[FEATURE] Maintain bounded profiling and eliminate evidenced startup and build bottlenecks #3

Open
opened 2026-09-12 07:56:23 +02:00 by erikinkinen · 6 comments
Owner

Problem and motivation

READY and command timings identify slow intervals but cannot attribute all loader, mapping, scheduler, I/O and terminal costs. A maintained profiler is needed to replace uncoordinated experiments and support reproducible optimization decisions.

Proposed behavior and scope

Build a versioned capture/report pipeline with exact provenance, bounded collectors, defensible symbol attribution, opt-in hooks where needed, and controlled startup/build baselines. Rank end-to-end bottlenecks, apply reviewed fixes, and retain comparable before/after evidence.

This issue records planned work; its unchecked criteria are not implementation proof. The normative basis is Phase 6 and AC1–AC24.

Native lifecycle captures

Add an optional bounded TCG profile selector to the native, scenario and bootloader VM entries. TCG_PROFILE_PLUGIN, TCG_PROFILE_OUTPUT, TCG_PROFILE_BLOCKS and TCG_PROFILE_BYTES must be supplied together with an explicit absolute QEMU_BIN. With all four absent, preserve the original command and perform no profiler-path lookup. A shared helper checks the entire selection before each entry's first build or VM effect: explicit regular plugin/QEMU files, fresh absolute output, option-delimiter safety, positive decimal budgets and checked capture arithmetic. Preparation validates without creating a capture.

The selector adds only -plugin and its derived option. Keep the original guest features, media, firmware, release/TCG/CPU settings, serial markers, QEMU exit gate, process owner and deadlines. It grants no guest capability or general QEMU argument interface. Retain VM outcome and profiler completeness independently; a successful guest cannot admit a missing, exhausted or killed capture. Add disabled-command, partial-selection, malformed-input, exact-forwarding and incomplete-capture controls, and document the operator interface.

Use an explicitly selected plugin-capable QEMU with the original writable media support. Admit it with the existing five-case plugin fixture, actual media probe and unchanged uninstrumented native scenarios. Compare original member-watch and initial full/zero workloads under the same selected binary and identical guest artifacts. Preserve raw captures and unmatched code; TCG counts are instruction upper bounds, not CPU shares or process attribution. Propose a Kernel optimization only after these captures identify relevant work, then preserve every original authority, lifecycle, cleanup and VM requirement in the equivalent-input comparison. This host observer does not establish EriX guest-build throughput.

Owned invocation measurements

Use a maintained host workload using actual kernel process, CSpace, endpoint and revocation objects plus public owned-invocation entry points. Measure queued-request progress, collected-result progress, repeated collection and full request cycles against explicitly selected retained populations. Separate setup, operation and cleanup durations; retain exact source/binary identity, parameters, actual completed-operation counts and independent caller-origin, FIFO, byte, capability, foreign-selector and final-disposal checks. Each sample uses one owned process and explicit inputs, deadlines and output budgets; no VM, network, signing credential or unrelated-process authority is required for measurement. Integration must retain complete process status and raw evidence, reject incomplete or mismatched samples, and report comparisons only for equivalent source/workload contexts. Build once per source into distinct targets and run uncontended before/after samples. Host measurements do not establish guest or self-host-build throughput. Registry and endpoint-binding scans are hypotheses until measured; any optimization must preserve exact identity, allocation-safe publication and accounted revocation/terminal cleanup. Ordinary functional tests gain no timing-ratio assertion.

Every sample checks exact bytes and SEND capability bindings, authenticated origin, foreign selection, duplicate completion refusal, FIFO position reuse, one-time collection and final invocation disposal. Numeric search positions remain non-authoritative: every fast path must still check current capability type/rights, endpoint identity/in-use state and exact CSpace/slot binding, with complete lookup on a stale hint. Original workload bytes and sample policy must remain unchanged for comparison.

Authority, security and reliability

Capture authority is explicit, finite and retired after success or failure. Exclude keys, capability contents, consent input and arbitrary process memory; diagnostic access must not grant control or namespace access. Bind PCs to exact process generations and module identities, and keep wall time, CPU time, blocking and samples distinct.

Apply the priority order: security, reliability, then performance. Keep suspected vulnerabilities in the repository’s restricted SECURITY.md reporting channel.

Acceptance criteria

Checklist
  • Inventory existing READY/command timers and private QMP/PC evidence; define which
    collectors are reusable, incomplete, obsolete or unsafe to retain.
  • Specify a versioned profile format with source/image/config hashes, toolchain,
    hardware/QEMU/acceleration/CPU/memory/worker settings, time domain, capture interval, exact
    process generation and module identity.
  • Define explicit capture authority, lifecycle and permitted fields; exclude key
    material, capability contents, consent input and arbitrary process memory. Diagnostics must
    not confer process control or namespace access.
  • Add a bounded host collector with hard/silence limits, owned PID cleanup,
    interruption handling, raw-log preservation and true failure status.
  • Implement exact image/ELF/PT_LOAD/source binding and symbolization; reject stale
    maps and ambiguous userspace addresses without process/module identity. Do not infer process
    ownership from an overlapping virtual address.
  • Add authenticated, opt-in guest timing/counter hooks where host samples cannot
    attribute a cost; compile/runtime gate them appropriately and prove their absence from clean
    release images when disabled.
  • Record boot service stages: image resolution/read, relocation,
    mapping/materialization, process start, startup delivery, initialization, publication,
    cleanup, final READY and full shell-caret display.
  • Record command stages: Enter acceptance, parse/expand/resolve, authority
    request/decision, file reads, link/load, spawn/scheduling, first execution/output, output
    drain, exit/wait and fresh caret.
  • Measure IPC count/bytes/waits, runnable versus blocked time, context switches,
    mapping scans/batches, page allocation/faults, VFS/provider fanout, block request size/LBA,
    cache hits/misses and TTY/framebuffer work.
  • Bound buffers through explicit resource policy, report dropped records, clock
    discontinuities and partial captures; never label incomplete data a complete profile or
    fabricate a percentage of total CPU time.
  • Implement reproducible reports: service timeline, command critical path,
    per-component counts, latency distributions and symbol/sample summaries. Separate wall time,
    CPU time, blocking time and sampling observations.
  • Add deterministic parser/attribution/error fixtures, malformed/stale identity
    tests, dropped-record tests and collector cancellation/timeout tests.
  • Add VM profiler tests that identify known synthetic CPU, wait and I/O workloads,
    reject cross-realm capture and verify all capture authority is retired. Compare profiled
    versus unprofiled overhead without subtracting it from acceptance measurements.
  • Establish controlled cold/warm startup and build baselines with isolated repeats,
    explicit resource limits and confidence/variance reporting. Preserve the stricter carried integration#6
    startup thresholds without moving endpoints.
  • Rank evidenced bottlenecks by measured end-to-end impact; create individual integration#2
    findings with one causal hypothesis, owner and acceptance test.
  • Optimize the highest-impact proven path, then reprofile and repeat: inspect
    loader/provider fanout, mapping/process startup, event waits, build dependency scheduling and
    terminal delivery according to evidence, not guesses.
  • For each optimization, prove unchanged authorization, revocation, permissions,
    object identity, failure ordering, output semantics and fairness; retain strict tests and
    isolated same-configuration before/after measurements.
  • Add reviewed startup/build throughput and resource regression budgets to CI, using
    dedicated uncontended performance jobs and calibrated full-build deadlines. Ordinary
    functional jobs still have forced stall limits.
  • Remove superseded diagnostic hooks/helpers and document how to capture, interpret
    and reproduce profiles. Keep only evidence-backed tooling.

For each implementation slice, retain actual formatting, strict Clippy, unit/doctest and warning-denied build results for all altered Rust repositories and valid configurations. Add relevant runtime VM coverage, monitor older unit/VM regressions in exact-head CI, and update canonical component documents and affected technical-manual/API material. Every authored code file must remain below 1,000 physical lines, with meaningful inline documentation and missing_docs enforcement in Rust crates.

Alternatives and tradeoffs

A raw address can overlap several userspace images, and sample counts are not CPU shares. Reuse sound host observers first; add authenticated guest hooks only where they resolve a real attribution gap and prove their absence when disabled.

Tracking and rollout

Track bounded implementation and measured fixes in the owning component issues, with each result linked from its review. Keep actual status in comments and check a criterion only when its complete scope is established. Native lifecycle captures and host invocation measurements do not establish either EriX guest build generation or the external Rust/LLVM guest rebuild.

Related component issues and contracts
Component Links
bootloader PR 3
docs issue 11, PR 4
exsh issue 4, issue 9, compiler-test-profiling.md
integration issue 1, issue 18, issue 20, issue 3, issue 30, issue 31, issue 39, issue 40, issue 45, issue 47, issue 54, issue 65, issue 69, issue 70, issue 71, issue 72, issue 73, issue 74, issue 75, issue 76, issue 77, issue 78, issue 79, issue 80, issue 81, issue 82, issue 83, issue 84, issue 91, issue 96, issue 97, issue 98, issue 99, PR 12, tcg-profiling.md
kernel issue 22, issue 23, issue 24, issue 25, issue 26, issue 27, issue 28, issue 29, issue 30, issue 31, issue 32, issue 33, issue 34, issue 35, issue 36, issue 37, issue 38, issue 39, issue 40, issue 41, issue 42, PR 3, owned-address-spaces.md
lib-cstd PR 3
meta issue 2
procd issue 5
## Problem and motivation READY and command timings identify slow intervals but cannot attribute all loader, mapping, scheduler, I/O and terminal costs. A maintained profiler is needed to replace uncoordinated experiments and support reproducible optimization decisions. ## Proposed behavior and scope Build a versioned capture/report pipeline with exact provenance, bounded collectors, defensible symbol attribution, opt-in hooks where needed, and controlled startup/build baselines. Rank end-to-end bottlenecks, apply reviewed fixes, and retain comparable before/after evidence. This issue records planned work; its unchecked criteria are not implementation proof. The normative basis is [Phase 6 and AC1–AC24](https://git.erikinkinen.fi/erix/docs/src/branch/main/phases/6.md). ### Native lifecycle captures Add an optional bounded TCG profile selector to the native, scenario and bootloader VM entries. `TCG_PROFILE_PLUGIN`, `TCG_PROFILE_OUTPUT`, `TCG_PROFILE_BLOCKS` and `TCG_PROFILE_BYTES` must be supplied together with an explicit absolute `QEMU_BIN`. With all four absent, preserve the original command and perform no profiler-path lookup. A shared helper checks the entire selection before each entry's first build or VM effect: explicit regular plugin/QEMU files, fresh absolute output, option-delimiter safety, positive decimal budgets and checked capture arithmetic. Preparation validates without creating a capture. The selector adds only `-plugin` and its derived option. Keep the original guest features, media, firmware, release/TCG/CPU settings, serial markers, QEMU exit gate, process owner and deadlines. It grants no guest capability or general QEMU argument interface. Retain VM outcome and profiler completeness independently; a successful guest cannot admit a missing, exhausted or killed capture. Add disabled-command, partial-selection, malformed-input, exact-forwarding and incomplete-capture controls, and document the operator interface. Use an explicitly selected plugin-capable QEMU with the original writable media support. Admit it with the existing five-case plugin fixture, actual media probe and unchanged uninstrumented native scenarios. Compare original member-watch and initial full/zero workloads under the same selected binary and identical guest artifacts. Preserve raw captures and unmatched code; TCG counts are instruction upper bounds, not CPU shares or process attribution. Propose a Kernel optimization only after these captures identify relevant work, then preserve every original authority, lifecycle, cleanup and VM requirement in the equivalent-input comparison. This host observer does not establish EriX guest-build throughput. ### Owned invocation measurements Use a maintained host workload using actual kernel process, CSpace, endpoint and revocation objects plus public owned-invocation entry points. Measure queued-request progress, collected-result progress, repeated collection and full request cycles against explicitly selected retained populations. Separate setup, operation and cleanup durations; retain exact source/binary identity, parameters, actual completed-operation counts and independent caller-origin, FIFO, byte, capability, foreign-selector and final-disposal checks. Each sample uses one owned process and explicit inputs, deadlines and output budgets; no VM, network, signing credential or unrelated-process authority is required for measurement. Integration must retain complete process status and raw evidence, reject incomplete or mismatched samples, and report comparisons only for equivalent source/workload contexts. Build once per source into distinct targets and run uncontended before/after samples. Host measurements do not establish guest or self-host-build throughput. Registry and endpoint-binding scans are hypotheses until measured; any optimization must preserve exact identity, allocation-safe publication and accounted revocation/terminal cleanup. Ordinary functional tests gain no timing-ratio assertion. Every sample checks exact bytes and SEND capability bindings, authenticated origin, foreign selection, duplicate completion refusal, FIFO position reuse, one-time collection and final invocation disposal. Numeric search positions remain non-authoritative: every fast path must still check current capability type/rights, endpoint identity/in-use state and exact CSpace/slot binding, with complete lookup on a stale hint. Original workload bytes and sample policy must remain unchanged for comparison. ## Authority, security and reliability Capture authority is explicit, finite and retired after success or failure. Exclude keys, capability contents, consent input and arbitrary process memory; diagnostic access must not grant control or namespace access. Bind PCs to exact process generations and module identities, and keep wall time, CPU time, blocking and samples distinct. Apply the priority order: security, reliability, then performance. Keep suspected vulnerabilities in the repository’s restricted SECURITY.md reporting channel. ## Acceptance criteria <details> <summary>Checklist</summary> - [ ] Inventory existing READY/command timers and private QMP/PC evidence; define which collectors are reusable, incomplete, obsolete or unsafe to retain. - [ ] Specify a versioned profile format with source/image/config hashes, toolchain, hardware/QEMU/acceleration/CPU/memory/worker settings, time domain, capture interval, exact process generation and module identity. - [ ] Define explicit capture authority, lifecycle and permitted fields; exclude key material, capability contents, consent input and arbitrary process memory. Diagnostics must not confer process control or namespace access. - [ ] Add a bounded host collector with hard/silence limits, owned PID cleanup, interruption handling, raw-log preservation and true failure status. - [ ] Implement exact image/ELF/PT_LOAD/source binding and symbolization; reject stale maps and ambiguous userspace addresses without process/module identity. Do not infer process ownership from an overlapping virtual address. - [ ] Add authenticated, opt-in guest timing/counter hooks where host samples cannot attribute a cost; compile/runtime gate them appropriately and prove their absence from clean release images when disabled. - [ ] Record boot service stages: image resolution/read, relocation, mapping/materialization, process start, startup delivery, initialization, publication, cleanup, final READY and full shell-caret display. - [ ] Record command stages: Enter acceptance, parse/expand/resolve, authority request/decision, file reads, link/load, spawn/scheduling, first execution/output, output drain, exit/wait and fresh caret. - [ ] Measure IPC count/bytes/waits, runnable versus blocked time, context switches, mapping scans/batches, page allocation/faults, VFS/provider fanout, block request size/LBA, cache hits/misses and TTY/framebuffer work. - [ ] Bound buffers through explicit resource policy, report dropped records, clock discontinuities and partial captures; never label incomplete data a complete profile or fabricate a percentage of total CPU time. - [ ] Implement reproducible reports: service timeline, command critical path, per-component counts, latency distributions and symbol/sample summaries. Separate wall time, CPU time, blocking time and sampling observations. - [ ] Add deterministic parser/attribution/error fixtures, malformed/stale identity tests, dropped-record tests and collector cancellation/timeout tests. - [ ] Add VM profiler tests that identify known synthetic CPU, wait and I/O workloads, reject cross-realm capture and verify all capture authority is retired. Compare profiled versus unprofiled overhead without subtracting it from acceptance measurements. - [ ] Establish controlled cold/warm startup and build baselines with isolated repeats, explicit resource limits and confidence/variance reporting. Preserve the stricter carried [integration#6](https://git.erikinkinen.fi/erix/integration/issues/6) startup thresholds without moving endpoints. - [ ] Rank evidenced bottlenecks by measured end-to-end impact; create individual [integration#2](https://git.erikinkinen.fi/erix/integration/issues/2) findings with one causal hypothesis, owner and acceptance test. - [ ] Optimize the highest-impact proven path, then reprofile and repeat: inspect loader/provider fanout, mapping/process startup, event waits, build dependency scheduling and terminal delivery according to evidence, not guesses. - [ ] For each optimization, prove unchanged authorization, revocation, permissions, object identity, failure ordering, output semantics and fairness; retain strict tests and isolated same-configuration before/after measurements. - [ ] Add reviewed startup/build throughput and resource regression budgets to CI, using dedicated uncontended performance jobs and calibrated full-build deadlines. Ordinary functional jobs still have forced stall limits. - [ ] Remove superseded diagnostic hooks/helpers and document how to capture, interpret and reproduce profiles. Keep only evidence-backed tooling. For each implementation slice, retain actual formatting, strict Clippy, unit/doctest and warning-denied build results for all altered Rust repositories and valid configurations. Add relevant runtime VM coverage, monitor older unit/VM regressions in exact-head CI, and update canonical component documents and affected technical-manual/API material. Every authored code file must remain below 1,000 physical lines, with meaningful inline documentation and missing_docs enforcement in Rust crates. </details> ## Alternatives and tradeoffs A raw address can overlap several userspace images, and sample counts are not CPU shares. Reuse sound host observers first; add authenticated guest hooks only where they resolve a real attribution gap and prove their absence when disabled. ## Tracking and rollout Track bounded implementation and measured fixes in the owning component issues, with each result linked from its review. Keep actual status in comments and check a criterion only when its complete scope is established. Native lifecycle captures and host invocation measurements do not establish either EriX guest build generation or the external Rust/LLVM guest rebuild. <details> <summary>Related component issues and contracts</summary> | Component | Links | | --- | --- | | bootloader | [PR 3](https://git.erikinkinen.fi/erix/bootloader/pulls/3) | | docs | [issue 11](https://git.erikinkinen.fi/erix/docs/issues/11), [PR 4](https://git.erikinkinen.fi/erix/docs/pulls/4) | | exsh | [issue 4](https://git.erikinkinen.fi/erix/exsh/issues/4), [issue 9](https://git.erikinkinen.fi/erix/exsh/issues/9), [compiler-test-profiling.md](https://git.erikinkinen.fi/erix/exsh/src/commit/ef26d89c45e7b65534e8157df0307edd36b73132/docs/compiler-test-profiling.md) | | integration | [issue 1](https://git.erikinkinen.fi/erix/integration/issues/1), [issue 18](https://git.erikinkinen.fi/erix/integration/issues/18), [issue 20](https://git.erikinkinen.fi/erix/integration/issues/20), [issue 3](https://git.erikinkinen.fi/erix/integration/issues/3), [issue 30](https://git.erikinkinen.fi/erix/integration/issues/30), [issue 31](https://git.erikinkinen.fi/erix/integration/issues/31), [issue 39](https://git.erikinkinen.fi/erix/integration/issues/39), [issue 40](https://git.erikinkinen.fi/erix/integration/issues/40), [issue 45](https://git.erikinkinen.fi/erix/integration/issues/45), [issue 47](https://git.erikinkinen.fi/erix/integration/issues/47), [issue 54](https://git.erikinkinen.fi/erix/integration/issues/54), [issue 65](https://git.erikinkinen.fi/erix/integration/issues/65), [issue 69](https://git.erikinkinen.fi/erix/integration/issues/69), [issue 70](https://git.erikinkinen.fi/erix/integration/issues/70), [issue 71](https://git.erikinkinen.fi/erix/integration/issues/71), [issue 72](https://git.erikinkinen.fi/erix/integration/issues/72), [issue 73](https://git.erikinkinen.fi/erix/integration/issues/73), [issue 74](https://git.erikinkinen.fi/erix/integration/issues/74), [issue 75](https://git.erikinkinen.fi/erix/integration/issues/75), [issue 76](https://git.erikinkinen.fi/erix/integration/issues/76), [issue 77](https://git.erikinkinen.fi/erix/integration/issues/77), [issue 78](https://git.erikinkinen.fi/erix/integration/issues/78), [issue 79](https://git.erikinkinen.fi/erix/integration/issues/79), [issue 80](https://git.erikinkinen.fi/erix/integration/issues/80), [issue 81](https://git.erikinkinen.fi/erix/integration/issues/81), [issue 82](https://git.erikinkinen.fi/erix/integration/issues/82), [issue 83](https://git.erikinkinen.fi/erix/integration/issues/83), [issue 84](https://git.erikinkinen.fi/erix/integration/issues/84), [issue 91](https://git.erikinkinen.fi/erix/integration/issues/91), [issue 96](https://git.erikinkinen.fi/erix/integration/issues/96), [issue 97](https://git.erikinkinen.fi/erix/integration/issues/97), [issue 98](https://git.erikinkinen.fi/erix/integration/issues/98), [issue 99](https://git.erikinkinen.fi/erix/integration/issues/99), [PR 12](https://git.erikinkinen.fi/erix/integration/pulls/12), [tcg-profiling.md](https://git.erikinkinen.fi/erix/integration/src/commit/4fa27f942bc2c6511eedceb5b20b8212f2bc94b0/docs/tcg-profiling.md) | | kernel | [issue 22](https://git.erikinkinen.fi/erix/kernel/issues/22), [issue 23](https://git.erikinkinen.fi/erix/kernel/issues/23), [issue 24](https://git.erikinkinen.fi/erix/kernel/issues/24), [issue 25](https://git.erikinkinen.fi/erix/kernel/issues/25), [issue 26](https://git.erikinkinen.fi/erix/kernel/issues/26), [issue 27](https://git.erikinkinen.fi/erix/kernel/issues/27), [issue 28](https://git.erikinkinen.fi/erix/kernel/issues/28), [issue 29](https://git.erikinkinen.fi/erix/kernel/issues/29), [issue 30](https://git.erikinkinen.fi/erix/kernel/issues/30), [issue 31](https://git.erikinkinen.fi/erix/kernel/issues/31), [issue 32](https://git.erikinkinen.fi/erix/kernel/issues/32), [issue 33](https://git.erikinkinen.fi/erix/kernel/issues/33), [issue 34](https://git.erikinkinen.fi/erix/kernel/issues/34), [issue 35](https://git.erikinkinen.fi/erix/kernel/issues/35), [issue 36](https://git.erikinkinen.fi/erix/kernel/issues/36), [issue 37](https://git.erikinkinen.fi/erix/kernel/issues/37), [issue 38](https://git.erikinkinen.fi/erix/kernel/issues/38), [issue 39](https://git.erikinkinen.fi/erix/kernel/issues/39), [issue 40](https://git.erikinkinen.fi/erix/kernel/issues/40), [issue 41](https://git.erikinkinen.fi/erix/kernel/issues/41), [issue 42](https://git.erikinkinen.fi/erix/kernel/issues/42), [PR 3](https://git.erikinkinen.fi/erix/kernel/pulls/3), [owned-address-spaces.md](https://git.erikinkinen.fi/erix/kernel/src/commit/5e818e8b6d1d6da7698cfff34820d0aa9de8c05a/docs/owned-address-spaces.md) | | lib-cstd | [PR 3](https://git.erikinkinen.fi/erix/lib-cstd/pulls/3) | | meta | [issue 2](https://git.erikinkinen.fi/erix/meta/issues/2) | | procd | [issue 5](https://git.erikinkinen.fi/erix/procd/issues/5) | </details>
erikinkinen changed title from [FEATURE] [P04] Maintain bounded profiling and eliminate evidenced startup and build bottlenecks to [FEATURE] Maintain bounded profiling and eliminate evidenced startup and build bottlenecks 2026-09-12 08:02:02 +02:00
Author
Owner

The native runners now accept the bounded TCG selector. Captures for all six original lifecycle scenarios have complete footers and no missed blocks, with original VM gates preserved. Exact packaged-code reports retain unknown and ambiguous candidates. Earlier baseline firmware equivalence was not captured, so these measurements establish no elapsed-cost ratio or lifecycle optimization. The fuller startup/build profiling and optimization criteria remain open.

[The native runners](https://git.erikinkinen.fi/erix/integration/commit/6e05590754d311a54bf7db4aae24b522ab735c6e) now accept the bounded TCG selector. Captures for all six original lifecycle scenarios have complete footers and no missed blocks, with original VM gates preserved. Exact packaged-code reports retain unknown and ambiguous candidates. Earlier baseline firmware equivalence was not captured, so these measurements establish no elapsed-cost ratio or lifecycle optimization. The fuller startup/build profiling and optimization criteria remain open.
Author
Owner

906cfb5a removes temporary token slices and repeated parsing of unchanged LLVM lines. Each verifier still parses its own original input independently. Across nine modules, complete derived bytes, mappings and semantic lists match.

Plain wall time in the paired experiment fell from 97.66 to 91.19 seconds; the instrumented variant fell from 100.51 to 99.38 seconds. These were single baseline-first pairs on the host, with some modules slower and no observer overhead subtracted. The result supports the narrow redundant-work removal; complete compiler and guest-build requirements remain open.

[906cfb5a](https://git.erikinkinen.fi/erix/exsh/commit/906cfb5af32c0befee27126ae1d8847afe6647ee) removes temporary token slices and repeated parsing of unchanged LLVM lines. Each verifier still parses its own original input independently. Across nine modules, complete derived bytes, mappings and semantic lists match. Plain wall time in the paired experiment fell from 97.66 to 91.19 seconds; the instrumented variant fell from 100.51 to 99.38 seconds. These were single baseline-first pairs on the host, with some modules slower and no observer overhead subtracted. The result supports the narrow redundant-work removal; complete compiler and guest-build requirements remain open.
Author
Owner

I cancelled four superseded runs that had not started any jobs: 2666, 2669, 2672, and 2675. All twelve jobs still had no assigned task after cancellation.

The active older run continues. The complete workflow for current source 7769fb52 remains queued; its test catalog is unchanged.

I cancelled four superseded runs that had not started any jobs: [2666](https://git.erikinkinen.fi/erix/integration/actions/runs/2666), [2669](https://git.erikinkinen.fi/erix/integration/actions/runs/2669), [2672](https://git.erikinkinen.fi/erix/integration/actions/runs/2672), and [2675](https://git.erikinkinen.fi/erix/integration/actions/runs/2675). All twelve jobs still had no assigned task after cancellation. The active older run continues. The complete workflow for [current source 7769fb52](https://git.erikinkinen.fi/erix/integration/actions/runs/2678) remains queued; its test catalog is unchanged.
Author
Owner

The full VM run finished with 448 passes and 77 failures across all 525 scenarios; Rust and Markdown passed. Of the failed cases, 63 print the same Vfsd fault. The matching binary contains large nested stack frames, so the next fix focuses on Vfsd’s stack use. The remaining fourteen failures still need separate diagnosis.

The [full VM run](https://git.erikinkinen.fi/erix/integration/actions/runs/2684) finished with 448 passes and 77 failures across all 525 scenarios; Rust and Markdown passed. Of the failed cases, 63 print the same Vfsd fault. The matching binary contains large nested stack frames, so the next fix focuses on [Vfsd’s stack use](https://git.erikinkinen.fi/erix/vfsd/issues/33). The remaining fourteen failures still need separate diagnosis.
Author
Owner

CI2714 on ede995b7 finishes 491 passed / 34 failed / 525 VM cases. All twelve workers have complete summaries; the Rust and Markdown jobs pass.

The original log shows 26 serial-progress stalls at 45 seconds, four native call failures with slot 2637/status 3, one no-pager fault and two guest exits with status 41. One remaining failed name has no matching case-start record, so its cause remains unassigned. These groups describe the observed failures; they do not establish a common cause.

Launchd's resolver diagnostic tags are published and pass CI. Catalog adoption is being prepared for the explicit traced investigation of the native launch failure.

[CI2714](https://git.erikinkinen.fi/erix/integration/actions/runs/2714) on `ede995b7` finishes **491 passed / 34 failed / 525 VM cases**. All twelve workers have complete summaries; the Rust and Markdown jobs pass. The original log shows 26 serial-progress stalls at 45 seconds, four native call failures with slot 2637/status 3, one no-pager fault and two guest exits with status 41. One remaining failed name has no matching case-start record, so its cause remains unassigned. These groups describe the observed failures; they do not establish a common cause. Launchd's resolver diagnostic tags are published and pass CI. Catalog adoption is being prepared for the explicit traced investigation of [the native launch failure](https://git.erikinkinen.fi/erix/integration/issues/135).
Author
Owner

Added an explicit snapshot profiler and changed object preparation to copy first, then persist each file and directory through its sealed identity. On the same 159-object local workload, mean snapshot time fell from 1.681 s to 1.180 s (29.8%). Directory flushes fell from 238 to 120; all 159 file flushes remain.

The report keeps actual phase costs and failures. It does not change source admission or process deadlines. Broader build measurements remain open. See the implementation, catalog adoption, and profiling guide.

Added an explicit snapshot profiler and changed object preparation to copy first, then persist each file and directory through its sealed identity. On the same 159-object local workload, mean snapshot time fell from 1.681 s to 1.180 s (29.8%). Directory flushes fell from 238 to 120; all 159 file flushes remain. The report keeps actual phase costs and failures. It does not change source admission or process deadlines. Broader build measurements remain open. See the [implementation](https://git.erikinkinen.fi/erix/integration/commit/e1350dac4837fbf344056fa5cbf513cfcabb008b), [catalog adoption](https://git.erikinkinen.fi/erix/integration/commit/f3d5b32b92ab8e5b33a6b76dc9f474ca2f1b8d04), and [profiling guide](https://git.erikinkinen.fi/erix/integration/src/commit/e1350dac4837fbf344056fa5cbf513cfcabb008b/docs/source-snapshot-performance.md).
erikinkinen added this to the Alpha 1 milestone 2026-10-30 09:34:08 +01:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erix/integration#3
No description provided.