Replace Makefile with Taskfile.yml and move all generation-related logic there by denik · Pull Request #5050 · databricks/cli

denik · 2026-04-21T12:09:50Z

Each task describes inputs (sources) and outputs as precise as possible.

The tasks mostly match what we had in Makefile but slightly more consistent approach:

All go related tasks (tidy, lint, test) now have 3 subtasks for each submodule (root, tools, codegen) which are aggregated into main one. Previously we did not run tests & linters for tools and codegen modules.
The targets that use lintdiff.py to do incremental run have -q suffix (old fmtfull -> new fmt, old fmt -> new fmt-q).
generate:genkit is task that runs genkit + follow ups. “generate” is aggregate over all generation work.

This also consolidates all knowledge about generation in one place:

tools/post-process.sh is removed, the followup commands are now part of generate:genkit task
testmask no longer contains of dependencies of CI targets, it reads those from Taskfile.yml

The runner (go-task) is packaged as a go tool, no installation need. Shortcut ./task is available to run it. Makefile is temporarily a wrapper that calls ./task.

This enables caching - go-task will only re-run if sources changed. This helps when checking after agent work - if they did run linters / tests already, then running ./task is very quick.

~/work/cli-trees/task-file % time ./task
…
./task 612.63s user 310.22s system 726% cpu 2:06.95 total

~/work/cli-trees/task-file % time ./task
…
./task 2.43s user 10.58s system 133% cpu 9.727 total

Additionally, all golangci-lint tasks are fixed to use per-worktree & per-submodule tmp directory, which enables parallel runs within worktree and across worktrees.

Replace the hardcoded prefix table with parsing of Taskfile.yml so the list of paths that triggers each CI test target is derived from the test:exp-aitools, test:exp-ssh, and test:pipelines tasks' `sources:`. Also makes the tool CWD-independent via `git rev-parse --show-toplevel`. Co-authored-by: Isaac

Introduce Taskfile.yml as the single source of truth for build, test, lint, format, and codegen commands. Provide a thin Makefile wrapper that forwards all targets to task so existing `make <target>` invocations keep working for single-word targets. Move python/Makefile into the same Taskfile under python:* tasks. Task is vendored via a separate tools/task module to avoid its charmbracelet dependencies conflicting with tools/go.mod (golangci-lint pins older charmbracelet versions). Also update acceptance test helper to build yamlfmt directly — the Makefile wrapper can no longer forward the old `tools/yamlfmt` target since it's now a Task-only dependency. Co-authored-by: Isaac

Replace `make <target>` with `go tool -modfile=tools/task/go.mod task <target>` in CI workflows and the setup-build-environment action. This skips the Makefile wrapper and surfaces task's real target names (e.g. task:exp-aitools, fmt:full) in CI logs. release-build.yml inlines the python wheel build steps (which used to live in python/Makefile) so the release job does not depend on task. Co-authored-by: Isaac

Replace make commands with their task equivalents in the build/test and code-quality sections so AI agents and human readers see the current invocation style. Co-authored-by: Isaac

Each submodule gets its own minimal .golangci.yaml so lint:tools and lint:codegen can run from those directories without inheriting the root config. The root config enables the ruleguard gocritic check against `libs/gorules/rule_*.go`, which golangci-lint resolves relative to the working directory — from inside tools/ or bundle/internal/tf/codegen/ that path does not exist and lint aborts before running. Co-authored-by: Isaac

Shortens `go tool -modfile=tools/task/go.mod task <target>` to `./task <target>` and resolves the modfile via `git rev-parse --show-toplevel` so the script works from any subdirectory. Co-authored-by: Isaac

Replace every `go tool -modfile=tools/task/go.mod task` invocation in CI workflows, the Makefile wrapper, and the dbr:test deco command with the shorter `./task`. The dbr:test command previously pointed at tools/go.mod, which does not include task as a tool — the wrapper fixes that too. Co-authored-by: Isaac

- build: correct generates from `databricks` to `cli` (Go module name) - drop build:vm (unused) - snapshot, snapshot:release, schema:for-docs, docs: add sources/generates - lint:tools, lint:codegen: include .golangci.yaml, go.mod, go.sum in sources - python:docs, python:codegen: include docs and databricks inputs - split fmt into fmt:python, fmt:go, fmt:yaml; fmt and fmt:full dispatch the three in parallel via deps Co-authored-by: Isaac

Co-authored-by: Isaac

Renames lint:check/lint:tools/lint:codegen to lint:go:root/lint:go:tools/ lint:go:codegen to make it explicit they're Go-specific. Adds lint:go that runs all three in parallel so a single invocation covers the whole repo. Co-authored-by: Isaac

The `fmt` aggregate uses the incremental `fmt:go` (via lintdiff.py) for a fast interactive loop, while `fmt:full` uses `fmt:go:full` to sweep the whole tree. Python and YAML tasks are shared between the two aggregates. Co-authored-by: Isaac

The root CLI binary only imports from bundle/, cmd/, experimental/, internal/, libs/, so test-only trees (acceptance/, integration/), the tools/ and bundle/internal/tf/codegen/ submodules, and _test.go files are excluded from build/snapshot/snapshot:release source globs. This lets Task's checksum cache skip rebuilds when only tests change. Task v3 uses `- exclude: <pattern>` for glob exclusion (the `!` prefix from doublestar isn't honored by Task's source matching). Also remove the _build-yamlfmt task — nothing referenced it, and acceptance tests build yamlfmt directly via BuildYamlfmt with exeSuffix for Windows. Co-authored-by: Isaac

Introduce `generate` as an aggregator that runs each generator in sequence: - `generate:commands` (universe + genkit; was top-level `generate`) - `generate:schema` (was `schema`; now has tight sources, no longer gated by `git diff` against merge-base) - `generate:schema-docs` (was `schema:for-docs`) - `generate:docs` (was `docs`) - `generate:validation` (now has tight sources) - `generate:direct` (unchanged) - `generate:openapi-json` (was `codegen:openapi-json`) Each reflection-based generator (schema, schema-docs, validation, docs) lists its Go sources explicitly and includes `go.mod`/`go.sum`, so the aggregator picks up SDK version bumps from `generate:commands` automatically. Test files are excluded from the source globs. Trim `tools/post-generate.sh` to only post-process genkit output (consistency check, tagging.py relocation, whitespace fix). The former `make schema`/`make schema-for-docs`/`make generate-validation`/ `make -C python codegen` calls are gone — those run as standalone tasks in the aggregator now. Update `.github/workflows/push.yml`, `AGENTS.md`, `.agent/rules/auto-generated-files.md`, `.agent/skills/pr-checklist/SKILL.md`, `bundle/docsgen/README.md`, and schema test comments to reference the new task names. Also add `.databricks/` (snapshot binary output) to `.gitignore`. Co-authored-by: Isaac

…efault - Move refschema regeneration into a standalone generate:refschema task with a tight sources list so the cache tracks what actually affects `bundle debug refschema` (cmd, dresources adapters, libs/structs, go.mod). - generate:commands now only runs genkit + post-process. - Aggregator runs generate:refschema between commands and the other generators so fresh out.fields.txt feeds generate:direct. - default now invokes the full generate aggregator instead of just generate:schema + generate:validation. - Fix acceptance/install_terraform.py to skip doc-comment lines when reading ProviderVersion (regressed by the recent doc comment added to bundle/internal/tf/codegen/schema/version.go). Co-authored-by: Isaac

Convention change: the plain task name is the full variant. Quick / incremental variants carry a -q suffix on the top-level namespace. - fmt:full -> fmt, fmt:go:full -> fmt:go (full is the default) - fmt (old incremental) -> fmt-q; fmt:go (old incremental) -> fmt-q:go - lint:full removed; lint is now the aggregator across all Go modules - lint (old incremental root+fix) -> lint-q:go:root; lint-q alias added - lint:go aggregator runs modules sequentially (concurrent golangci-lint invocations have shown unreliable behavior) - default task now runs the -q variants for a quick dev loop - new `all` task runs the full suite (checks, fmt, lint, test, generate, python:lint, python:test) — replaces old default semantics Close the gap where CI only linted the root module: CI now runs `./task lint` which covers root + tools + bundle/internal/tf/codegen. lintdiff.py gains --root-module to skip paths under nested go.mod files; needed for `golangci-lint run` (typechecks) but not for `fmt` (filesystem walk), so only lint-q:go:root opts in. Co-authored-by: Isaac

Drop the --root-module flag; infer from the wrapped subcommand instead. `golangci-lint run` typechecks and needs the filter; `golangci-lint fmt` walks the filesystem and wants to see changed files across modules. Callers no longer need to know which mode applies. Co-authored-by: Isaac

generate:commands is expensive (bazel/genkit build) and requires a universe checkout. Its outputs are committed, so a fresh worktree at any commit already has them. Run `./task generate` explicitly when codegen inputs change. Co-authored-by: Isaac

- Use {{.GO_TOOL}} everywhere (redefine with {{.ROOT_DIR}} so tasks with `dir:` can reference it). - Drop dead `{{.X | default ...}}` fallbacks in task-local vars; the top-level vars are always set and pass through via `deps:` + `vars:`. - Serialize `checks` (tidy/ws/links) and `test:update-all` — they were racing on the same files via parallel `deps:`. - Add `deps: [tidy]` to `snapshot` and `snapshot:release` (matches `build`). - Add `desc:` to `generate:direct-apitypes` and `generate:direct-resources`. - Expand `lint-q:go:root` desc to note it skips tools/ and codegen/. - Drop `sources:` from `ws` (script is fast; no point caching). - Fix code-generation section comment to list all reflection-based generators (refschema, schema-docs missing). Co-authored-by: Isaac

- `lint:go:root` and `lint-q:go:root` now exclude tools/** and bundle/internal/tf/codegen/** from sources. golangci-lint `./...` stops at module boundaries, so changes there never affect root-module lint output — no reason to invalidate the cache. - Split `tidy` by module, matching the `lint:go` layout: - `tidy` — aggregator across all three modules - `tidy:root` — root (previously `tidy`) - `tidy:tools` — tools/ - `tidy:codegen` — bundle/internal/tf/codegen `build`, `snapshot`, `snapshot:release` switch to `deps: [tidy:root]` (they only build the root module); `checks` still calls the aggregator. Co-authored-by: Isaac

Mirrors main's e24ae4e (Remove Slow tests #5032): drop test:slow, test:slow-unit, test:slow-acc, and the `-short` flag from test:unit and test:acc. Slow tests support in the acceptance runner was removed because the split between short and fast tests adds complexity and makes it possible to miss failures. Co-authored-by: Isaac

Switch lint:go and tidy aggregators from sequential cmds: to parallel deps:. Each lint:go module sets its own TMPDIR under .tmp/ so concurrent golangci-lint invocations — both siblings here and sibling worktrees running lint at the same time — don't serialize on the shared /tmp lock. The TMPDIR path lives under the repo (but not at ROOT_DIR itself, which would trip Go's "go.mod in os.TempDir" check). Add /.tmp/ to .gitignore. Co-authored-by: Isaac

Rename generate:direct-{apitypes,resources,clean} to generate:direct:{apitypes,resources,clean} so the direct engine subtasks follow the same colon-namespaced convention as lint:go:root, tidy:root, etc. Also broaden generate:refschema sources to all of bundle/ (excluding _test.go) plus go.mod / go.sum — the command reflects over types reachable from bundle/, so narrowing to a handful of packages risks missing changes that affect the schema. Co-authored-by: Isaac

…nerate.sh Renamed because the task does more than generate CLI commands: it also emits the tagging workflow + tagging.py, bumps the SDK version, and post-processes the output. Moved the steps from tools/post-generate.sh (previously invoked via .codegen.json post_generate hook) inline into the task so ./task generate is self-contained and the tree is left clean afterwards. Also dropped generate:direct:clean which was out of place. Co-authored-by: Isaac

The tasks target the python/databricks-bundles package (pydabs), not Python tooling in general (fmt:python stays, since that one does apply repo-wide). Renaming avoids confusion and matches how the project refers to the package elsewhere. Co-authored-by: Isaac

The old `test:unit` only ran `go test ./...` in the root module; tests in tools/ and bundle/internal/tf/codegen weren't run. Split into per-module tasks plus an aggregator that concatenates their JSON outputs. Also makes tools/testmask tolerate non-string `sources:` entries (e.g. `- exclude: tools/**`), which previously made yaml.Unmarshal fail once those excludes were introduced. Co-authored-by: Isaac

Previously `cat … > test-output-unit.json` (and the top-level `test` merge) ran every invocation even when all sub-tasks were cache-hit. Declaring the inputs as `sources:` and the merged file as `generates:` lets Task skip the cat when nothing upstream changed. Co-authored-by: Isaac

Consistent with other colon-namespaced tasks (test:unit:*, test:acc). testmask now derives the Taskfile task name from the CI output name by replacing "-" with ":", so the list of targets no longer needs both forms — a single slice suffices. CI target identifiers (job names, cache keys) remain dash-separated. Co-authored-by: Isaac

Rename to integration-short, dbr-integration, dbr-test. These are single-task variants without a parent aggregator, so dash suffixes fit better than colon namespacing. Co-authored-by: Isaac

Redundant — the `deps:` structure already makes parallel execution obvious to anyone reading the task definition, and users running `./task --list` don't need that implementation detail in the summary. Co-authored-by: Isaac

Previous comments claimed generate:genkit bumps the SDK version in go.mod/go.sum; it doesn't. Genkit regenerates CLI command stubs from .codegen/_openapi_sha and refreshes .gitattributes, but SDK bumps are a manual `go get` step. TestConsistentDatabricksSdkVersion is what keeps the two in sync. Co-authored-by: Isaac

Replace every `:` with `-` in task names, so every task is a valid target name for tools that don't allow `:` — primarily `make`, where `make test:unit` parses as target `test` with prerequisite `unit` rather than as a single target `test:unit`. With dashes everywhere: * `make X` and `./task X` invoke the same task for every X, making the Makefile wrapper a true drop-in for former `make` users. * testmask no longer needs the CI-name ↔ Taskfile-name translation: the CI job output names are the task names. CI job identifiers (job names, cache-key strings) are unchanged; they were already dash-separated. Co-authored-by: Isaac

The previous `%: @./task "$@"` pattern rule didn't trigger for phony targets in GNU Make 3.81 (Apple's default); `make test-unit` printed "Nothing to be done". `.DEFAULT` is the built-in hook for unmatched targets and fires reliably. Co-authored-by: Isaac

Co-authored-by: Isaac

Mirrors `default` but swaps in the non-incremental fmt/lint variants. Also tightens the `all:` description to mention regeneration scope (skips genkit) and test updates. Co-authored-by: Isaac

Drop the leading \`-\` on test-update and test-update-templates cmds. The dash tells Taskfile to ignore a non-zero exit, which masked broken acceptance tests — we want these to fail the dev loop. Co-authored-by: Isaac

Removing as unused. out.test.toml is regenerated by the normal test-update flow. Co-authored-by: Isaac

Replace the broad \`**/*.go + acceptance/**\` source globs with the same curated list the \`build\` task uses (acceptance_test.go builds the CLI in-process via BuildCLI), plus acceptance/** for the test harness and fixtures, plus go.mod/go.sum. Two separate source lists because out.* files play different roles: - test-acc: out.* are golden inputs. A change to a committed out.* must re-run the test to confirm it still matches. - test-update*: out.* are outputs. The task rewrites them, so keeping them in sources would change the checksum each run and force re-execution on every invocation. The latter uses a shared &ACC_SOURCES_UPDATE YAML anchor across test-update, test-update-templates, and test-update-aws. Co-authored-by: Isaac

janniklasrose · 2026-04-21T12:30:26Z

@@ -1,3 +1,3 @@
 ---
 description: Rules for how to deal with auto-generated files
 globs:


Let's add a task that parses Taskfile.yml to auto-gen this?

this could be a follow up

janniklasrose · 2026-04-21T12:39:30Z

+    generates:
+      - test-output.json
+    cmds:
+      - cat test-output-unit.json test-output-acc.json > test-output.json


cating two JSON doesn't create a JSON. Better to use jq here (ditto for others)

these are json-lines files

should we use .jsonl then?

we could rename, but out of scope there (and might need updating other places like deco repo)

janniklasrose · 2026-04-21T12:43:29Z

+          sed -i 's|tagging.py|internal/genkit/tagging.py|g' .github/workflows/tagging.yml
+        fi
+      - "{{.GO_TOOL}} yamlfmt .github/workflows/tagging.yml"
+      - ./tools/validate_whitespace.py --fix


can we reference ws-fix task?

yes, 4d19c6b

The test failed on shallow CI clones (default fetch-depth: 1) because HEAD~2 doesn't resolve. The well-known empty tree SHA (4b825dc) is always available to git regardless of clone depth, and diffing HEAD against it produces the full file set — still exercising the non-empty result path. Co-authored-by: Isaac

@janniklasrose

Addresses review feedback from @janniklasrose — dedupe the direct validate_whitespace.py --fix call. Co-authored-by: Isaac

janniklasrose · 2026-04-21T14:23:23Z

+    generates:
+      - test-output.json
+    cmds:
+      - cat test-output-unit.json test-output-acc.json > test-output.json


should we use .jsonl then?

Drops the \`git rev-parse --show-toplevel\` subshell in favor of \`dirname "$0"\`, which is simpler and works even outside a git repo (e.g. in archive extractions or release artifacts). Co-authored-by: Isaac

The system \`task\` binary in some PATHs is broken / missing; the repo's wrapper at ./task always works. Update AGENTS.md / CLAUDE.md, .agent/rules/auto-generated-files.md, bundle/docsgen/README.md, and the comment block in bundle/internal/schema/main_test.go. Co-authored-by: Isaac

Co-authored-by: Isaac

Why: the minimal configs in tools/ and bundle/internal/tf/codegen/ existed only to opt nested modules out of root's ruleguard check (whose rules path is cwd-relative and unresolvable from those dirs). Disabling gocritic via --disable is equivalent and removes the duplicated enable lists. Co-authored-by: Isaac

denik added 30 commits April 21, 2026 14:08

docs: update AGENTS.md command references for task

5dc976e

Replace make commands with their task equivalents in the build/test and code-quality sections so AI agents and human readers see the current invocation style. Co-authored-by: Isaac

Add ./task wrapper script

9f2873a

Shortens `go tool -modfile=tools/task/go.mod task <target>` to `./task <target>` and resolves the modfile via `git rev-parse --show-toplevel` so the script works from any subdirectory. Co-authored-by: Isaac

Document fmt:go module-spanning behavior

a5fb2b2

Co-authored-by: Isaac

Taskfile: flatten integration:short, dbr:integration, dbr:test

a9d88d2

Rename to integration-short, dbr-integration, dbr-test. These are single-task variants without a parent aggregator, so dash suffixes fit better than colon namespacing. Co-authored-by: Isaac

Taskfile: drop "(in parallel)" from test task descriptions

2642025

Redundant — the `deps:` structure already makes parallel execution obvious to anyone reading the task definition, and users running `./task --list` don't need that implementation detail in the summary. Co-authored-by: Isaac

denik added 7 commits April 21, 2026 14:08

tools: tidy duplicate require in go.mod

6f920d8

Co-authored-by: Isaac

Taskfile: add full task for more complete dev loop

d155086

Mirrors `default` but swaps in the non-incremental fmt/lint variants. Also tightens the `all:` description to mention regeneration scope (skips genkit) and test updates. Co-authored-by: Isaac

Taskfile: fail loudly on test-update errors

bb5fae3

Drop the leading \`-\` on test-update and test-update-templates cmds. The dash tells Taskfile to ignore a non-zero exit, which masked broken acceptance tests — we want these to fail the dev loop. Co-authored-by: Isaac

Taskfile: drop generate-out-test-toml

039a4b8

Removing as unused. out.test.toml is regenerated by the normal test-update flow. Co-authored-by: Isaac

janniklasrose reviewed Apr 21, 2026

View reviewed changes

denik added 2 commits April 21, 2026 15:39

Taskfile: reference ws-fix from generate-genkit

4d19c6b

Addresses review feedback from @janniklasrose — dedupe the direct validate_whitespace.py --fix call. Co-authored-by: Isaac

janniklasrose approved these changes Apr 21, 2026

View reviewed changes

task wrapper: resolve modfile via script dirname instead of git

c7cc94c

Drops the \`git rev-parse --show-toplevel\` subshell in favor of \`dirname "$0"\`, which is simpler and works even outside a git repo (e.g. in archive extractions or release artifacts). Co-authored-by: Isaac

denik requested review from pietern and simonfaltum April 21, 2026 14:44

denik added 3 commits April 21, 2026 16:46

lintdiff: hard-code nested module list

ed7da79

Co-authored-by: Isaac

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Replace Makefile with Taskfile.yml and move all generation-related logic there#5050

Replace Makefile with Taskfile.yml and move all generation-related logic there#5050
denik wants to merge 43 commits intomainfrom
denik/task-file

denik commented Apr 21, 2026

Uh oh!

janniklasrose Apr 21, 2026

Uh oh!

denik Apr 21, 2026

Uh oh!

janniklasrose Apr 21, 2026

Uh oh!

denik Apr 21, 2026

Uh oh!

janniklasrose Apr 21, 2026

Uh oh!

denik Apr 21, 2026

Uh oh!

janniklasrose Apr 21, 2026

Uh oh!

denik Apr 21, 2026

Uh oh!

janniklasrose Apr 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

denik commented Apr 21, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants