tests: faster dev loop + hpack dep dedup; misc hasql fixes

[tonyalaribe]

Summary

• make live-test-dev: lib + integration tests as one GHCi target via a new test-dev test-suite, with --match filtering. Compiles src/ + test/integration/ together so iteration on test code skips the library re-link step.
• hpack-includes/lib-deps.yaml: single source of truth for library dependencies, referenced from both library and tests.test-dev via !include. Eliminates the manual dep mirror (which had already drifted — vector-split was missing from test-dev, breaking the build after rebase). hpack 0.38 doesn't splice list-position !includes, so a handful of test-only deps (hspec, uuid-quasi, ki, pg-transact, aeson-pretty, process) also live in the shared file; the library declares them unused, silenced by the existing repo-wide -Wno-error=unused-packages.
• Follow-up fixes on top of the hasql-interpolate migration that landed in hasql-interpolate + record-dot migration: derived ENUM/CITEXT decoders #409 / Ingestion/derive bulk insert cap #410: bigint COUNT decoded as Int64 in BackgroundJobs.hs; doctest pin on length otelColumns; minor DeriveUtils adjustments.

Test plan

• hpack regenerates cleanly
• cabal repl monoscope (ghcid): All good, 109 modules
• cabal build test-dev: exit 0, 157 modules compiled and linked
• make live-test-dev on a real spec (e.g. TEST_MATCH=/MonitoringSpec/ make live-test-dev)
• CI green on the integration / unit / doctest jobs

[claude]

Review

The core ideas here — lib-deps.yaml as a single source of truth, the test-dev combined target for faster iteration, and the tmux pane-pinning rework — are all solid. A few things worth fixing before merge:

---

Bug: base16-bytestring duplicated in lib-deps.yaml

hpack-includes/lib-deps.yaml lists base16-bytestring twice: once near line 27 and again at line 137 (just before minio-hs). This is a carry-over from the same dup in the old package.yaml. hpack silently deduplicates at resolve time so it won't break the build, but the YAML is confusing and contradicts the stated goal of deduplication.

- - base16-bytestring   # remove the line 137 occurrence
  - minio-hs

---

Incomplete Int → Int64 fix

The PR correctly fixes jobStats/stuckJobs at lines 700–701, but the same Int/bigint mismatch exists in several nearby sections that weren't touched:

Line	Annotation	SQL
656	[(ProjectId, Text, Int)]	COUNT(*)::bigint
677	[(ProjectId, Text, Int)]	COUNT(*)::bigint
757	[(ProjectId, Text, Text, Int, Int)]	::bigint AS reported, ::bigint AS ingested

These will decode fine on 64-bit Linux (where Int is 64 bits) but the decoder contract is technically violated — hasql's int8 decoder returns Int64, not Int. Worth fixing in this PR since you're already in this file for the same reason.

---

live-test-dev missing from .PHONY

The new target isn't in the .PHONY declaration (line 244). If a file named live-test-dev ever exists in the repo root, make will silently skip running the recipe.

-.PHONY: ... tmux-pin-here tmux-unpin ...
+.PHONY: ... live-test-dev tmux-pin-here tmux-unpin ...

---

Minor: TEST_MATCH is unquoted in the match flag

--test ':main $(if $(TEST_MATCH),--match=$(TEST_MATCH))'

If TEST_MATCH contains spaces or glob characters the shell will mis-parse it. --match="$(TEST_MATCH)" (inner double-quotes) would be safer, since the whole string is already inside single quotes for ghcid:

--test ':main $(if $(TEST_MATCH),--match="$(TEST_MATCH)")'

---

Observation: test-only deps in the library

The comment in lib-deps.yaml is clear and the -Wno-error=unused-packages suppression is already repo-wide, so this is understood to be a hpack 0.38 workaround. One alternative for the record: ki is already in the library as ki-effectful (which re-exports the structured concurrency primitives), so the bare ki entry might be droppable from the extras list depending on what test/integration/ actually imports.

[claude]

Review

Overall this is a solid DX/infra PR. A few concrete things worth addressing:

---

BackgroundJobs.hs — Int → Int64 (bug fix ✓)

Correct fix. hasql's decoder for bigint (int8) requires Int64, not Int; the previous code would fail at decode time when the result exceeded maxBound :: Int32 on 64-bit GHC or silently truncate values.

---

BackgroundJobs.hs — minor succinctness nit

-- current
stuck = fromMaybe 0 $ listToMaybe stuckJobs

-- with `extra` (already in deps)
stuck = headDef 0 stuckJobs

headDef from Data.List.Extra removes the listToMaybe/fromMaybe dance.

---

package.yaml / generated .cabal — duplicate -threaded

test-dev's ghc-options in package.yaml includes -threaded, but it's already present in the global ghc-options. hpack concatenates them, so the generated cabal ends up with -threaded twice. Harmless, but removing -threaded from the test-dev-specific options would clean it up.

---

lib-deps.yaml — test-only deps leaking into library

The comment explains the hpack 0.38 limitation, and -Wno-error=unused-packages silences it. That said, it's worth noting these six packages (hspec, uuid-quasi, ki, pg-transact, aeson-pretty, process) now appear in the library's build graph and affect cabal outdated, cabal check, and lock-file diffs. If hpack ever gains proper list-position !include support, extracting them to a separate test-only-deps.yaml would be the right clean-up.

---

Makefile — tmux set-option -g @build-pane scope

Using -g (global) means the pinned pane ID is shared across all windows and sessions in tmux. If someone runs make tmux-pin-here from two project windows concurrently, the second pin silently clobbers the first. -w (window-scoped) would be more conservative here.

---

DeriveUtils.hs — doc comments ✓

The new Haddock lines for WrappedEnum and WrappedEnumShow clearly distinguish the TEXT-column decoder path from the PG-ENUM path. Good addition.

---

Telemetry.hs doctest ✓

length otelColumns pin updated from 89 → 88. Correct maintenance after the column removal in #409/#410.

[claude]

See review below

[claude]

Code Review - PR 411

Overview: Adds a test-dev GHCi target (src/ + test/integration/ as one unit), consolidates lib deps into hpack-includes/lib-deps.yaml, fixes Int->Int64 for COUNT()::bigint, updates doctest pin, adds haddock to DeriveUtils.

Bug - proto-lens-protoc missing from package.yaml test-dev build-tools: package.yaml lists only hspec-discover:hspec-discover, but the generated cabal has both hspec-discover and proto-lens-protoc. Since test-dev compiles all Proto modules via generated-other-modules, the next hpack run would drop proto-lens-protoc and break the build. Add it explicitly or confirm it comes from a global build-tools stanza.

Generated Cabal: Duplicate Flags: test-dev ghc-options contain both -fwrite-ide-info and -fno-write-ide-info, and -threaded twice. Global options prepend before stanza overrides. Last occurrence wins so it is functional, but noisy.

Test-Only Deps Leaking into Library: lib-deps.yaml adds hspec/uuid-quasi/ki/pg-transact/aeson-pretty/process to the shared file, silenced by -Wno-error=unused-packages. Accepted given the hpack 0.38 limitation, but it permanently degrades -Wunused-packages signal on the library. Consider a separate hpack-includes/test-dev-extras.yaml with two !include lines in the tests: test-dev: dependencies block — the splice limitation may only affect the library: stanza.

Makefile - tmux Option Namespace: tmux set-option -g @build-pane is machine-global across all sessions. Namespacing as @monoscope-build-pane avoids clobbering other projects.

Positive: lib-deps.yaml dep consolidation is clean and fixes the vector-split drift; tmux_run pane-id tracking is more robust than grepping command names; Int->Int64 for COUNT bigint is correct; doctest pin and haddock additions are good.

[claude]

Code Review

Overview

Clean dev-loop improvement: the combined lib+test GHCi target (test-dev) and !include-based dep dedup solve real friction points. The Int → Int64 fix for COUNT(*)::bigint is a genuine correctness fix that was silently wrong.

---

Correctness / Bugs

Int → Int64 (BackgroundJobs.hs) Correct fix. PostgreSQL COUNT(*)::bigint is always 64-bit; Int happens to be 64-bit on most 64-bit hosts but is not guaranteed by the Haskell standard. The explicit Int64 is right.

headDef 0 stuckJobs (line 703) Good use of Data.List.Extra.headDef over fromMaybe 0 $ listToMaybe — already in-scope via the existing extra import.

---

Biggest Concern: Test-only Deps Leaked into Library

hpack-includes/lib-deps.yaml includes at the bottom:

# === extras shared with test-dev ===
- hspec
- uuid-quasi
- ki
- pg-transact
- aeson-pretty
- process

These are now declared as explicit library build-depends in the generated monoscope.cabal. The -Wno-error=unused-packages silences GHC, but:

• Accidental import risk: process, pg-transact, hspec etc. are now importable in production modules with no warning. One stray import would compile silently.
• ki is redundant: ki-effectful already transitively brings in ki. Adding it explicitly buys nothing and makes it importable directly in prod code.
• process is a non-trivial dep — the library almost certainly does not need shell process spawning in prod paths.

Suggestion: Keep the shared file as just the pure lib deps and list the 6 test-only extras inline in the test-dev stanza of package.yaml. Still far less duplication than before, and the test-only extras stay test-only:

# package.yaml, test-dev.dependencies:
dependencies: !include hpack-includes/lib-deps.yaml
  # test-only extras:
  - hspec
  - uuid-quasi
  - ki
  - pg-transact
  - aeson-pretty
  - process

This is exactly what the existing integration-tests section already does (lib-include + inline extras) — just move those 6 lines out of lib-deps.yaml back into the test-dev stanza.

---

Makefile

Tmux pane pinning — Using tmux set-option -wv @build-pane is more robust than grepping for command names. tmux-pin-here / tmux-unpin are a nice ergonomic addition.

TEST_MATCH quoting: The variable is embedded as --match="$(TEST_MATCH)" inside a single-quoted --test string. Shell metacharacters are inert here, but a value containing a literal " would break the inner quoting. Low severity (dev-only), but single-quoting the inner match avoids the edge case.

-osuf dyn_o -hisuf dyn_hi note: These options let ghcid reuse .dyn_o artifacts cabal already wrote. Worth noting in the comment that the first run on a clean checkout still does a full build.

---

DeriveUtils.hs

The two doc comments on WrappedEnum / WrappedEnumShow are a useful navigability improvement. Minor: 'Just "schema.type"' in the haddock will be interpreted by Haddock as a type-reference link attempt rather than a code span — @Just "schema.type"@ is more consistent.

---

Summary

Area	Assessment
Int64 correctness fix	Ship it
headDef cleanup	Cleaner
!include dep dedup approach	Good idea, implementation has the leak issue
Tmux improvements	Better
Test-only deps in library	Address before merge
live-test-dev target	Useful

The core idea is solid. The main thing worth addressing before merge is keeping the 6 test-only extras out of the library's dep graph — they are importable in prod code with no warning once they are in build-depends.

[claude]

Code Review

Overall this is a clean, well-motivated PR. The lib-deps.yaml single-source-of-truth approach, the test-dev unified GHCi target, and the tmux pane-pinning rework are all good ideas.

---

Factual correction to the existing review comment

The previous review flags four issues — all four are already addressed in this diff:

• issueCounts/alertCounts/rows Int→Int64 "not fixed" — all are fixed (BackgroundJobs.hs hunks at lines 653, 674, and 754)
• live-test-dev "missing from .PHONY" — it is present in the updated .PHONY line
• TEST_MATCH "unquoted" — the diff already uses --match="$(TEST_MATCH)" with inner double-quotes
• base16-bytestring "duplicated in lib-deps.yaml" — appears exactly once; the duplicate was in the old package.yaml and was cleaned up

---

Genuine observations

BackgroundJobs.hs — headDef

headDef 0 stuckJobs is a nice simplification over fromMaybe 0 $ listToMaybe. Import from Data.List.Extra alongside the existing chunksOf/groupBy is correct.

lib-deps.yaml — test-only deps leaking into library stanza

The workaround (hspec, uuid-quasi, ki, pg-transact, aeson-pretty, process appearing in the library's declared deps) is documented and silenced by -Wno-error=unused-packages. Fine for an application. One small suggestion: extend the comment to say where a new test-only dep should go, so the next contributor doesn't add it in the wrong place and reintroduce drift.

package.yaml test-dev — proto-lens-protoc absent from build-tools

The generated monoscope.cabal has proto-lens-protoc:proto-lens-protoc in test-dev's build-tool-depends, but it is not listed in the test-dev stanza of package.yaml. If hpack infers it from generated-other-modules that is fine, but worth confirming — if that inference ever stops, the cabal will silently drop the tool dep and proto codegen breaks.

Makefile — log file gitignored?

tee build-test-dev.log creates a new log file. Make sure build-test-dev.log is in .gitignore alongside build.log.

---

Summary

The Int→Int64 fixes are correct and complete across all affected queries, dep deduplication via !include is the right call, and the Makefile improvements are solid. Main things to confirm before merge: hpack's proto-lens-protoc inference for test-dev, and the .gitignore entry for the new log file.