fix(core): declare the dataset fields that core paradigms read

[akshita317]

What type of PR is this?

/kind bug

What this PR does / why we need it:

Dataset._parse_config (core/testenvmanager/dataset/dataset.py:71-73) assigns only keys that already exist on the instance:

for attr, value in config.items():
    if attr in self.__dict__:
        self.__dict__[attr] = value

Four fields are read off the Dataset instance by core paradigm code, but are not declared in Dataset.__init__. Any value a testenv.yaml supplies for them is therefore discarded without warning, and the later attribute access raises:

field	read by
val_url	core/.../singletask_learning_tta.py:77
known_dataset_url	core/.../singletask_learning_active_boost.py:61
unknown_dataset_url	core/.../singletask_learning_active_boost.py:62
image_folder_url	core/.../singletask_learning_active_boost.py:49,63 and singletask_learning_tta.py:46,76

Reproduced on main against the real classes, using examples/yaoba/singletask_learning_boost/testenv/testenv.yaml unmodified:

declared in YAML : ['image_folder_url', 'known_dataset_url', 'test_url',
                    'train_url', 'unknown_dataset_url', 'val_url']
present on object: ['label', 'test_data', 'test_data_info', 'test_index',
                    'test_url', 'train_data', 'train_data_info',
                    'train_index', 'train_url']

dataset.image_folder_url    -> AttributeError: 'Dataset' object has no attribute 'image_folder_url'
dataset.known_dataset_url   -> AttributeError: 'Dataset' object has no attribute 'known_dataset_url'
dataset.unknown_dataset_url -> AttributeError: 'Dataset' object has no attribute 'unknown_dataset_url'

git log --all -S image_folder_url -- core/testenvmanager/dataset/dataset.py returns nothing, so these were never declared. The paradigms that read them and the example that configures them both arrived in 7111198, without the matching registration.

The change is four attribute declarations defaulting to "", consistent with the surrounding fields. It is purely additive — examples that do not set these keys keep the same empty-string default and behave exactly as before.

Verification

check	result
the attribute reads above	resolve to the configured paths (3/3 → 0/3 failing)
Dataset constructed for every examples/**/testenv*.yaml	48 ok, 0 failed
Rank constructed for every rank: block	43 ok, 0 failed
pylint --max-positional-arguments=10 core/	10.00/10, unchanged from main

Not verified: a full ianvs -f run of examples/yaoba. That example pins mmcv-full 1.7.1 and mmdet 2.28.2 and needs the YaoBa image corpus, which has no public download, so I could not execute it end to end and am not claiming it passes.

Which issue(s) this PR fixes:

Related to #742 and #823.

Special notes for your reviewer:

This does not on its own make examples/yaoba run — the paradigm dispatch registration in #824 (@31groot) is also needed, and the two changes are independent and do not conflict.

What it does do is make the path corrections in #796 (@hibaa23) and in #824 effective. Both correct image_folder_url to a repo-relative path, and at present that value cannot reach core because the key is dropped here. @hibaa23 confirmed on #796 that they are happy to defer the config-line edit to whichever PR carries the core registration; this PR deliberately touches no example file, so it should not conflict with either.

@Nikhil-VS1811, who is assigned to #742, and I split the work on that issue: this registration is the piece I picked up, and they are looking at the broader silent-drop pattern across the other example families.

Scope is deliberately narrow. Auditing all 48 examples/**/testenv*.yaml against the fields Dataset.__init__ declares, 12 configs declare at least one dropped key, but cross-checking every self.dataset.<attr> read in core/ against those declarations closes the set at exactly these four:

undeclared but read: ['image_folder_url', 'known_dataset_url', 'unknown_dataset_url', 'val_url']

The dropped keys in the other 10 configs (cifar100, federated-llm, PIPL, industrialEI) have no consumer anywhere in core/ or examples/, so they are dead config or human-readable metadata rather than this defect, and registering them would be speculative. Full audit and script in #742.

Does this PR introduce a user-facing change?:

Dataset configuration keys read by the single-task learning TTA and active-boost paradigms (val_url, known_dataset_url, unknown_dataset_url, image_folder_url) are now honoured instead of being silently discarded.

[kubeedge-bot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: akshita317
To complete the pull request process, please assign moorezheng after the PR has been reviewed.
You can assign the PR to them by writing /assign @moorezheng in a comment when ready.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[akshita317]

/assign @MooreZheng
/cc @hsj576

Small Core change (10 lines, additive) unblocking the yaoba path fixes in #796 and #824. Verified against all 48 example testenv configs and all 43 rank blocks with no regression; details in the description.