `Closes #488`, CI green, request Argus review.

orion referenced this issue from a pull request that will close it,

2026-10-26 00:55:25 +00:00

feat(os): Frequency and processor counts, times and usage #491

orion commented

2026-10-26 01:08:36 +00:00

Author

Collaborator

[loop:torbscript] fix: regenerated skills/ (std/os page + three index summaries) and docs/standard-library/index.md summary for the new Processor text, pushed 96d749f7; PR 491 body now has Closes #488. No torb binary was available outside CI, so the skills change is a patch of docs/standard-library/os.md plus the link rewrite the generator does and the summary line; tier A's torb docs skill docs skills --check will confirm or show the exact diff. The self.other in ProcessorTimes.subtract, the tuple result Ok((total: found, each: each)) and withoutPrefix are already in the code, not yet compiled. CI of 96d749f7 had not started after ~20 min (runner queue busy with other commits), so no result yet. next: fix (continuation). Remainder: read CI of 96d749f7 once it ran (bootstrap, tier A, tier B), fix compile/gate findings (including a skills --check diff, regenerate natives files if a gate diffs), push, then request Argus review on PR 491 when green.

[loop:torbscript] fix: regenerated `skills/` (std/os page + three index summaries) and `docs/standard-library/index.md` summary for the new Processor text, pushed 96d749f7; PR 491 body now has `Closes #488`. No `torb` binary was available outside CI, so the skills change is a patch of `docs/standard-library/os.md` plus the link rewrite the generator does and the summary line; tier A's `torb docs skill docs skills --check` will confirm or show the exact diff. The `self.other` in `ProcessorTimes.subtract`, the tuple result `Ok((total: found, each: each))` and `withoutPrefix` are already in the code, not yet compiled. CI of 96d749f7 had not started after ~20 min (runner queue busy with other commits), so no result yet. next: fix (continuation). Remainder: read CI of 96d749f7 once it ran (bootstrap, tier A, tier B), fix compile/gate findings (including a skills `--check` diff, regenerate natives files if a gate diffs), push, then request Argus review on PR 491 when green.

orion commented

2026-10-26 01:10:36 +00:00

Author

Collaborator

[loop:torbscript] release fix: the fix worker timed out at its limit and its continuation finished; PR 491 head 96d749f7 is waiting in the forge CI runner queue, no verdict yet, next: waiting (CI on PR 491, then fix findings and request Argus review)

[loop:torbscript] release fix: the fix worker timed out at its limit and its continuation finished; PR 491 head 96d749f7 is waiting in the forge CI runner queue, no verdict yet, next: waiting (CI on PR 491, then fix findings and request Argus review)

orion commented

2026-10-26 07:16:54 +00:00

Author

Collaborator

[loop:torbscript] claim fix by session:loop-torbscript-20261026-0724 until 2026-10-26T13:16:00Z (PR 491: tier A and gates red, base moved after PR 487 landed)

orion commented

2026-10-26 07:24:37 +00:00

Author

Collaborator

[loop:torbscript] fix (checkpoint): rebased onto main 4f8e15cf (the generated runtime/machine_natives.c merged by taking main's thunks plus the five new processor thunks, renumbered and sorted by C name); fixed the red causes of run 1837 (tier A): Posix.configuration knew no NPROCESSORS_ONLN (processor tests failed on Linux), five audit rows for the new natives in docs/design/NATIVES.md (keep 190, total 546), std/os example awaits at the top level, page summary under 240 chars, two lines over 120 columns broken. New head eefbcb29 pushed (force-with-lease), CI is running; format --check could not be run locally (no C compiler for a seed), so that gate may still name files. next: waiting (CI on PR 491, then Argus review). Usage of this worker so far: about 110k tokens of context, no more than a few hundred thousand in total.

[loop:torbscript] fix (checkpoint): rebased onto main 4f8e15cf (the generated runtime/machine_natives.c merged by taking main's thunks plus the five new processor thunks, renumbered and sorted by C name); fixed the red causes of run 1837 (tier A): Posix.configuration knew no NPROCESSORS_ONLN (processor tests failed on Linux), five audit rows for the new natives in docs/design/NATIVES.md (keep 190, total 546), std/os example awaits at the top level, page summary under 240 chars, two lines over 120 columns broken. New head eefbcb29 pushed (force-with-lease), CI is running; format --check could not be run locally (no C compiler for a seed), so that gate may still name files. next: waiting (CI on PR 491, then Argus review). Usage of this worker so far: about 110k tokens of context, no more than a few hundred thousand in total.

orion commented

2026-10-26 09:15:52 +00:00

Author

Collaborator

[loop:torbscript] claim fix (continuation 1) by session:loop-torbscript-20261026-0724 until 2026-10-26T15:30:00Z (PR 491 head eefbcb29: commit messages red, tier A/B running)

[loop:torbscript] claim fix (continuation 1) by session:loop-torbscript-20261026-0724 until 2026-10-26T15:30:00Z (PR 491 head eefbcb29: commit messages red, tier A/B running)

orion commented

2026-10-26 09:22:59 +00:00

Author

Collaborator

[loop:torbscript] fix (checkpoint, continuation 1): run 1853 had only the commit messages job red: the head commit subject had 115 characters (limit 100); the six commits are squashed into one Conventional Commit (feat(os): Processor, ProcessorTimes, ProcessorUsage and Frequency) and rebased onto main b4852363 (the three generated index files merged cleanly, tools/check-commits.sh is clean). Run 1853 was cancelled by the update; the old tier A/B result is void. New head 798dfe13 pushed (force-with-lease), runs 1911/1912 are still queued in the runner queue, nothing started yet. next: waiting (CI on PR 491, then fix findings and request Argus review). Usage of this worker: about 60k tokens.

[loop:torbscript] fix (checkpoint, continuation 1): run 1853 had only the commit messages job red: the head commit subject had 115 characters (limit 100); the six commits are squashed into one Conventional Commit (feat(os): Processor, ProcessorTimes, ProcessorUsage and Frequency) and rebased onto main b4852363 (the three generated index files merged cleanly, tools/check-commits.sh is clean). Run 1853 was cancelled by the update; the old tier A/B result is void. New head 798dfe13 pushed (force-with-lease), runs 1911/1912 are still queued in the runner queue, nothing started yet. next: waiting (CI on PR 491, then fix findings and request Argus review). Usage of this worker: about 60k tokens.

orion commented

2026-10-26 10:54:47 +00:00

Author

Collaborator

[loop:torbscript] release fix (continuation 1): the fix worker timed out after its checkpoint; PR 491 head 798dfe13 (single squashed commit, commit-message subject fixed) is in forge CI, tier A and tier B started, the rest queued. next: waiting (CI on PR 491 head 798dfe13, then fix findings and Argus review)

[loop:torbscript] release fix (continuation 1): the fix worker timed out after its checkpoint; PR 491 head 798dfe13 (single squashed commit, commit-message subject fixed) is in forge CI, tier A and tier B started, the rest queued. next: waiting (CI on PR 491 head 798dfe13, then fix findings and Argus review)

orion commented

2026-10-26 12:28:13 +00:00

Author

Collaborator

[loop:torbscript] fix: the commit messages job was red on 22d494a5 because style is not an allowed Conventional Commit type (tools/check-commits.sh); the format fix is folded into the single feat(os) commit, PR 491 head is now 69cddc8 (same tree as 22d494a5), force-pushed. next: waiting (CI on PR 491 head 69cddc8, then Argus review)

[loop:torbscript] fix: the commit messages job was red on 22d494a5 because `style` is not an allowed Conventional Commit type (tools/check-commits.sh); the format fix is folded into the single `feat(os)` commit, PR 491 head is now 69cddc8 (same tree as 22d494a5), force-pushed. next: waiting (CI on PR 491 head 69cddc8, then Argus review)