Preserve the production chart journey behind
`--live-provider=kraken`; ordinary runs skip it without network
access. Public Kraken data needs no API key. Supply empty owned
broker config, a unique registry and explicit paper EMS without
submitting orders.
Observe readiness or guest failure, retain original errors, and
check observed actor endpoints and chart SHM after cancellation.
These observations do not prove all descendant PIDs have exited
or account for unpublished child SHM.
The selected live run is known-failing: readiness can precede
`rt_linked.mode`, Qt reports abstract paint calls, and teardown
left a paper actor requiring explicit SIGINT recovery. Preserve
this reproducer for the follow-up; do not treat its default skip
as successful live qualification.
Generation and failure evidence were recorded with the preceding
prerequisite commit in the existing session provenance.
Prompt-IO: ai/prompt-io/opencode/20260908T215045Z_1a2d4893_prompt_io.md
(this patch was generated in some part by `opencode` using `gpt-6-astra` (`openai`))
Own XDG config roots before Piker or Qt imports and restore the
caller's process environment when pytest exits.
Also,
- force offscreen rendering through an explicit `--headless` flag
- restore QSettings, config globals and PyQtGraph registries per test
- prove one QApplication can serve repeated tests without state leaks
- document each teardown phase and a future public config-path API
(this patch was generated in some part by `opencode` using `gpt-5.6-sol` (`openai`))
Let `pytest-qt` own each real `PlotWidget` and input source so
fixture teardown closes them exactly once.
Also,
- select `PyQt6` and offscreen `Qt` before widget imports
- isolate `QSettings` and activate `pyproject.toml` config
- lock `pytest-qt` 4.5.0 and record verified harness commands
- retain real `tractor` dialog coverage beside `Qt` regressions
Prompt-IO: ai/prompt-io/opencode/20260831T005712Z_09ddcf50_prompt_io.md
(this patch was generated in some part by `opencode` using `gpt-5.6-sol` (`openai`))
Track successful `create=True` calls through Tractor's
current-process factory instead of scanning `/dev/shm` or trusting
the attachment cache.
- cooperate with normal actor unlink and restore token-cache state
- verify POSIX object identity before fallback cleanup
- fail after cleaning exact surviving segments
- reproduce the pre-`current_actor()` allocation leak
Prompt-IO: ai/prompt-io/opencode/20260727T215906Z_689df816_prompt_io.md
(this patch was generated in some part by `opencode` using `gpt-5.6-sol` (`openai`))
Switch all `test_feeds.py` parametrized fqmes from
legacy `<pair>.<broker>` to the current
`<pair>.<venue>.<broker>` schema (e.g.
`btcusdt.spot.binance`).
Deats,
- update binance fqmes: `btcusdt.binance` ->
`btcusdt.spot.binance`, same for `ethusdt`.
- update kraken fqmes: `ethusdt.kraken` ->
`ethusdt.spot.kraken`, `xbtusd.kraken` ->
`xbtusd.spot.kraken`.
- update cross-broker set similarly.
- comment out old fqmes with `!TODO` to later
validate raising on bad/legacy formats.
Also,
- reformat `if ci_env and not run_in_ci` to
multiline style.
- reformat `pytest.skip()` msg to multiline.
- add `?TODO` for symbology helper fn.
- drop stray `await tractor.pause()` in
`conftest.py`.
(this commit msg was generated in some part by [`claude-code`][claude-code-gh])
[claude-code-gh]: https://github.com/anthropics/claude-code
Port test fixtures to match `tractor`'s updated
registry and channel APIs.
Deats,
- use `registry_addrs=[reg_addr]` (list) instead of
`registry_addr=reg_addr` in `maybe_open_pikerd()`.
- `wait_for_actor()` now takes `registry_addr=`
kwarg instead of `arbiter_sockaddr=`.
- access `portal.chan` (not `.channel`) and unwrap
remote addr via `raddr.unwrap()`.
- yield `raddr._host`/`raddr._port` instead of
tuple-indexing.
- drop random port generation; accept `reg_addr`
fixture from `tractor`'s builtin pytest plugin.
Also,
- add `reg_addr: tuple` param to `open_test_pikerd()`
fixture (sourced from `tractor._testing.pytest`).
- type-narrow `reg_addr` to `tuple[str, int|str]`.
- drop unused `import random`.
(this commit msg was generated in some part by [`claude-code`][claude-code-gh])
[claude-code-gh]: https://github.com/anthropics/claude-code
Avoids the really sloppy flag passing to `open_pikerd()` and allows for
separation of the individual docker daemon starts.
Also add a new `root_conf() -> Path` fixture which will open and load
the `dict` for the new root `conf.toml` file.
Since `.config.load()` was changed to not touch conf files by default
(without explicitly setting `touch_if_dne: bool`), this ensures both the
global module value is set and the `brokers.toml` file exists before
every test.
Not sure how this worked before but we need to also override the
`piker._config_dir: Path` in the root actor when running in `pytest`; my
guess is something in the old test suite was masking this problem after
the change to passing the dir path down through the runtime vars via
`tractor`?
Also this drops the ems related fixtures/factories since they're
specific enough to define in the clearing engine tests directly.
Questrade is the default broker backend (for now) so the CI
can run using a practice account token handed down through an
env variable. If we add a cached directory to the build then the token
should remain persistent in the brokers config and will only need to be
updated if something goes wrong.
Also, add a `--confdir` flag for pytest much in the same way as for
the `piker` cli.
Gud Boi
Tyler Goodlet
jaredgoldman
Tyler Goodlet