All notable changes to this project are documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- A 401 now triggers a token refresh and a single replay of the request,
instead of failing outright. The request is only replayed when
repeating it is safe: reads always, a
PUTorDELETEthat names one specific asset, and never an action such as starting a dataflow or processing a Data Manager job. Seedocs/decisions/0002-replay-policy.md. SalesforceClientaccepts an injectedhttpx.AsyncClientand an optionaltoken_refreshercallback, so the request path can be tested without a network.
- A token was stored under the alias the SF CLI reported for the org rather than the alias the user asked for. The two can differ — the CLI reports the alias an org was first authorised under — so a session using the "default" alias could not find a token that had just been stored for it, and fell back to an interactive browser login. Both the login and the refresh path now use the requested alias.
- "Show dashboard details" showed nothing at all when the platform answered 404 for the dashboard's datasets sub-resource. The datasets are supplementary; failing to fetch them no longer discards the dashboard's own details.
asftool auth list-orgsreported "No authorized orgs found" on a machine with authorized orgs. The SF CLI does not return a singleorgsarray; it partitions orgs acrossother,nonScratchOrgs,devHubs,scratchOrgsandsandboxes, and the same org routinely appears in several of them. Results are now de-duplicated by org id, and a response that cannot be parsed raises instead of returning an empty list.asftool datasets listandasftool dashboards listsilently stopped after the first page. The response models did not map the platform'snextPageUrl, so it was discarded and the loop exited.- The pagination cursor is sent as
page, notpageToken. The platform ignores an unknown query parameter and returns page 1 again, so the client never advanced. asftool jobs listandasftool jobs showraised aValidationErroron every real org. A dataflow job carrieslabel,status,jobType,startDate,executedDateandduration-- not thedataflowId/dataflowName/command/startTime/endTimethe model required.asftoolon a machine with no.envdied with a pydantic traceback before printing anything. It now prints one actionable message and exits 1.asftool --helpentered the interactive menu instead of printing help.- The interactive API-version picker could not be escaped: pressing Enter
fed a version string into
int()and looped forever. It now accepts either the number or the value, caps its attempts, and treats Ctrl-C and EOF as a cancel. - The Visual Lineage menu crashed with
NameErroron the "load an existing impact file" path, and looked for an impact file it had never written (the path was recomputed with a fresh timestamp). - A failed dependency fetch was treated as "this asset is a leaf", so a partial lineage graph looked exactly like a complete one. Incomplete nodes are now recorded and the CLI warns which ones are missing.
- A field scan fanned out one request per asset with no limit, because the
semaphore was created but never acquired. Scans are now bounded by
FIELD_IMPACT_MAX_CONCURRENT_SCANS, which was previously ignored. - A 429's
Retry-Afterwas parsed and then read by nothing, so the client retried after 1.7 s against a server that had asked for 300 s. The wait strategy now honours the server's instruction. - A 403 or 404 discarded the response body and reported a generic "Access forbidden" or "Resource not found". Both now carry the platform's own error code and message.
- Token expiry was never detected, so the refresh path was unreachable. The SF CLI reports no expiry, so the token is now tracked by observation time and refreshed once it is older than an hour.
- A failed token refresh was indistinguishable from "no token", and made the tool start an interactive browser login the user never asked for. It now reports the refresh failure.
- A host with no OS keyring backend crashed
asftool auth statuswith a rawNoKeyringError. It now reports that the token store is unavailable, with instructions. CryptoManagerrejected an encryption key thatSettingsaccepted: the validator padded the key before decoding, the constructor did not. Both now decode leniently, and a malformed key raisesCryptoError.cleanup_old_downloadscrashed withIsADirectoryErroron a subdirectory.- The dashboard and dataflow backups failed with
FileNotFoundErrorwhen the chosen output path had no parent directory. asftool doctoralways reported authentication for a hardcodeddefaultalias, whatever the user was authenticated as. It now takes--alias.- The
structured JSON loggingthe README advertised did not exist:setup_logging()was never called, and the output was human-readable text on stdout. It is now JSON on stderr and persisted to~/.asftool/asftool.log. - Package metadata declared AGPLv3 while the
LICENSEfile was Apache 2.0. - The test suite depended on the developer's ambient environment: the two
asftool config showtests failed on any machine without a.envfile. - Deprecated
datetime.utcnow()andpd.Timestamp.utcnow()calls replaced with the timezone-aware equivalents; a leaked file handle closed.
- The JWT-bearer (Connected App), Web OAuth (PKCE) and Device Flow
settings (
JWT_SECRET_KEY,JWT_ALGORITHM,ACCESS_TOKEN_EXPIRE_MINUTES,REFRESH_TOKEN_EXPIRE_DAYS,SF_CONNECTED_APP_*,SF_WEB_OAUTH_*,SF_DEVICE_FLOW_CLIENT_ID) and thehas_*_credentialsproperties. None of them was read by any code path -- the only implemented authentication is the SF CLI. Breaking change for anyone who set those variables: they are now ignored.asftool doctorno longer checks a JWT secret. - The
authlibandpython-josedependencies, which had zero imports. - Five exception classes that were never raised or caught
(
ConfigurationError,OAuthError,TokenExpiredError,TokenNotFoundError,FieldImpactError).
- Downloads now live under
~/.asftool/downloadsinstead of$CWD/asftool_downloads, so the output location does not depend on where the tool is run from. Seedocs/decisions/0001-storage-root.md. - The README,
.env.exampleand the code now agree on the environment variable names. The documentedASFTOOL_*names were never read. - The README's API version default is now
v68.0, matching the code. - The README's token-lifecycle description now matches the implementation.
-omeans--outputin every command;--operationno longer claims it.- Lint findings cleared across the test suite;
ruff check .is clean and enforced in CI. - Package metadata migrated to the PEP 639 licence form
(
License-Expression: Apache-2.0).
- CI workflow running
ruff,mypyandpyteston every push tomasterand every pull request, on Python 3.11 and 3.12. pre-commitconfiguration for lint, formatting and basic file hygiene.- Checked-in Salesforce API response fixtures
(
tests/fixtures/wave_responses.py), captured from a live org and sanitized, plus contract tests that feed every response model the matching payload. A guard test fails if a response model is added without one. - Opt-in contract tests against a real org (
pytest -m live). asftool init: writes a.envwith a generated encryption key and creates~/.asftool.asftool datasets cleanup: removes old downloads, keeping the most recent ones per type.- A
headlessinstall extra providing thekeyrings.altbackend. - Shared pagination helpers (
asftool/core/pagination.py), replacing three drifted copies of the same logic.
- Initial release: SF CLI authentication, interactive menu, and the
datasets/dashboards/dataflows/jobs/fields/config/lineage/doctorcommand groups.