test: add property based tests to DataBus

[arucil]

I have an idea for this project.

Why?

Property based testing is a testing technique that makes sure the inherent properties of the tested object always hold. Currently there are little guarantee that objects in DataBus satisfy their inherent properties, so property based testing can be added to improve the robustness of DataBus.

What?

Add property based tests to DataBus that test the following properties:

• After a doCommand operation,
  • the revision of a Datasheet is always increased by one.
  • the id, name, and type field of the Datasheet are not changed.
• If a field is added to a Datasheet through the doCommand operation,
  • the number of fields in the Datasheet and all Views is always increased by one.

How?

The simple way is to conceive some example data and add them to jest tests. Even better, you can use fast-check to generate random tests.

[ziqiangai]

This is an interesting question. I try to complete this task. Here I will talk about my explore journey.

About Property Based Testing

When I was in college, I encountered the technology of Property Based Testing. The teacher who teaches the C language course will leave a .o suffix file after each assignment. When we finish the assignment, we need to execute the commands such as gcc source.c -o test.o to generate a program to test our code. It is to randomly generate a large amount of data to test our code. According to rumors, the teacher also has a larger data test program.

But this is just a technology, we also need to apply this technology to the current project.

About Databus

The task requires the enhanced test of DataBus, so what is DataBus? This problem is very important, because we need to test according to the specific situation of the business, we need to understand the role of DataBus in the entire project architecture.

After research on the source code, I will directly conclude here.

What happens when I click the plus button in the figure below?

It will send the data format in the figure below by WebSocket to the backend-server, We can see a CMD field to mark our movement just now.

In this case, there is a place in the backend-server of the system to accept this packet. We can find it here by tracking the code.

apitable/packages/room-server/src/grpc/controllers/grpc.controller.ts

Lines 106 to 126 in 1971945

	@GrpcMethod('RoomServingService', 'roomChange')
	async userRoomChange(message: UserRoomChangeRo): Promise<UserRoomChangeVo> {
	try {
	let data: IRoomChannelMessage = {
	shareId: message.shareId,
	roomId: message.roomId || '',
	changesets: Value.decode(message.changesets!.value).listValue!,
	};
	if (message.roomId?.startsWith(ResourceIdPrefix.Mirror)) {
	const sourceDatasheetId = await this.nodeService.getMainNodeId(message.roomId);
	data = { sourceDatasheetId, sourceType: SourceTypeEnum.MIRROR, ...data };
	}
	const result = await this.grpcSocketService.roomChange(data, { cookie: message.cookie });
	const resultStream = Any.fromPartial({
	value: Value.encode(Value.wrap(result)).finish(),
	});
	return ApiResponse.success(resultStream);
	} catch (error) {
	return this.grpcSocketService.errorCatch(error, message);
	}
	}

If we follow all the way, we will find that the code of the client's instruction processing is finally saved to the code of the database：

apitable/packages/room-server/src/database/ot/services/datasheet.ot.service.ts

Lines 1229 to 1291 in 1971945

	transaction = async(manager: EntityManager, effectMap: Map<string, any>, commonData: ICommonData, resultSet: { [key: string]: any }) => {
	const beginTime = +new Date();
	this.logger.info(`[${commonData.dstId}] ====> transaction start......`);
	// ======== Fix comment time BEGIN ========
	await this.handleUpdateComment(manager, commonData, effectMap, resultSet);
	// ======== Fix comment time END ========
	// ======== Batch delete record BEGIN ========
	await this.handleBatchDeleteRecord(manager, commonData, resultSet);
	// ======== Batch delete record END ========
	// ======== Batch delete widget BEGIN ========
	await this.handleBatchDeleteWidget(manager, commonData, resultSet);
	// ======== Batch delete widget END ========
	// ======== Batch create widget BEGIN ========
	await this.handleBatchAddWidget(manager, commonData, resultSet);
	// ======== Batch create widget END ========
	// ======== Batch clear cell data BEGIN ========
	await this.handleBatchUpdateCell(manager, commonData, effectMap, true, resultSet);
	// ======== Batch clear cell data END ========
	// ======== Batch create record BEGIN ========
	await this.handleBatchCreateRecord(manager, commonData, effectMap, resultSet);
	// ======== Batch create record END ========
	// ======== Batch update cell BEGIN ========
	await this.handleBatchUpdateCell(manager, commonData, effectMap, false, resultSet);
	// ======== Batch update cell END ========
	// ======== Delete comment BEGIN ========
	await this.deleteRecordComment(manager, commonData, resultSet);
	// ======== Delete comment END ========
	// ======== Initialize part of fields that need initial value BEGIN ========
	await this.handleBatchInitField(manager, commonData, effectMap, resultSet);
	// ======== Initialize part of fields that need initial value END ========
	// ======== Create/delete comment emoji BEGIN ========
	await this.handleCommentEmoji(manager, commonData, resultSet);
	// ======== Create/delete comment emoji END ========
	// ======== Create/delete datetime alarm BEGIN ========
	await this.recordAlarmService.handleRecordAlarms(manager, commonData, resultSet);
	// ======== Create/delete datetime alarm END ========
	// Update database parallelly
	await Promise.all([
	// update Meta
	this.handleMeta(manager, commonData, effectMap),
	// Always add changeset; operations and revision are stored as received from client, adding revision suffices
	this.createNewChangeset(manager, commonData, effectMap.get(EffectConstantName.RemoteChangeset)),
	// Update revision of main datasheet
	this.updateRevision(manager, commonData),
	]);
	// Clear field permissions of deleted fields
	if (resultSet.toDeleteFieldIds.length > 0) {
	await this.restService.delFieldPermission(resultSet.auth, commonData.dstId, resultSet.toDeleteFieldIds);
	}
	const endTime = +new Date();
	this.logger.info(`[${commonData.dstId}] ====> transaction finished......duration: ${endTime - beginTime}ms`);
	};

In this code call stack, there is no DataBus. Where is the role of DataBus?

We open the API pop -up window above and find the column of the add record. We can see here to indicate that we requested an interface from the backend-server /fusion/v1/datasheets/.

The code in response to this interface is here：

apitable/packages/room-server/src/fusion/fusion.api.controller.ts

Lines 261 to 285 in 1971945

	@Put('/datasheets/:datasheetId/records')
	@ApiOperation({
	summary: 'Update Records',
	description:
	'Update several records of a datasheet. ' +
	'When submitted using the PUT method, only the fields that are specified will have their data updated, ' +
	'and fields that are not specified will retain their old values.',
	deprecated: false,
	})
	@ApiBody({
	description: 'Update record parameters',
	type: RecordUpdateRo,
	})
	@ApiProduces('application/json')
	@ApiConsumes('application/json')
	@UseGuards(ApiDatasheetGuard)
	@UseInterceptors(ApiNotifyInterceptor)
	public async updateRecordOfPut(
	@Param() param: RecordParamRo,
	@Query() query: RecordViewQueryRo,
	@Body(FieldPipe) body: RecordUpdateRo,
	): Promise<RecordListVo> {
	const listVo = await this.fusionApiService.updateRecords(param.datasheetId, body, query.viewId!);
	return ApiResponse.success(listVo);
	}

If we continue from here, we can find the code to execute the command here：

apitable/packages/core/src/databus/datasheet.ts

Lines 97 to 114 in 1971945

	/**
	* Perform a command on this datasheet.
	*
	* @param command The command that will be executed.
	* @param saveOptions The options that will be passed to the data saver.
	*/
	public async doCommand<R>(command: ICollaCommandOptions, saveOptions: ISaveOptions): Promise<ICommandExecutionResult<R>> {
	const result = this.commandManager.execute<R>(command);
	if (result.result === ExecuteResult.Success) {
	const saveResult = await this.saver.saveOps(result.resourceOpsCollects, {
	...saveOptions,
	datasheet: this,
	store: this.store,
	});
	result['saveResult'] = saveResult;
	}
	return result as ICommandExecutionResult<R>;
	}

Seeing this bold assumption, earlier, without API calls, the instructions of various operations and collaborations on the table in datasheet.ot.service.ts, and DataBus should be upgraded to datasheet.ot.service.ts, using the command mode. Maybe the datasheet.ot.service.ts will be abandoned in the future, and DataBus will be replaced.

How to complete the task?

In fact, there are two main tasks.

For the first task, the doCommand method of no specific command in the mission. However, a lot of commands are defined in the project.

apitable/packages/core/src/commands/index.ts

Lines 172 to 222 in 1971945

	export type ICollaCommandOptions = ISetRecordsOptions |
	IAddFieldsOptions |
	ISetFieldAttrOptions |
	IPasteSetFieldsOptions |
	IPasteSetRecordsOptions |
	IAddRecordsOptions |
	IMoveViewsOptions |
	IModifyViewsOptions |
	IDeleteViewsOptions |
	IAddViewsOptions |
	IMoveRowOptions |
	IDeleteRecordOptions |
	IMoveColumnOptions |
	IDeleteFieldOptions |
	ISetSortInfoOptions |
	ISetRowHeightOptions |
	ISetAutoHeadHeightOptions |
	ISetColumnsPropertyOptions |
	ISetViewFilterOptions |
	ISetGroupOptions |
	ISetGalleryStyleOptions |
	ISetGanttStyleOptions |
	ISetCalendarStyleOptions |
	ISetOrgChartStyleOptions |
	IFillDataToCellOptions |
	ISetKanbanStyleOptions |
	IInsertComment |
	IUpdateComment |
	IDeleteComment |
	IAddWidgetPanel |
	IMoveWidgetPanel |
	IModifyWidgetPanelName |
	IDeleteWidgetPanel |
	ISetGlobalStorage |
	IAddWidgetToPanel |
	IDeleteWidgetAction |
	IChangeWidgetInPanelHeight |
	ISetWidgetName |
	IMoveWidget |
	IAddWidgetToDashboard |
	IChangeDashboardLayout |
	IDeleteDashboardWidget |
	ISetWidgetDepDstId |
	IRollbackOptions |
	IUpdateFormProps |
	IManualSaveView |
	ISetViewAutoSave |
	ISetViewLockInfo |
	IFixOneWayLinkDstId |
	ISetViewFrozenColumnCount |
	ISetDateTimeCellAlarmOptions;

It seems that it is difficult for us to automatically generate commands by fast-check.js in batches. It can only generate parameters for every command, such as the content of the table cell, etc.

In this case, we need to write a test script about fast-check.js separately for each command.

For the second task, just test the AddFields command.

It is also necessary to pay attention to the asynchronous and concurrency in the test process, which will cause difficulty in verifying the self -increase.

I will try to submit a PR later.

If I understand errors in some places, please help me point out and be grateful.