[slint]

Depends on inveniosoftware/invenio-app#113

• feat: rpc-server
  • use rpc-server which runs invenio commands. this is a performance
    enhancement
• fix: warning
• README: add new features
• feat: add data path
• feat(config): configure app-rdm version
  • usefull for beta.dev versions
• feat(rpc): make the use configurable

refactor(rpc): talk to the RPC server directly over its socket

• The client used to spawn a full uv run rpc-server send subprocess per command, paying the interpreter and import cost the server exists to avoid, and that subprocess always exited 0, so failed commands were reported as successful steps.
• send_command() now returns an RPCCall that speaks the JSON-lines protocol with stdlib socket+json and maps the response onto a ProcessResponse, so exit codes and stderr propagate and only the server needs the virtualenv.
• The server is spawned lazily when the first command executes (not while the step list is being built), probed with a bounded retry loop instead of busy-spinning subprocess pings, and terminated at exit only if this process started it.
• The socket lives at <instance_path>/rpc.sock, so RPC silently defers to plain commands until the instance path is known (first install).

fix(rpc): parse use_rpc as a boolean and document the socket setup

• configparser .get() returns the raw string, so use_rpc = false in the .invenio file still enabled RPC. Read it with getboolean instead.
• Drop the unused rpc_port option (the server no longer has a port) and update the README for the socket-based setup and the use_rpc flag.

feat(rpc): let CommandStep execute RPC calls

• CommandStep now accepts an RPCCall as its cmd: it executes it over the socket, surfaces the buffered output (terminal or log file, like run_interactive does), and applies the same skippable-to-warning mapping.
• This lets any step-based command switch to send_command() without changing the step plumbing.
• The output handling is shared with the assets loop via a new echo_output() helper.

feat(rpc): route services setup through the RPC server

• All invenio commands in services setup (including the --force cleanup and demo steps) now go through send_command(), so with use_rpc enabled the whole sequence runs on one long-lived server instead of paying app startup per command.
• Without use_rpc the behavior is unchanged.
• The pybabel translation steps stay as plain subprocesses since they are not invenio CLI commands.

feat(rpc): stream output via fd passing and unify command ops

• invenio_command() (formerly send_command) now always returns a callable op: RPCOp when the RPC server is in use, LocalOp wrapping a plain subprocess otherwise. Both take the same arguments, so callers no longer branch on whether they got an argv list or a callable.
• It is meant for short-lived, non-interactive invenio commands; everything else keeps using run_command().
• RPCOp sends the client's stdout/stderr (or the log file) along with the request using SCM_RIGHTS, and the server writes command output directly to them. Output shows up live in the terminal or log file like with a normal subprocess, webpack builds included, which also fixes the --node-log-file regression.
• The captured-output mode is still available as op(capture=True) for callers that need the output programmatically, like update_instance_path.

feat(rpc): apply env when spawning the server

• RPCOp no longer drops env. It is added to the Popen environment when the call is the one that spawns the server, which is the only place it can take effect: INVENIO_* variables and FLASK_DEBUG become app config at create_app time, so the auto-spawned server picks up e.g. pnpm's INVENIO_WEBPACKEXT_NPM_PKG_CLS from the assets steps.
• Per-request env would not work, because the server caches the app and config-derived variables are only read at startup.
• A server that is already running keeps the environment it was started with, so setting WEBPACKEXT_NPM_PKG_CLS in invenio.cfg remains the safe option for pnpm.

fix(rpc): connect directly instead of pinging before each call

• The client no longer pings before sending: the single-threaded server queues the connection of a second client until the current command finishes, whereas the old 2s ping timeout made a busy server look absent and triggered a doomed second spawn.
• Only a refused or missing socket (nothing listening) now causes a spawn, probed with a cheap connect check that also recognizes a busy server.
• If the spawned server dies during startup, the client rechecks the socket before reporting failure, so losing a spawn race against another invenio-cli whose server won the socket is treated as success.

fix(services): stream docker compose output and fail on errors

• start_containers() captured the compose output through run_cmd and ensure_containers_running() discarded the response, so docker compose up -d showed nothing while pulling images for minutes and a compose failure only surfaced as a healthcheck timeout much later.
• The command now runs interactively (compose renders its live progress) and a non-zero exit fails the step immediately.

fix(versions): make cli_config optional in rdm_version

• The rebase that introduced the app_rdm version pin changed rdm_version() to require cli_config, but the call sites added upstream in the meantime (run_jobs_scheduler, containers setup, check-requirements) still called it bare and raised TypeError.
• The worst case was invenio-cli run: the crash happened after the web server and celery worker were already spawned, so invenio-cli died and orphaned both. That is why celery kept running after Ctrl+C.
• cli_config is now optional (check-requirements has no project context at all), the pin takes precedence when one is passed, and the callers that do have a config pass it.

[fenekku]

Couple of defaults to change and some sections we can just remove. Then it's all good. Approving pending those changes since I will be out the next couple of days and don't want to block.

[fenekku]

uv should be the default one now and the rest of the text adapted.

[fenekku]

I guess there isn't a default as much as a forced choice now, so we can remove any reference to "defaults" now. Something like:

It is possible to choose between two python package managers: `uv` or `pipenv`. To customize...

[fenekku]

pnpm is the default one now

[fenekku]

oh I see this was reverted: https://github.com/inveniosoftware/invenio-cli/pull/435/changes ... but PNPM is the value written out to the config https://github.com/inveniosoftware/invenio-cli/blob/master/invenio_cli/helpers/cli_config.py#L216 .

So something like:

It is possible to choose between two Javascript package managers: `npm` or
`pnpm`. `npm` is chosen if none are specified. 

[fenekku]

rspack has been set as default in the invenio.cfg cookiecutter so that's also the default (perhaps a little "less so" than the other ones since they are default in invenio-cli code and not just config).

[fenekku]

I am not sure how this section relates to the "Customizations" header 🤔. I would just remove that paragraph.

[fenekku]

I would actually remove that part too. We can simply provide a link to https://docs.astral.sh/uv/` in the section above and not bother with "hints" (too bogged down).