oficina / sheets Public

build: add a stamped release target, and refuse dirty trees

The root-page release shipped with a plain `go build`, so rastrillo's
BuildVersion kept its "dev" default and the live instance answered
GET /api/version with "dev". That is the field used to tell what is
actually deployed, so an unstamped binary makes the fleet unreadable
while still reporting a successful deploy — and `carlos deploy` printed
"live" against it.

The stamp was never encoded anywhere: earlier releases carried a sha
because the ldflags were typed by hand, which is why one build without
them was enough to lose it. `make release` is now the only supported way
to build a shipping binary.

It also refuses a dirty tree, because a sha that names a commit other
than what is in the binary is worse than no stamp at all. Both behaviours
were watched: the refusal fires on a dirty tree and writes no output
file.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01APtNStXDZ5H2fAc1ZuGqtS

Paul Campbell pushed by paul@keymail.dev 1 month ago f013d2ef0cce5f0e32a2bd928f9f8def957129ee parent aa3d7f2

2 files changed, +20 −1

.gitignore +1 −0
Makefile +19 −1

Intent

full ledger: core

diff --git a/.gitignore b/.gitignore

index 62171f9..35c6f85 100644

--- a/.gitignore

+++ b/.gitignore

@@ -16,3 +16,4 @@

.claude/workflows

.claude/launch.json

.claude/settings.local.json

+/sheets-arm64

diff --git a/Makefile b/Makefile

index b695aad..0e99e22 100644

--- a/Makefile

+++ b/Makefile

@@ -1,4 +1,4 @@

-.PHONY: build test test-vet test-fmt test-build test-migrations test-go test-race

+.PHONY: build release test test-vet test-fmt test-build test-migrations test-go test-race

# One binary, CGO_ENABLED=0, cgo-free SQLite (AGENTS.md / Deploys). Set

# per-target below rather than exported globally, because test-race is

@@ -6,6 +6,24 @@

build:

CGO_ENABLED=0 go build ./...

+# The ONLY way a binary should be built for shipping. rastrillo's

+# BuildVersion defaults to "dev", and an unstamped binary answers

+# GET /api/version with "dev" — which is the field used to tell what is

+# actually deployed, so an unstamped release makes the fleet unreadable

+# while still looking like a successful deploy. Shipping one by hand

+# with a plain `go build` is exactly the mistake this target removes.

+# A dirty tree is refused: the sha would name a commit that is not what

+# is in the binary, which is worse than no stamp at all.

+RELEASE_OUT ?= ./sheets-arm64

+release:

+ @git diff --quiet && git diff --cached --quiet || \

+ { echo "refusing to build a release from a dirty tree"; exit 1; }

+ CGO_ENABLED=0 GOOS=linux GOARCH=arm64 go build -trimpath \

+ -ldflags "-s -w -X github.com/carlosframework/rastrillo.BuildVersion=$$(git rev-parse --short HEAD)" \

+ -o $(RELEASE_OUT) ./cmd/sheets

+ @echo "built $(RELEASE_OUT) stamped $$(git rev-parse --short HEAD)"

# The gate, split into named targets so CI can report them individually

# (.amadan/ci.d calls these one at a time, each its own step). `make

# test` runs exactly the same six things in the same order, so the