fix(repo): prevent daemon write replay after ack loss - #3843
Open
KooshaPari wants to merge 364 commits into
Open
fix(repo): prevent daemon write replay after ack loss#3843KooshaPari wants to merge 364 commits into
KooshaPari wants to merge 364 commits into
Conversation
Terminal-Forge scanline + mint-cursor identity demo. 5s loop rendered via playwright + ffmpeg from CSS-animated SVG. Co-authored-by: Phenotype Agent <agent@phenotype.ai>
… addition) (#86) Upstream fork annotation: this is a Phenotype-org addition. It is NOT present in tailcallhq/forgecode. Adds `crates/forge_display/src/theme.rs` with the Terminal-Forge family palette per the vision-pillar L96 lock-in: tf-deep-charcoal #0d1117 (panel / background) tf-deep-charcoal-2 #161b22 (nested panel / surface) tf-amber-crt #ffb454 (forgecode dominant) tf-synthwave #ff7edb (accent) tf-mint-prompt #7ee787 (success / prompt) Exports: - `TerminalForgePalette` (struct with hex-string fields, Copy) - `terminal_skin_from_theme(&TerminalForgePalette) -> MadSkin` helper that wires the palette into termimad's MadSkin so MarkdownFormat (and downstream consumers) inherit Terminal-Forge identity without per-call-site color literals. Wires `pub mod theme;` into forge_display/src/lib.rs and re-exports the public symbols. Tests (+4): - terminal_forge_palette_matches_tokens_css_family_3: pins the L96 decision (any drift must come from coordinated tokens.css update) - parse_hex_accepts_canonical_terminal_forge_tokens: 5 hex inputs - parse_hex_rejects_malformed_input: 4 negative cases - terminal_skin_from_theme_produces_termimad_skin: smoke-test MadSkin wire-up Local gate: cargo test -p forge_display → 29 tests pass, 0 fail (was 25 before this PR). Co-authored-by: Phenotype Agent <agent@phenotype.ai>
…#91) Co-authored-by: Phenotype Agent <agent@phenotype.ai>
* feat(packaging): forge-dev binary + install guide (T18) Add a feature-gated forge-dev [[bin]] target alongside the existing forge binary so operators can install the Phenotype build side-by-side with the upstream release. The forge-dev target is opt-in via the empty dev-binary feature flag, keeping the default cargo install behavior unchanged. - crates/forge_main/Cargo.toml: declare [features] dev-binary = [] and a new [[bin]] name = "forge-dev" with required-features = ["dev-binary"]. The shared src/main.rs path produces the documented dual-target cargo warning, which is expected. - docs/forge-dev-install.md: 2-paragraph install guide covering the side-by-side rationale, the --features dev-binary gate, and the cargo install command from the git fork. cargo check -p forge_main defaults to building only the forge bin (no warning). cargo check -p forge_main --features dev-binary builds both forge and forge-dev bins and finishes clean. Refs: docs/packaging/FORGE_DEV_PACKAGING.md §2.1, §6.5. Co-Authored-By: ForgeCode <noreply@forgecode.dev> * audit(v38): forgecode C03 scorecard — Agent Readiness L30.1-L30.12 Authored by parent (not codex) because the codex standalone dispatch stalled at the read-evidence phase (log 161KB of source-file reads, no file written, no proc visible after 3 minutes idle). Same pattern as the earlier Tracera failure. DEBUG: every pillar cites real file:line refs (AGENTS.md:1-246, CLAUDE.md:1-133, 2888 #[test] markers, etc.). OPTIMIZE: 12 pillars × ~10 lines each ≈ 120 lines total. ENRICH: every gap has explicit effort (S/M) for follow-up. CLUSTER_TOTAL score=18/36 pct=50% grade=D * audit(v38): forgecode C05 scorecard — 5/30 F * audit(v38): forgecode C06 scorecard — 12/24 C * audit(v38): forgecode C07 scorecard — 16/30 C * audit(v38): forgecode C04 scorecard — Security (18/30) * audit(v38): forgecode C08 scorecard — Eval Coverage (8/30 ) * audit(v38): forgecode C09 scorecard — Accessibility+UX (23/42 ) * audit(v38): forgecode C10 scorecard () * audit(v38): forgecode C09 scorecard (soft_goal_delta: friction capture is not yet instrumented18/42 ) * audit(v38): forgecode C10 scorecard (10/33 ) * audit(v38): forgecode scorecard (audit/.lane-c11/C11.md:7/30 ) Parent-authored (continuation of v38 scorecard sweep). * audit(v38): forgecode C03.md refinement (25/36 ) Re-scan refinement with additional evidence. CLUSTER_TOTAL marker preserved. * audit(v38): forgecode C05.md refinement (5/30 ) Re-scan refinement with additional evidence. CLUSTER_TOTAL marker preserved. --------- Co-authored-by: ForgeCode Bot <forgecode-bot@users.noreply.github.com> Co-authored-by: ForgeCode <noreply@forgecode.dev> Co-authored-by: Phenotype Agent <agent@phenotype.ai>
…se policy (#92) * renames(helioslite): add canonical binary + publish-surface metadata + provenance docs * renames(helioslite): update channel + doctor banner + packaging matrix + landing/Caddy * ci(helioslite): add main branch-protection config (file-only) * renames(helioslite): redirect banner + env-var legacy fallback + tombstone * docs(helioslite): add redirect-chain + install-time tombstone sections
…glob_pattern (#93) 3 utility modules adapted from the substrate gateway library (L123 + L131) into forge_infra. These primitives are useful for forge's existing retry / watcher / tool-pattern code paths without depending on the substrate workspace. - backoff: Fixed / Linear / Exponential retry schedule capped at max. 4 unit tests including capped-at-max regression guard. Suitable for MCP server retries and provider endpoint backoff. - cyclic_check: CRC-8 (poly 0x07), CRC-16-CCITT (poly 0x1021), CRC-32-IEEE (poly 0xEDB88320). 4 unit tests with the standard '123456789' known-vector vectors. Useful for file integrity checks on tool output + JSON RPC framing. - glob_pattern: ASCII glob matcher with * ? [abc] [!abc]. 8 unit tests. Useful for file include/exclude lists, watcher path filters, and tool-pattern allowlists across the MCP and tool-macro layers. All three registered as in crates/forge_infra/src/lib.rs. cargo test -p forge_infra shows 16 new tests passing. Ref: l133-forgecode-utility-cross-pollination Co-authored-by: Phenotype Agent <agent@phenotype.ai>
- get_all_provider_models now gracefully handles individual provider failures instead of failing the entire /models endpoint when one provider returns a 500. Individual errors are logged as warnings; the operation only fails if every configured provider fails. - Added /providers remove TUI command (AppCommand::ProviderRemove) that lets users interactively select and remove a configured provider from the REPL, matching the behavior of the existing /logout command.
Re-lands Primary3090/Legacy1080 lane intent lost during GH restore. Includes WSL and Windows llama.cpp CVD profile adapters for pheno-harness.
Pin workflow actions, tighten token permissions, fix trufflehog install, narrow cargo-deny to source checks, and address rustfmt/clippy/build failures. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Use a resolvable Swatinem/rust-cache commit SHA so nextest and lint jobs can start. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Replace go install with the official pinned TruffleHog action because v3.91.0 cannot be installed as a Go tool due upstream replace directives. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Use resolvable commit SHAs for install-action, cache, SLSA, and release-drafter pins; verify all pinned action refs through the GitHub commits API. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Use the TruffleHog Docker tag format expected by the pinned composite action. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Run cargo-deny source checks only and install cargo-nextest explicitly via pinned install-action. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Install Zig in Rust workspace CI lanes so forge_daemon build scripts can run during clippy, nextest, coverage, and advisory autofix. Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Skip the Zig daemon C core unless explicitly requested so normal workspace builds and Linux CI do not fail on Zig/libc translation issues. Keep dispatch failure explicit when the core is not enabled and fix forge_tui clippy output formatting.
Make proc-stat CPU parsing return 0.0 on malformed data instead of using ? in an f64-returning helper, and mark the Linux sysconf extern block unsafe for Rust 2024 workspace builds.
Co-Authored-By: ForgeCode <noreply@forgecode.dev>
Co-Authored-By: ForgeCode <noreply@forgecode.dev>
# Conflicts: # crates/forge_gpu/Cargo.toml # crates/forge_gpu/src/lib.rs
heliosdoctor --integrity-only runs a fast PRAGMA integrity_check on both the write and legacy databases without COUNT queries. Plain output prints a dedicated 'integrity check' block; porcelain/json add write_db/legacy_db lines. Includes two CLI parse tests. Also fixes the pre-existing forge_app test-gate compile break (E0046 from split-DB EnvironmentInfra extension): database_stats default impls added to test mocks in changed_files, command_generator, orch_runner, terminal_context, and user_prompt. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
* fix(ci): repair Rust 1.96 quality gates * fix(ci): refresh release workflow and safe snapshot hashing Co-Authored-By: ForgeCode <noreply@forgecode.dev> --------- Co-authored-by: KooshaPari <koosha@example.com> Co-authored-by: ForgeCode <noreply@forgecode.dev>
…h test, dedupe mock - forge_embed: normalize CRLF->LF at template registration so rendered output (and insta snapshots) is identical on Windows and Linux; git stores templates as LF, autocrlf checkouts were leaking CRLF into snapshots (command_generator, compact). - workspace_status: gate test_file_statuses with #[cfg(unix)] - the POSIX absolute-path semantics differ on Windows, matching the pattern established for fd.rs/path.rs tests. - orch_runner: drop duplicate database_stats left by the origin/main merge (E0201). - fmt: apply rustfmt to feature files (api, forge_api, command_generator, services, user_prompt, env, ui, app_config, forge_embed). Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
release build of 6c98c8c installed to Programs\heliosLite + Programs\Forge; version 2.10.7 confirmed; doctor/export results. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
test_file_statuses asserts Unix-rendered absolute paths (/workspace/...) and was never runnable on Windows; the forge_app test target did not compile here until the test-build blocker fixes. Same class as the fd.rs/path.rs cfg(unix) gates. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
…o, forge_main - database_stats mock impls: use async fn syntax (field_reassign / async_fn_in_trait) - env.rs: struct-update init for HeliosdoctorDbStats, collapse nested if-lets - conversation_repo: collapse if-lets, needless borrows, allow dead_code on QueryableByName ExportRow field - executor.rs: drop needless borrows in CRLF-tolerant comparisons - provider_repo / openai_responses: async fn database_stats mocks - ui.rs: remove unused std::fs import in tests Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
lib+bin split (lib exposes client+protocol so forge_repo can depend on it); daemon executes real upsert/update/delete SQL against the split-DB write path (single writer connection, batching kept); Windows named-pipe transport alongside the Unix socket; DELETE keys on the conversation_id PK (workspace predicate could silently no-op when cwd hashing diverges); JSON frames; drop unused bincode dep. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
…-in) Environment gains dbd_enabled() (FORGE_DBD_ENABLED) and dbd_socket_path() (FORGE_DBD_SOCKET); DaemonConversationRepository routes upsert/upsert_ref/update_parent_id/delete through forge_dbd with direct-diesel fallback on any daemon failure (trait surface unchanged, reads stay direct); ForgeRepo wires it when enabled — off by default, zero behavior change otherwise. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
test_create_with_nonexistent_absolute_path used a Unix-only "/this/..." fixture; is_absolute() is false for prefix-less paths on Windows, so canonicalize-fallback bailed and unwrap() panicked. Gate the Unix fixture and add a drive-qualified Windows twin (test-only change). Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Daemon (server.rs): exits cleanly after the last client disconnects and the idle timeout (default 300s) elapses, draining the write queue and unlinking the socket; active-connection tracking on both the Unix-socket and Windows named-pipe transports; new_with_idle for tests. Client (daemon_repo.rs): on the first routed write that cannot connect, spawns forge_dbd exactly once per process (FORGE_DBD_BIN or PATH lookup, inheriting FORGE_WRITE_DB_PATH), polls the socket ~2s, then falls back to the direct diesel path; a second failure never respawns. Environment (env.rs): dbd_bin_path() reads FORGE_DBD_BIN, mirroring the existing env-var pattern. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
ToolDescription reads description .md files at compile time via fs::read_to_string; on Windows autocrlf checkouts those carry CRLF while committed snapshots are LF, breaking snapshot determinism. Normalize to authored LF bytes, matching the forge_embed template fix. Verified: forge_repo --lib 369 passed (incl. catalog snapshot test). Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
heliosdoctor --integrity-only is PRAGMA-only at the forge_infra layer, but the runtime services infra is ForgeRepo<ForgeInfra>, whose EnvironmentInfra impl forwarded database_stats only. database_integrity therefore fell back to the trait default (full COUNT stats), so porcelain/json output under --integrity-only surfaced db_total counts instead of the cheap probe. ForgeServices had the same gap. Forward database_integrity in both wrappers and add a regression test that constructs ForgeRepo over a mock infra with distinguishable database_stats/database_integrity results. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
ForgeRepo/ForgeServices now forward database_integrity so heliosdoctor --integrity-only stays PRAGMA-only at runtime. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
* fix(helioslite): keep default home separate from Forge Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(snapshot): match current Forge conversation schema Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(snapshot): reuse equivalent later exports Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(snapshot): recover matching publish races Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(helioslite): route published destinations to verifier Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(ci): repair strict workspace quality gates Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(ci): satisfy strict workspace lint Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(helioslite): block symlinked snapshot destinations Co-Authored-By: ForgeCode <noreply@forgecode.dev> --------- Co-authored-by: KooshaPari <koosha@example.com> Co-authored-by: ForgeCode <noreply@forgecode.dev>
Co-authored-by: KooshaPari <koosha@example.com>
Co-authored-by: KooshaPari <koosha@example.com>
* fix(provider): retry transient OpenAI server errors Co-Authored-By: ForgeCode <noreply@forgecode.dev> * fix(provider): preserve OpenAI stream error classification Co-Authored-By: ForgeCode <noreply@forgecode.dev> --------- Co-authored-by: KooshaPari <koosha@example.com> Co-authored-by: ForgeCode <noreply@forgecode.dev>
Co-Authored-By: ForgeCode <noreply@forgecode.dev>
github-actions
Bot
added
type: fix
|
|
Co-Authored-By: ForgeCode <noreply@forgecode.dev>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Safety contract
Direct fallback remains available only after initial connect/spawn failure.
DbClient::sendtransport failures,Response::Error, and unexpected responses are explicit indeterminate errors, preventing duplicate writes.Verification
cargo test -p forge_repocargo clippy --all-targets --all-features -- -D warningscargo +1.96 fmt --checkgit diff --check