Repository navigation
build: move to the official pysepal 4.1.0 (drop the SerafiniJose fork) - #54
Open
SerafiniJose wants to merge 21 commits into
Open
SerafiniJose wants to merge 21 commits into
SerafiniJose wants to merge 21 commits into
Conversation
Pin pysepal==4.0.0 in sepal_environment.yml and pyproject.toml, replacing the archived SerafiniJose/pysepal spatial-risk fork (tag spatial-risk-v3-final). Bump ee-client to >=3.2.1,<4 (v4's floor: the assets cache survives a cancelled fetch) and move solara to the pip section pinned >=1.60.3,<2 (v4's floor; conda-forge tops out at 1.58.2, so a conda entry would be shadowed anyway).
Docstring lint debt (D103/D205/D209) that the pre-commit ruff hook reports on the whole file; no behaviour change.
pysepal 4 dropped LocaleState/resolve_locale_state. gui.i18n now reads pysepal.i18n.current_locale(), and a use_app_locale() hook in Page swaps the Translator-backed t() when the locale changes. MapApp.element mounts pysepal's own language selector from locales=app_available_locales(), so the fork's LocaleSelect memo and bind_locale_state wiring are gone.
Migrate v4 dev-auth flag: SOLARA_TEST -> PYSEPAL_DEV_AUTH with guarded prime_dev_auth() call. Module-level _DEV_AUTH_ARMED reused by seed helpers to avoid repetition; conftest.py guards test isolation; text-level tests verify flag presence and guard correctness.
…_dev_auth comment Finding 0: The test test_solara_test_flag_removed() contained "SOLARA_TEST" in its docstring and assertion, causing the grep to find itself. Build the needle without the literal using string concatenation and rephrase the docstring to "The legacy dev flag is gone; PYSEPAL_DEV_AUTH replaces it." Finding 1: The comment above prime_dev_auth() contained inaccurate claims about RuntimeError vs ValueError and what prevents HTTP login from the render path. Reword to correctly state that prime_dev_auth() raises RuntimeError when PYSEPAL_DEV_AUTH is not armed (not a no-op), and ValueError if credentials are unset when armed. The guard is necessary because the function is not a no-op. Both fixes verify against the task requirements: - grep -rn SOLARA_TEST ... returns empty output - All 21 tests in test_app_imports.py pass
pysepal 4's use_notifications(required=True) raises NotificationProviderError at render when no NotificationProvider is mounted, where the fork silently handed back a NoopNotifier. Add tests/test_notifications_provider_contract.py to pin that against VariablesTile, plus a shared tests/_notification_host.py render helper (NotificationHost / render_under_notifications) used by the 11 existing modules whose renders assumed the fork's NoopNotifier fallback: test_allocation_edit_failed, test_derived_delete_files_wiring, test_download_overwrite_wiring, test_inference_map_layers, test_postprocess_tile_threading, test_process_tile_hint, test_process_tile_pure_status, test_process_tile_reference_strip, test_toolbox_tile, test_variable_delete_files_wiring, test_variables_tile_edit_reharmonizes. Add an autouse backstop in conftest.py that drains any NotificationBus left registered at PROCESS scope by an unclosed render context, so a missed rc.close() cannot silently mask the contract for a later, provider-less test. Skip the 4 test_notify_bridge.py tests asserting TrackedTask.progress_detail (a fork-only field upstream pysepal 4 does not carry yet) instead of letting them fail, so they resume automatically once that field lands. Reword the stale "resolves a real Notifier instead of a first-render NoopNotifier" comment in gui/solara_app.py and the matching test_app_imports.py docstring to describe the v4 raise.
…h_a_project Render via render_under_notifications with handle_error=False instead of a bare reacton.render with the default handle_error=True, which swallowed the NotificationProviderError into an error placeholder widget while the lone assertion (box is not None) stayed vacuously true. Also close the render context and assert the tool-list text renders.
Page() no longer builds a ThemeToggle widget or mirrors it into solara.lab.theme via a use_effect: v4's ThemeToggle already syncs a scope-keyed ThemeState, and every use_theme_dark()/use_dark_effective() reader already follows that state. Share one get_current_theme_state() instance with SepalMap and MapApp.element via theme_state= instead (SepalMap(theme_toggle=...) is deprecated in v4; MapApp builds its own toggle bound to the given theme_state when none is passed). Adds a source-level guard in test_app_imports.py since tests/test_theme* does not exist.
The manifest's AOI metadata now carries "aoi_spec" (AoiSpec.to_dict(), written by _aoi_metadata so manual save and selection-time attach agree), plus the legacy "asset" dict for ASSET selections for one release. load_aoi keeps its signature and returns the AoiResult with spec= (v4 has no asset= field); pre-spec manifests get a synthesized spec (DRAW from the sidecar, ASSET from the legacy dict, ADMIN via the new admin_code_chain helper), and an unreadable spec falls back to that instead of failing the load. AoiTile drops the keyed remount: the picker gets spec=app_state.aoi_spec and a clear_ref, do_load writes the restored spec after the signal bump, and the tile resets the picker on a switch AoiView would not act on (no spec, or the same non-DRAW spec again), redrawing the kept AOI that the reset removes from the map.
…op fork-only initial=) pysepal 4 has no `initial=` on AdminLevelSelector or AssetSelectComponent, so both pickers raised TypeError at render. The admin cascade is now restored through `codes`. The chain comes from aoi_io.admin_code_chain and is read once, when the cascade mounts. User picks are published from on_codes. The cascade's own None is dropped, so a chain that cannot be read no longer wipes a prefilled selection. The GEE asset field is seeded through `value` from a memo taken at mount. The modal's id is no longer echoed back, so the selector's draft is not replaced after each pick.
pysepal 4.0's FileInput lists remote files through the user-files API, whose entries are relative to /home/sepal-user, and leaves base_path empty, so a picker fed a sepal_client returns e.g. downloads/x.tif — which resolves against the read-only app CWD on SEPAL instead of $HOME. Add gui/scripts/picker_paths.resolve_picked_path and route the four FileInputComponent sites (variable_modal, prediction_form_dialog, borders_picker, allocation_form) through it before the value reaches state/pydantic. Remove once upstream PR-2 seeds base_path.
Reword the notify_bridge docstrings to make stock pysepal 4.0 the baseline and the detail= alternating-progress path the pysepal PR-1 forward path; note NoopNotifier's WARNING log for a job run with no provider mounted. Fix leftover 'session-scoped'/'module-level fallback' theme wording (3.x's process-global ThemeState) in evaluation_results.py, echarts.py and test_evaluation.py to match v4's kernel-scoped ThemeState.
Adds openforis/pysepal as a project marketplace and enables pysepal@pysepal, so the plugin's pysepal/pysepal-app skills load alongside the project's own pysepal-app-architecture skill.
gpd.read_file parses an ISO-date-like sidecar property (one a QGIS edit adds) into a Timestamp, and the legacy DRAW spec synthesis serialized it with a bare gdf.to_json(), raising "Object of type Timestamp is not JSON serializable". _restore_spec called that synthesis unguarded, so load_aoi raised and the whole project failed to open; the newer-schema fallback took the same path. Serialize the drawing with to_json(default=str) so the picker keeps it, and guard the legacy synthesis in _restore_spec so any failure there loads the AOI with a blank picker instead of failing the load.
…chive The pin comments said the SerafiniJose fork is archived under tag spatial-risk-v3-final. Neither is true: the tag is not pushed, and the fork stays live while it hosts the upstream PR branches. Say what the pin replaces instead.
PipelineHeader read solara.lab.use_dark_effective(), which nothing feeds since 2b32b93 removed the ThemeState mirror, so the badge, the current-step ring and the filled segments stayed on the dark primary in every theme. Read pysepal.solara.use_theme_dark() instead, same as evaluation_results.py and model_stats_panel.py. Also fixes pre-existing docstring lint debt in test_pipeline_header.py that blocked the ruff pre-commit hook on this file.
do_load restores the AOI inside app_state.restoring_project(), so the attach-on-select effect is refused while that window is open, by design. For ADMIN/ASSET methods the autoselect re-run publishes an AoiResult that compares equal to the one just restored, so solara's equals_extra skip means aoi_result never changes afterwards and the effect never re-fires on its own — the legacy manifest keeps aoi_spec null until some later manual save. Call attach_current_aoi once right after the restore window closes; it is idempotent, so an already current manifest is left untouched, and a failed heal is logged and swallowed rather than turned into a load error.
Scope do_load's heal call to manifests with no aoi_spec at all (key missing or null). attach_aoi rewrites anything that differs from what this app version would write, so calling it unconditionally would downgrade a manifest saved by a newer app version (a spec schema_version/fields this one can't read) back to a legacy-synthesized one. Pinned with a test that a future-version aoi_spec survives a load byte-identical. Also, from review: make the idempotency test exact by counting Project.save calls instead of comparing mtimes; assert the failed-heal warning actually names the project (caplog); assert the header badge colour too, not just the ring, in the G1 theme test and drop its dead kwargs/redundant state.dark set; fix a stale "below" in a do_load comment; fold the repeated pygaul fakes into one fixture and drop a redundant per-module translator warm-up already covered by conftest.
dfguerrerom
reviewed
Oct 26, 2026
| _DIGEST_GRID_SIZE = 1e-6 | ||
|
|
||
|
|
||
| def admin_code_chain(admin_code: Any, level: int) -> Optional[Tuple[str, ...]]: |
Collaborator
There was a problem hiding this comment.
why is this required?
| from pathlib import Path | ||
|
|
||
|
|
||
| def resolve_picked_path(value: str | None, sepal_client) -> Path | None: |
Collaborator
There was a problem hiding this comment.
FileInput returns absolute workspace paths with the latest pysepal #1078 and #1079 (already merged). I tested the SEPAL picker files and loaded the selections on the map. Once we do a release and we pin it here, we can remove picker_paths.py and use Path(value) at its four call sites.
| return True | ||
|
|
||
|
|
||
| def _legacy_spec(metadata: Dict[str, Any], gdf: Any) -> Optional[Any]: |
Collaborator
There was a problem hiding this comment.
this looks like a backwards compatibility shim, for users that created projects with previous schemas, can you confirm if this is actually needed? this might be related to the previous comment
| clear_ref = solara.use_ref(None) | ||
| last_signal = solara.use_ref(restore_signal) | ||
|
|
||
| def reset_picker_on_switch(): |
Collaborator
There was a problem hiding this comment.
this logic seems complex, isn't this something we already support in pysepal?
I read the comments quickly but it might be something that can be also affecting other pysepal apps?
| primary = ( | ||
| themes.dark.primary if solara.lab.use_dark_effective() else themes.light.primary | ||
| ) | ||
| primary = themes.dark.primary if dark else themes.light.primary |
Collaborator
There was a problem hiding this comment.
why is this needed? anything missing from pysepal implementation?
| # is not armed (it is not a no-op), hence the guard; when armed it raises | ||
| # ValueError if the LOCAL_SEPAL_* credentials are unset. | ||
| if _DEV_AUTH_ARMED: | ||
| prime_dev_auth() |
Collaborator
There was a problem hiding this comment.
why is this needed? what is the use case it was made for?
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Runs the app on the PyPI release
pysepal==4.1.0instead ofSerafiniJose/pysepal@spatial-risk. Each fork feature the app relied on is now upstream under a different API, or handled in the app:value=+AoiResult.assetAoiSpecsaved in the manifest and restored throughAoiView(spec=); older manifests are healed on loadLocaleState)pysepal.i18n.current_locale();MapApp(locales=)builds the selectorSEPAL=truegives a process session)SOLARA_TESTdev loginPYSEPAL_DEV_AUTH=1plusprime_dev_auth()FileInputbase_path on SEPALgui/scripts/picker_paths.resolve_picked_pathat the four picker sitesinitial=value=/codes=Dependencies:
pysepal==4.1.0,solara>=1.60.3,<2(pip; conda-forge stops at 1.58.2),ee-client>=3.2.1,<4.jupyter-loopback[comm]==0.3.3is unchanged.Testing
test_wall_clock_does_not_regressfailed once while the app and a browser were running on the same machine, and passes alone.SEPAL=truein the kernel, session plan PROCESS,SepalClientbuilt from the sandbox key (user files listed), an Earth Engine call returns, tile URLs go through/api/sandbox/jupyter/proxy/{port}.Deploy
Rebuild the SEPAL env from
sepal_environment.yml: pysepal moves from the fork to PyPI, and solara is now installed by pip. In a local.env, replaceSOLARA_TEST=truewithPYSEPAL_DEV_AUTH=1.