Files
settled-reach/tooling/main.py
T
jpmschweitzerandClaude Opus 5 c924b0934e feat(config): T-1277 — detach, and a failed job that looked busy
core/process.py spawns a child that outlives its parent: its own session, so a
signal to the parent's group or a timeout kill does not take the work with it;
re-execing reach by BARE NAME, because an absolute path would freeze the child
to whichever checkout was current at spawn time and silently run the wrong
source after a repoint; and streams kept separate exactly as in the foreground,
events to <id>.jsonl and real output to <id>.out.

Testing a case the ticket did not name found a real hole. Recording completion
inside @command looked right and was wrong: a child that fails BEFORE any
command runs — bad arguments, an unknown verb, an import error — never reaches
that decorator. `reach --detach check bogus` left its metadata reading
"running" forever with the process long gone. That is the exit-0 trap wearing a
new disguise and worse than the original, because a failed job that looks busy
sits somewhere nobody is watching, and a caller polling for completion would
wait indefinitely on something that failed in milliseconds.

So completion is recorded at the PROCESS's exit instead. main.py gains main(),
wrapping cli() in a single try/finally, and the entry point moves to
main:main. Every exit path now passes through one place. Removed from @command
rather than left in both — two writers of one field is how they drift.

Verified on three paths: success records done/0, a real drift failure records
failed/1, and the parse failure that exposed the hole now records failed/2.

One narrow conformance exemption, with its reason inline so it does not read as
an oversight: the no-domain-imports-core.jobs invariant fired on main.py,
correctly by its letter and wrongly by its purpose. main.py is not a command;
it is the entry point, and it already owns --detach.

Still open, and carried to T-1278: a child killed outright cannot record
anything, so jobs list must reconcile against process liveness rather than
trusting the file.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 16:46:19 +02:00

203 lines
8.4 KiB
Python

"""`reach` — one command for every repo tool (D-263).
**This file is a router and nothing else.** No logic, no I/O, no pydantic, no
domain imports at module level. It is the file most likely to accumulate "just
one small thing", and the only defence is that it stays short enough that an
addition is obvious in review.
The root is a `typer.Typer`, and the reason is worth recording because the
original plan was different. T-1259 specified a `click.Group` root, on the
theory that it would keep typer off the `reach --help` path. **That is no longer
possible: typer vendors click as of 0.26.0** — there is no top-level `click`
package to import, and the docs are explicit that "extracting the internal Click
app" is unsupported. Mixing a real `click.Group` root with typer sub-apps would
mean two different Click implementations in one process.
So the customisation surface is `typer.Typer(cls=...)` with a `TyperGroup`
subclass, which is the supported path and is what `LazyDomainGroup` below uses
to register domains lazily.
`rich_markup_mode=None` is not a style preference — it is worth 94 ms of the
168 ms an empty `--help` otherwise costs, and it keeps `rich` and `pygments`
off the import path entirely (verified absent from `sys.modules`). It also
stops typer drawing box-art help, which it does **even when stdout is a pipe**,
so hook logs and agent output stay readable. One line to revert if the boxes
are ever wanted more than the milliseconds.
The callback below is not decoration. A `typer.Typer` with **no commands and no
callback** raises `RuntimeError: Could not get a command for this Typer
instance` at build time; with a callback it builds fine. Since every domain is
registered lazily and none is eager, the callback is what makes the root legal.
"""
from __future__ import annotations
import importlib
import typer
from tooling.core.cli import ReachGroup
# The domain registry: name -> (import target, one-line help).
#
# This table is the ONLY thing `reach --help` reads. The short help lives here
# as a literal string rather than being pulled off the loaded command, because
# reading it off the command is precisely what would import the world — see
# LazyDomainGroup.format_commands.
#
# Adding a domain is adding a line here plus a `router.py` that exposes `app`.
DOMAINS: dict[str, tuple[str, str]] = {
"check": (
"tooling.domains.check.router:app",
"Consistency gates — the checks the push hook runs",
),
}
class LazyDomainGroup(ReachGroup):
"""Lists domains without importing them; imports exactly the one invoked.
Extends `ReachGroup`, so bare `reach` prints the domain list and exits 0
like every domain group does. Only the laziness and the domain-specific
wording live here.
Three overrides, and the third is the one that matters. `TyperGroup`'s own
`format_commands` loops over `list_commands` calling `get_command` on each,
just to read a short help string off the loaded command — which, with lazy
loading underneath, imports every domain in the registry to render `--help`.
That would defeat the whole mechanism silently, while looking correct.
So `format_commands` is overridden to read help from `DOMAINS` and never
touch `get_command`. The cost of `reach --help` is then flat no matter how
many domains exist, which is the property that has to hold as this grows
from one domain to a dozen.
"""
def list_commands(self, ctx: typer.Context) -> list[str]:
return sorted({*super().list_commands(ctx), *DOMAINS})
def get_command(self, ctx: typer.Context, cmd_name: str):
if cmd_name in DOMAINS:
return _load_domain(cmd_name)
return super().get_command(ctx, cmd_name)
def resolve_command(self, ctx: typer.Context, args: list[str]):
# Closed-set enumeration (D-263). Click's default is "No such command
# 'x'" — which tells you that you are wrong without telling you what
# would be right, the exact gap measured in pql. The accepted set is
# sitting in DOMAINS, already loaded for --help, so naming it is free.
if args and args[0] not in DOMAINS and super().get_command(ctx, args[0]) is None:
listed = ", ".join(sorted({*DOMAINS, *super().list_commands(ctx)}))
ctx.fail(f"unknown domain {args[0]!r}\n\nChoose one of: {listed}")
return super().resolve_command(ctx, args)
def format_commands(self, ctx: typer.Context, formatter) -> None:
# Deliberately does NOT call get_command. See the class docstring.
rows = [(name, short) for name, (_target, short) in sorted(DOMAINS.items())]
for name in sorted(super().list_commands(ctx)):
command = super().get_command(ctx, name)
if command is not None and not command.hidden:
rows.append((name, command.get_short_help_str(80)))
if rows:
with formatter.section("Domains"):
formatter.write_dl(sorted(rows))
def _load_domain(name: str):
"""Import one domain's router and convert its Typer app to a command."""
target, _short = DOMAINS[name]
module_name, _, attr = target.partition(":")
module = importlib.import_module(module_name)
return typer.main.get_command(getattr(module, attr))
cli = typer.Typer(
name="reach",
cls=LazyDomainGroup,
help="Repo tooling for The Settled Reach.\n\nRun `reach <domain> --help` to see what a domain can do.",
no_args_is_help=True,
add_completion=False,
rich_markup_mode=None,
context_settings={"help_option_names": ["-h", "--help"], "max_content_width": 100},
)
def main() -> None:
"""Console entry point — `reach`.
Exists so a detached child records its completion at the PROCESS's exit
rather than at a command's. Recording it inside `@command` looked right and
was subtly wrong: a child that fails before any command runs — bad
arguments, an unknown verb, an import error — never reaches that decorator,
so its metadata said `running` forever. A failed job that looks busy is the
exit-0 trap wearing a new disguise, and worse than the original because
nothing is watching a background job.
Here, every exit path passes through one `finally`.
"""
exit_code = 0
try:
cli()
except SystemExit as exc:
exit_code = exc.code if isinstance(exc.code, int) else 1
raise
except BaseException:
exit_code = 1
raise
finally:
# Imported lazily and only when detached, so `reach --help` never pays
# for it — the laziness T-1260 protects applies here too.
import os
from tooling.core import jobs
if os.environ.get(jobs.ENV_JOB_ID):
from tooling.core import process
process.finish_if_detached(exit_code)
@cli.callback()
def root(
verbose: bool = typer.Option(
False, "--verbose", "-v", help="Show progress events and full tracebacks."
),
no_input: bool = typer.Option(
False, "--no-input", help="Never prompt. Hooks and agents should always pass this."
),
detach: bool = typer.Option(
False, "--detach", help="Run in the background; print a job id and return at once."
),
) -> None:
"""Global options, declared once here so every domain inherits them.
A per-domain copy of these is how the two would drift apart — one router
growing a `--verbose` that sets a different level, or forgetting `--no-input`
entirely.
Note `--no-input` is about PROMPTING, not output format. Console already
chooses JSONL versus rendered text from `isatty` with an `SR_OUTPUT_FORMAT`
override; making this flag a second, conflicting way to say the same thing
would leave nobody sure which one wins.
"""
from tooling.core import console, runtime
if verbose:
console.set_level("debug")
runtime.set_no_input(no_input)
if detach:
# Handled here, before any domain loads, because detaching is a property
# of the INVOCATION rather than of the verb — every command gets it and
# no command implements it. The child re-runs this same argv with
# --detach stripped, so it does the work instead of forking again.
from tooling.core import process
job_id = process.spawn_detached(process.current_argv())
console.out(job_id)
console.verdict(
f"started job {job_id} — this exit status means STARTED, not succeeded",
fix=None,
)
raise typer.Exit(0)