fix: format debt, GitLab CI wiring, and the (never-actually-run) live-download install path - #71
Merged
Merged
Conversation
226 files were unformatted, leaving the Mustfile's 'format' check (just fmt -> mix format --check-formatted) permanently red — nothing enforced formatting so it silently drifted. Pure mix format output, no manual edits; compile --warnings-as-errors and the full test suite (807 tests) are unaffected. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
… run Every mix-*/cargo-* job has been permanently dormant: mix.exs lives in opsm_ex/, not the repo root, and the estate's Rust code is nine independent first-party crates (opsm-ui/tui, two opsm_ex/native NIFs, six services/* microservices) — there is no root Cargo.toml for 'rules: exists: Cargo.toml' to ever match. - mix-* jobs (mix-audit, mix-format, credo, mix-test, mix-build): rules point at opsm_ex/mix.exs; scripts cd into opsm_ex first. - cargo-* jobs (cargo-audit, cargo-deny, rustfmt, clippy, cargo-test, cargo-build): rules list all nine crates' Cargo.toml paths (rules: exists needs literal paths, not variable expansion); scripts loop over the new $RUST_CRATE_DIRS variable, running each command per-crate. opsm_ex/deps/** (vendored Mix dependency cache) is deliberately excluded. - cache/artifact paths updated for the per-crate target/ layout. Verified: mix-format's own check (mix format --check-formatted, now green after the prior commit) and just toolchain-check both pass; YAML parses clean. The cargo-* loop bodies are not runnable in this sandbox (no GitLab remote, no network for cargo install) — reviewed by inspection against the confirmed crate list, same pattern as the already-working per-crate GitHub Actions job (nif-build.yml). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
…1.16.0
Discovered while wiring the live-download CI job (which needed a real
nickel install for the first time): 84 of 95 .ncl files under runtime/
failed to even typecheck against the canonical nickel 1.16.0 pinned in
opsm.toml [runtime]. opsm runtime install has never actually worked
against a real nickel binary — every prior test only reached the
{:error, {:no_plugin, _}} / {:error, {:nickel_not_installed, _}} paths.
Two independent bugs, both traced to a single shared file:
- runtime/contract/runtime-plugin.ncl had three fields with
'| default = X | doc "..."' annotation ordering. Nickel 1.16's
grammar requires doc before default in a field's annotation chain;
the reverse order is a parse error, and every plugin file imports
this contract, so the failure was transitive to all 84 dependents.
Fixed by swapping the two annotations (verified minimal repro).
- Every plugin file destructured the import as
'let { RuntimePlugin, .. } = import "../contract/runtime-plugin.ncl"',
but the contract file's final expression is the bare contract value
itself (a module, not a record wrapping it) —
so the destructuring pattern could never match. Changed to a plain
'let RuntimePlugin = import ... in' binding across all 83 affected
files (one-line mechanical change, same substitution everywhere).
Verified: all 95 .ncl files now nickel typecheck clean; all 94 files
under runtime/core/ (the actual plugin catalog) fully nickel-export to
JSON — the exact call Manager.load_plugin makes. Confirmed end-to-end
via a live zig install (see following commit for the http.ex fix that
was still needed to get a completed install).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
… streamed downloads
Two silent option-forwarding bugs in Opsm.Verified.Http, found while
finishing the runtime install path (following the nickel corpus fix):
- do_get/2 and do_post/2 read Keyword.get(opts, :timeout, 30_000), but
every call site in the codebase (~480 occurrences) passes
receive_timeout: — Req's actual option name. Every custom timeout
configured anywhere in OPSM has been silently discarded in favor of
the 30s default. Now reads :receive_timeout first, falling back to
the old :timeout key for compatibility, then the same default.
- do_get/2 never forwarded opts[:into] to Req.get/2 at all.
Runtime.Manager.download_archive/1 calls
VerifiedHttp.get(url, into: File.stream!(dest)) expecting Req to
stream the response body straight to disk; because :into was
dropped, Req buffered the full body in memory, returned {:ok, _},
and download_archive treated that as success — but dest was never
written. extract_archive/3 then handed tar a nonexistent file
({"", 2} from tar's own "Cannot open" exit). This is why a live zig
install still failed after the nickel fix.
Verified against a real 47MB zig 0.13.0 download from ziglang.org:
integration_test.exs (--include external_api --include live_download)
now passes 15/15 (0 failures) with a clean cache, including install,
which/1, current_version/1, remove/1, and reinstall idempotency. Full
default suite (807 tests) and mix format --check-formatted unaffected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
The workflow's own header comment has claimed :live_download tests are
"workflow_dispatch only" since it was written, but no job ever ran
them — workflow_dispatch only enabled the manual trigger, it was never
wired to a job scoped to those tests. Adds a live-download job, gated
on `github.event_name == 'workflow_dispatch'`, that:
- Installs a Rust toolchain (pinned to 1.97.0, matching opsm.toml
[runtime]) and cargo-installs nickel-lang-cli 1.16.0 — confirmed to
exist at that exact version on crates.io. Manager.install shells out
to the nickel binary to evaluate runtime/core/*.ncl plugins; without
it every install fails with {:nickel_not_installed, tool}.
- Runs integration_test.exs with --include external_api --include
live_download, exercising a real download+extract+install+remove
cycle (this is what the two prior commits in this branch actually
fixed and verified locally: 84 broken .ncl files, and a
VerifiedHttp bug that silently dropped streamed downloads).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
hyperpolymath
marked this pull request as ready for review
July 17, 2026 06:19
hyperpolymath
added a commit
that referenced
this pull request
Jul 21, 2026
Follow-up to #71, closing the gap I flagged there: the GitLab `cargo-*` jobs were "reviewed by inspection, not runnable in the sandbox." I have a working Rust toolchain here, so I actually ran all nine first-party crates through `cargo build`/`test`/`fmt`. Result: **8 of 9 build clean once fixed, 1 is a known-optional non-builder**. Three commits: ## 1. `fix(selur): add missing tokio AsyncReadExt import` (`02ff456`) `services/selur` has **never compiled**. `sign_image` had a function-local `use tokio::io::AsyncReadExt;`, but `generate_keypair` (a second call site) also does `file.take(..).read_to_string(..)` without the trait in scope — so `.take()` resolved to `Iterator::take` on `tokio::fs::File` and failed (`E0599`/`E0282`). The dormant GitLab jobs never caught it, and `nif-build.yml` doesn't cover `services/`. Hoisted the import to module scope, dropped the redundant local one, removed `mut` from two read-only bindings. `cargo build --release` + `cargo test` pass. ## 2. `style(rust): cargo fmt all first-party crates` (`1d743b2`) The `rustfmt` job (not `allow_failure`) runs `cargo fmt -- --check` per crate; all 8 failed because the code was never fmt-checked while the job was dormant — same story as the `mix format` sweep in #71, for Rust. Pure `cargo fmt` output, no logic changes. ## 3. `fix(gitlab-ci): quarantine the non-building quic_transport NIF` (`89a4c1b`) `opsm_ex/native/quic_transport` does **not** compile: pinned `h3 0.0.6` / `h3-quinn 0.0.7` predate `quinn 0.11`'s now-private `StreamId` field, and with no committed `Cargo.lock` the fresh resolve breaks. It's an **optional** QUIC/HTTP3 NIF — `Opsm.Transport.QuicNif` swallows a failed `:erlang.load_nif` and falls back to HTTP/2 (`nif_loaded?` is hard-coded `false`), so the Elixir side (807 tests) runs fine without it, and no CI builds it today. Rather than sink a full pre-1.0 `h3`-ecosystem migration into optional aspirational code, I removed it from `$RUST_CRATE_DIRS` and the `rules:`/artifacts lists, with a `QUARANTINED` header note stating the un-quarantine condition (migrate the `h3`/`h3-quinn`/`quinn` triple to a compatible set + commit a lockfile). **Flagged as a follow-up decision for you** — see below. ## Verified (real Rust 1.9x toolchain, this environment) - All 8 kept crates: `cargo build --release` ✓, `cargo fmt -- --check` ✓; `checky-monkey` tests 4/4, others build clean (0 tests defined) - `.gitlab-ci.yml` parses; only 2 `quic_transport` mentions remain (both the header note) - `just toolchain-check` ✓ ## For your call (not done here) - **quic_transport migration**: the `h3` crate is pre-1.0 and churns hard (`recv_data` now returns `impl Buf`, send-side API changes, etc.) — a real migration, not a version bump. Worth doing only if QUIC/HTTP3 transport is on the near-term roadmap; otherwise the quarantine + HTTP/2 fallback is the honest steady state. - The `cargo-audit`/`cargo-deny` job *bodies* still aren't exercised here (installing those tools from crates.io for 8 crates wasn't worth the sandbox time); the build/test/fmt jobs are now genuinely verified. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C --- _Generated by [Claude Code](https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
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
2 participants
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.
Three items from the sitrep backlog, done as five separate commits. The last two turned into a real chain of discovery — wiring a CI job to exercise the live-download path surfaced that
opsm runtime installhas, as far as I can tell, never actually completed a real install against a real nickel binary, for three independent reasons stacked on top of each other.1.
style: run mix format across the tree(ef87e0b)226 files were unformatted, leaving the Mustfile's
formatcheck permanently red with nothing enforcing it. Puremix formatoutput, no manual edits.2.
fix(gitlab-ci): fix root-relative rules:exists paths(10588f8)Every
mix-*/cargo-*job in.gitlab-ci.ymlwas permanently dormant:mix.exslives inopsm_ex/, and the Rust code is nine independent first-party crates (opsm-ui/tui, twoopsm_ex/nativeNIFs, sixservices/*) — no rootCargo.tomlforrules: exists:to ever match.mix-*jobs now targetopsm_ex/;cargo-*jobs loop over a$RUST_CRATE_DIRSlist. Verified: YAML parses,mix format's own check (now green) confirms the job would pass,just toolchain-checkpasses.3–5. The live-download chain
Task was "add a
workflow_dispatchjob for the live-download tests" (8d0e373). Getting a real nickel binary onto a runner to test it locally surfaced two more bugs, both now fixed and verified against a real 47MB zig install:fix(nickel): unbreak the entire runtime plugin corpus(3f67294) — 84 of 95.nclfiles failed to even typecheck against the canonical nickel 1.16.0 pinned inopsm.toml [runtime]. All 84 failures traced to one shared file, two bugs:runtime-plugin.nclhad three fields withdefaultbeforedocin the annotation chain (nickel 1.16's grammar requires the reverse — confirmed via minimal repro), and every plugin destructured its import aslet { RuntimePlugin, .. } = import ...against a module whose export is the bare contract value, not a record — the destructuring could never match. Both are one-line-per-field-or-file mechanical fixes. All 95 files now typecheck; all 94runtime/core/*.nclplugins now fullynickel exportto JSON.fix(verified-http): forward receive_timeout and :into to Req(f9b926e) — with nickel now working, the install still failed:Opsm.Verified.Http.do_get/2never forwardedopts[:into]toReq.get/2at all, soManager.download_archive/1'sVerifiedHttp.get(url, into: File.stream!(dest))silently buffered the response in memory instead of streaming to disk, returned{:ok, _}, andextract_archive/3got handed a file that was never written. Also found in the same function:do_get/do_postreadopts[:timeout], but every call site in the codebase (~480 occurrences) passesreceive_timeout:— Req's actual option name — so every custom timeout anywhere in OPSM has been silently discarded in favor of the 30s default. Both fixed.feat(e2e): add a live-download job(8d0e373) — the actual dispatch job: Rust toolchain (pinned 1.97.0) →cargo install nickel-lang-cli --version 1.16.0(confirmed to exist at that version on crates.io) →mix test integration_test.exs --include external_api --include live_download.Verified (OTP 28.3.1 / Elixir 1.19.5, real
nickel-lang-cli1.16.0 built from source)mix format --check-formatted— cleanmix compile --warnings-as-errors— cleansh scripts/toolchain.sh check— clean.nclfiles nickel-typecheck; all 94runtime/core/*fully exportmix test test/opsm/runtime/integration_test.exs --include external_api --include live_download— 15/15, 0 failures, with a cold cache: real download from ziglang.org, extract,which/1,current_version/1,remove/1, reinstall idempotency.gitlab-ci.yml/.github/workflows/e2e.ymlYAML parse cleanThe GitLab
cargo-*job bodies aren't runnable in this sandbox (no GitLab remote, and installingcargo-audit/cargo-denyfrom crates.io for 9 crates wasn't worth the time here) — reviewed by inspection against the confirmed crate list and the already-working per-crate pattern innif-build.yml.🤖 Generated with Claude Code
https://claude.ai/code/session_01Kq24sZCEohSrNFXSuEuz6C
Generated by Claude Code