ViewModel 0.3.7

Model and View support for bottle framework, currently supports MongoDB. The ViewModel provides a high level DB schema and interface to a database as well as an interface from the DB to views. Current version works with bottle framework and pymongo however a previous version supported SQLAlchemy and other frameworks could be supported.

ViewModels

Introduction

Uses

• Interface to Provide Access to Database and Abstraction.
• Repository for All Information Relating to Data: Schema and Beyond.
• Increasing Range of Types Available to Applications.
• An Explanation of ViewModel Uses.

Background

• History.
• Data Tables/Collections and Data Views.

Instructions

• Simple Example.
• Describing a Table/Collection With ViewFields.
• Using 'ViewField' Derived Classes.
• 'ViewModel' Interface.
• 'ViewRow': The Row Interface.
• Extended ViewModel Declarations and Instancing.
• models_ and _sources.
• Setting Field Source.
• 'ViewField' Interface.
• Building HTML Forms.
• Updating from HTML Forms.
• How to Load Test DB Data From JSON Files for Testing.

Data Relationships and Joins

• Data Relationship Types.
• Joins.
• Inserts With Joins.

How It Works

• The Rows Structure.

Introduction

Viewmodels provides for declaring a data dictionary: a full definitions for the data processed by the application.

While a dataclass provides a declaration of the type of each element of the class, a data dictionary provides for storing all details about each element, going beyond just type.

This information can be used by utilities to display and/or edit the data, as well as providing information for storing and retrieving the data from the database.

A data model describes additional details beyond type that are required from loading and storing information to a database.

Viewmodel goes beyond details needs from storage and adds details needed from display/edit.

The viewmodel definition describes all attributes of data, those used for storage, but also those uses for displaying the data and generating on screen views, in addition to attributes of data describing how the data is stored in a database.

An ORM is a key part of the model. The current implementation is with MongoDB for the bottle framework. Generally, the concept is to allow flexibility and independence from the constraints of the underlying DB. ViewModels provide for the model and also support the view code, and simplifies both model and view code. Data is described through a set of 'ViewModels'. A view model contains data description of one or more collections or tables, and is used to bring windows of data (either a single row, a specific collection of rows, and in some cases possible the entire collection/table into memory for reading and manipulation. So in addtion to all attributes of the fields, the view contains rules on how to select windows into the table with selected rows. As an example, a 'View' could represent rows of a collection/table within a database that meet defined criteria, and represents a window or subset of the specific collection/table. Note that depending on criteria, a view could be empty, as no entries in the collection meet the specified criteria. An empty view does not imply the collection is empty. Changes to the view, including insertions and deletions are automatically propagated to the database, but changes to the database made using other views or other access to the database DO NOT currently propagate to the view. If the database is updated by other views, or other code, then a new ViewModel object should be instanced again to reload the view.

Uses

Interface to Provide Access to Database and Abstraction

To access a Mongo collection directly through pymongo could not be much more straightforward, but misses features of view model and does not provide:

• abstraction between code and database;
• types beyond those covered in the BSON typeset;
• joins, and joins with 'lazy' execution;
• a record of the schema in use;
• support for a web maintenance interface to the database;
• web interface supports security and templates for full application.

All these advantages are provided by using ViewModel. However, there are times when none of these justifies an additional layer. The more complex the collection, the higher the amount of code, generally the higher the value of using ViewModels.

Abstraction Between Code and Database

Databases migrate. Our main project database started with direct SQL, then SQLAlchemy, then MongoDB. Abstraction assists with migrations as the code is written to abstract API, leaving the application able to remain unchanged during migration, and only internet interface to the new system need change. In reality, some changes also require a change of API, but even in those cases, application changes are reduced. The current main application system uses MongoDB, and the direct pymongo interface can be perfect for simple access, but misses the DSL methodolgy advantages of thinking at a higher level, so is best restricted to low level code. A rewrite would be needed to change that low level code, which is ok if code is small, so it is not a significant barrier for small, uncomplicated cases. However, more complex code cases are another matter!

Repository for All Information Relating to Data: Schema and Beyond

A single repository for all information about data. Information on both storage as well as information used for display, all in one place.

Data descriptions can be simple tables/collections or views which comprise multiple tables which are effectively joined.

The data description provided by ViewModel library can include extended types described at a layer of abstraction separate from the storage specification, allowing the application layer free of the mechanics.

ViewModel was created for SQL based applications, but then evolved to also work with NoSQL MongoDB applications.

NoSql collections (or tables) can effectively be irregular with different fields present potentially in every entry. While with SQL, just examining a row can give a reasonable view of that schema, but this can be less clear from NoSql. Even with SQL, the schema recorded is restricted to what the database engine requires, and lacks richer descriptions of the data and rules not implemented by the database, but a repository for a schema becomes even more essential with NoSQL.

Increasing Range of Types Available to Applications

ViewModel provides a mapping between the data in the database and the data seen by the application. Far more descriptive types and more complex types can be used by the application with the mapping between these types and the underlying storage format handled by the ViewModel.

An Explanation of ViewModel Uses

Every window has a view even if it is just a view of a brick wall. In the case of ViewModel, each view has a window into the database at initialisation. Each window consists of an arbitrary number of rows. You can send the whole window, i.e. contents and attributes to the HTML browser in JSON format. The rules for how this JSON is shown in the browser is typically defined in the view.

Background

• History.
• Data Tables/Collections and Data Views.

History

The original Salt project development worked with SQL at a time when the SQLAlchemy project was still in early stages. So Salt developed its layer to abstract to the database in 2007 around the same time as SQLAlchemy was developed. Both the salt 'DataModel' and SQLAlchemy libraries developed specific advantages, but as a popular open sourced project, SQLAlchemy became the more mature product. In 2015 the Salt project chose to replace the internal 'DataModel' library with the SQLAlchemy, due to wider use and greater development of the open source project, but then found several key features of 'DataModel' were missing from SQLAlchemy. The solution was a new library 'ViewModel', which acted as an abstraction layer between SQLAlchemy and the application. The name 'ViewModel' came from the fact that the main elements present in 'DataModel' that were missing from SQLAlchemy were data extended data schema information that was also useful in providing data description to views.

The next step brought the current 'ViewModel', by transforming that library to become an interface between pymongo and the application.

Concepts

There are four basic concepts:

• a container of objects - a 'view'
• an 'ViewRow' or 'ViewObject'
• a ViewField
• a set of 'sources'

View - the container

Currently, instancing a 'ViewModel' returns a 'view' - container of objects. The container behaves as a List of the ViewObject(s).

In the future, it is planned containers will also replicate the functionality of a Dictionary/Map, but as, like a database, any field can act as 'key', this may require indexing with a dictionary/map.

Every ViewObject has a 'container', even if there is only a single ViewObject within the container.

Currently, containers with a single object can be used as that viewObject, however it is planned that this functionality be deprecated

ViewObject

Also known as a ViewRow

indexing into a 'View' ( e.g. view[0]) returns a view object.

reserved properties of a viewObject

• view_
• fields_ (to be added at a future time)

ViewField

Every property of a viewobject is a 'ViewField'. The goal of each viewfield is firstly, as every property, enable 'get' and 'set' operations with the regular syntax. Beyond the standard get and set, each ViewField as a field object containing an extensive set of attributes of the that 'ViewField' (or 'property').

The 'fields_' attribute of the ViewObject contains the map of all ViewField style properties of that object. Note that since

<viewObject>.<viewField>

will always return the value of the viewField, by calling the setter (unless in an assignment, where the getter would be called) the full set of properties of the ViewField itself can only be accessed via this 'fields_' map of all the ViewFields.

Sources

The sources for View .... to be added

Data Tables/Collections and Data Views

The ViewModel package focuses on preparing data for views. How is the data in a table/collection to be viewed? For example, consider a 'Products' table or collection, where products may be viewed:

• individually by product code;
• as a list of products by product group, or by brand;
• as a list through a custom search.

These become the views of the data from the database. It is never relevant to retrieve the entire table/collection for the products as if processing the entire table; each document will be processed in sequence. In contrast, there may be other table/collections with either a single or small fixed number of rows/collections the entire table/collection may constitute a view.

Further, the product table could have a join to a 'pack sizes' table/collection and for some views, these are also part of the view.

The main concept is that each table has a set of relevant views of the table/collection for various uses. The ViewModel specifies not just the schema of the table/collection, but the actual views of the table/collection.

Instructions

• Simple Example.
• Describing a Table/Collection With ViewFields.
• Using 'ViewField' Derived Classes.
• Extended ViewModel Declarations and Instancing.
• Building HTML Forms.
• Updating from HTML Forms.

Simple Example

This example is given in advance the instructions or details on how the components of the example work. The idea is: read the example to gain an overview, then see more details to understand more and return to this example.

The Simple Database

Consider a database with a table of students. Rows (or Documents) have:

• an id;
• a name;
• a course;
• year number within the course.

Code to Describe Table Find an Entry

A class for the student data is declared, inheriting from 'ViewModel'.

class StudentView(ViewModel):

Attributes or 'fields' are declared at the class level, with 'ViewFields' assigned for each (the id, name, course, year number). e.g.

id = IdField()
name = TxtField()
course = IntField()
#  .... field definitions may  continue

The view can be given an optional 'viewName_' to be displayed to system users. The view name will default to the class name if no view name is supplied. Note that all attributes of the class that are not 'fields' are given a trailing underscore to reduce the changes of them colliding with a field name.

 viewName_ = "Students"

Full working code follows:

from ViewModel import ViewModel, IdField, TxtField, IntField
import pymongo

database = pymongo.MongoClient(dbserver).get_database("example")

class StudentView(ViewModel):
    viewName_ = "Students"
    models_ = database.Students

    id = IdField()
    name = TxtField()
    course = IntField()
    #  .... field definitions may  continue

student = StudentView({}, models=database.Students)
# could have used 'models_' within class to avoid needing 'models' parameter
# for the init
#   {} empty dictionary to ensure an empty view, not needed if the database
#   does not even exist yet, as with a new database, initial view will always
#   be an empty view

if len(student) > 0:
    print("oh no, we already have data somehow!")

students.insert_() # add an empty entry to our view

with student:  # use 'with', so changes written at the end of 'with'
    student.name = 'Fred'

# ok ... now we have a 'Student' table with one entry

Code to Read and Update Our Entry

A key concept is that while the class for the view describes a table, set of tables or joined tables (or collections in Mongo speak), an instance of a ViewModel is the set of data or a window of the tables. Instancing the view reads from the database in most straightforward cases, although in more complicated cases the data may be read from the database when accessed, the view instance logically includes all data from a 'read' operation:

# same class definition and imports as above

student = StudentView({'name': 'Fred'},model = database.Students)
# would save if we could have 'models_' in class definition!

if not student.course:
    with student:
        student.course_year = 2
        student.course = 'Computing'

Multiple Entry Views

So far our view has only one entry. An instance of our view is a window viewing part of the database. This window can be a single row/collection or a logical group of entries(from rows/collections), and for small tables, may even be the entire table/collection. The code that follows adds another entry, so the sample has more than one entry, then works with a multi-entry view:

StudentView.models_ = database.Students
# modify class, add 'models_' as an attribute,
# this saves specifying 'models_' each time instancing StudentView

student = StudentView()
# no dictionary, this gives an empty view (not multi entry yet)

student.insert_()
with student:  # adding a second student
    student.name = 'Jane'
    student.course = "Computing"
    student.course_year = 2

# now our multi entry view for all year 2 Students
students = StudentView({'course_year':2})

for student in students:
    print(student.name)

Note how multi-entry view instances can be treated as lists. In fact, single entry views can also be treated as a list, however for convenience view properties for single entry views also allow direct access as one entry. For a single entry view 'student':

student.name == student[0].name

Example Summary

The example bypasses the power of ViewModels to show you a simple introduction. A fundamental concept is that classes describe a table (or collection or set/join of tables). An instance of a ViewModel is one set specific subset, a set of data from a table (or set/join of multiple tables).

Describing a Table/Collection With ViewFields

When creating a class derived from a ViewModel, add class attributes which are 'ViewFields' for each field in the table or collection.

The example (Simple Example. ) uses several types of view fields. However each 'ViewField' can contain information well beyond the type of data. An alternative name, a short and long description, formatting and other display defaults, value constraints and many other settings, as well as a 'default value' set with the 'value=' init parameter. Note that when a new row is inserted into a view, no fields are set to their default value, and instead all fields, even those with default values, remain 'unset'. However 'unset' fields return their default value when accessed. This means that if a ViewModel can have a new field (or even merely a new default value for an existing field) added after several rows are already in the database. Existing records will behave automatically return the 'default value' even though they were saved prior to the default being defined. This makes ViewModels stable and safe for software updates which add new fields without the need to update the database itself.

In the example, only the 'value' attribute of the "name" ViewField is accessed. 'student.name' does not access the ViewField, but instead returns "value" attribute of the "name" ViewField. To access the actual ViewField (or IntField, TextField etc) and have access to these other attributes use 'student["name"]' thus:

student.name == student["name"].value

Using 'ViewField' Derived Classes

All 'fields' are sub-classed from ViewField and represent individual data types. Each field contains the following properties:

• name: set explicitly, or defaulting to the property name;
• label: set explicitly but defaulting to the name;
• hint: defaults to '' for display;
• value: returns value when a field is an attribute of a row object.

'ViewModel' Interface

The 'ViewModel' provides a base class defines a database table/collection, and each instance of a ViewModel. Note all system properties and methods start of end with underscore to avoid name collision with database field names.

ViewModel Interface Methods

• insert_()
• labelsList_()
• update_()
• <iterate> for row in <ViewModel instance>
• <index> <ViewModel instance>[row]

ViewModel Interface Properties

• viewName_
• models_
• dbModels_

ViewModel Details

The insert_() method adds an empty new row (ViewRow instance) to the current ViewModel instance. At the next update_(), an actual database document/row will be created, provided some values have been set in the new row.

Note that a record is currently marked for insert if there is no '_id', and otherwise for update. So if a record created by insert_() has an '_id' added, currently this record will then allow changes by update, without reading the record first.

The labelsList_() method returns a list of the labels from the rows of the current ViewModel instance. It computes the list of labels by, first, looking for the row_label attribute if that fails then it will search through all possible fields for anything called rowLabel and then set row_label to the corresponding value of rowLabel. If rowLabel is not declared as True in the view definition, the rowLabel will default to 'no labels'.

The update_() method is called automatically at end of a 'with ' statement (python keyword 'with'), or can be called directly, to update the actual database with values changed by assignments through '. = statements'.

viewName_ is merely a title for the view for display purposes.

models_ is a list of the names of tables, or actual database tables objects used by the view

dbModels_ is a dictionary of database table objects used by the view, with the model names as keys.

Note: all 'ViewModel' instances with one row implements all of the ViewRow interfaces in addition to the methods and properties discussed. 'ViewModel' instances with more than one row will raise errors if the 'ViewRow' interface as it is ambiguous which row/document to use.

'ViewRow': The Row Interface

ViewRow objects and ViewModel objects both implement the 'ViewRow' interface.

Where a ViewModel contains one logical row, the operations can be performed on the ViewModel, which also supports this interface for single row instances.

ViewRow Interface Methods

• <iterate>: for field in <ViewRow instance>
• loop_(case=<case>): for field in a <ViewRow instance>
• <index>: <ViewRow instance>[<fieldname>]
• <attribute> <ViewRow instance>.field_name

ViewRow Interface Properties

• fields_
• view_
• label_
• idx_

ViewRow Details

The statement: for <field> in <ViewRow instance>: provides for using a 'for loop' to iterate over the fields in a row of a viewfield.

Note that this iteration can be for building a view, and as such the iteration allows for selecting which fields are included in the view. When fields are declared (see 'ViewField' Interface), they can set a 'case' where they are applicable for views. For example, this can be in a view, on an edit panel, or the field is for calculation purposes and part of the model, but not revealed in a view.

Using <ViewRow instance>[<field name>] (or indexing), retrieves the instance of the ViewField named. For example:

student['name'].value = 'Jane'
print(student['name'].value)

# is equivalent to
student.name = 'Jane'
print(student.name)
# but the point of using indexing to access other field attributes
assert student['name'].wide == 16 # check the name field is 16 characters wide

fields_ returns a 'ViewRow' is a logical entry in a ViewModel. Consider the example ( Simple Example. ). The line of code:

student.name = 'Fred'

Is using the ViewRow set attribute interface to set the 'value' of the 'name' field within the 'row' created by the insert_() method.

In this example, because the 'student' ViewModel has only one row, the 'name' field can be accessed directly in the ViewModel. However, if there were, for example, three students in the view, which 'name' is to be changed? As stated previously, ViewModel objects support the ViewRow interface but report an error if there is more then one row.

There are two main ways to access 'ViewRow' objects (apart from simple treating the ViewModel as also a ViewRow, which only works for single row views). If our 'student' ViewModel contains three students, there will be a row for each student, and these 'rows' could be accessed as:

students = StudentView({})
assert len(students) == 3  # check we have 3 students
student_0 = students[0]
student_2 = students[2]
for student in students:
    <print details from student>

From the ViewModel, indexing or iterating can access the ViewRows.

This interface allows retrieving and setting data 'fields' or ViewField entries by name as object attributes. All internal attributes of ViewRow have either a trailing underscore to avoid name collisions with field names of the database, or a leading underscore to indicate that these attributes should not be accessed externally of the ViewRow or ViewModel.

Provided database fields have no leading or trailing underscore, they will not collide with the names of internal workings of these classes.

Extended ViewModel Declarations and Instancing

getRows

The __init__() method calls getRows_ which is designed for subclassing. getRows_ can return either:

1. An empty list (for an empty view);
2. The raw data from a find (where all data is from a single source and in this case the 'source' parameter to the class is used to build dbRows_ automatically;
3. A list of dicts (for the rows, dict with one entry for each 'source', and that entry itself being a dictionary of the fields of that 'source'.

Previous versions of the library required (2) to be instead a list of ObjDicts. This is no longer supported. The statement:

# below statement no longer will produce functioning code
# remove it
result = [ObjDict(res) for res in result]

... would convert the result of a find into a list of ObjDicts, where each ObjDict is a row. What is now required is such data is embedded in a 'source' dictionary. A replacement for the above line, (which is not need as the standard class init method will make this adjustment automatically), would be the line:

result = [Objdict(((row,res),)) for res in self._dbRows]

models_ and _sources

As the names suggest, 'models' is for 'public' use (or in this case declaration) and _sources is 'private'. The data to construct _sources is provided in but the _sources class variable, or the 'sources' parameter to a viewmodel constructor.

If sources (either _sources class variable or sources parameter), is not a list then internal logic treats it as a one element list: [sources], so even if only one value is provided, consider that value a one element list.

Each value in the 'models' list can be one of the legacy values of 'None' or a MongoDB collection, or (preferred) an object instanced using a class based on the DBSource class. Currently, four such classes exist: DBNoSource; DBVMSource; DBMongoSource and DBMongoEmbedSource.

DBNoSource

When generating a sources list from 'models', a value of None is used as a legacy alternative to creating a DBNoSource object, but the preferred way is an explicit object. Fields with a 'NoSource', as the class name suggests, have no database source and thus no storage and as such are temporary values only. Since a collection or table name is not part of a 'NoSource' object, the source name must be described explicitly or will be 'None'. Note that at the time of writing, any string entry in a source list that beginning with an underscore will be taken as a DBNoSource object with the name of that string.

DBVMSource

A DBVMSource is used for data that exists within another ViewModel. This allows nested views. This time, this is merely a provision for the future.

DBMongoSource

The source used for mongo collections, and instanced from legacy MongoDB collections, as well as from the preferred explicit instances. The 'name' of a DBMongoSource is the name of the collection. So the collection 'students' would have the string name 'students'.

DBMongoEmbedSource

These are used when the table is embedded within a document inside a mongo collection. The source is specified as ".<object-list_name>", where the object list name is the object containing the entire embedded collection as a list of objects.

Declaring 'models_'

Models (models_) may be declared as a class variable, or passed as a parameter ('models') to the __init__() method for the ViewModel.

In either case, the value is a list of each source, with each entry of one of the 'DBSource' types listed above, or an application specific class derived from DBSource. Note that while models are in theory a list, the code will convert a single entry into a list, eliminating the need to have a single entry as a list.

Setting Field Source

Any field can belong to any 'source', as described above. The first 'source' for a view is considered the default source, so if using the first source, or 'default source', it is possible to omit the 'src=' parameter. Any field which is from a view other than the first view needs to specify the view by name with the 'src' parameter:

src=<name of the source as a string>

For an embedded source, the name will use 'dot notation'.

Further, a field may be embedded in another object. The name of the object should also be a specified through source. Examples:

models_ = DBMongoSource('students'), DBMongoSource('courses')

num1 = IntField()  # no 'src' specified -- field is in default 'students' collection
num2 = IntField(src='courses')  # field is in 'courses' table/collection
num3 = IntField(src='courses.scores') # field is in scores object in courses table
num4 = IntField(src='students.scores') # field is in scores object in students table
num5 = IntField(src='.scores') # alternative using default notation, same location as 'num4'

'ViewField' Interface

Getting and Setting 'Row Member' Values

To be added

Building HTML Forms

To be added

Updating from HTML Forms

To be added

How to Load Test DB Data From JSON Files for Testing

Loading tables (collections) for testing is made easier by using the JSONLoad class provided in ViewModel. The class allows you to load previously downloaded JSON tables (Mongo collections -- just make sure they are created as JSON array types -- see How to Export Mongo Databases/Collections to JSON for more about this). The JSONLoad class is in "json_load.py".

The JSONLoad class sets the following defaults:

• The default JSONLoad location is "dumped_data". It is located at the same level as the test file (test_file.py) that is using the JSONLoad class (see below):

    project_root/
    |-- ...
    |-- tests/
        |-- dumped_data/
        |-- test_file.py/
    |-- ...
  

  To override the default location, import "DEFAULT_DUMP_DATA_FOLDER_NAME" and set it to what you want it to be.

• The default host name & port number is:

  host_name = localhost port = 271017

• The default DB name is '' by design and is a required parameter i.e. db_name defaults to '' so must be passed in when you use JSONLoad:

  JSONLoad(db_name="MY_TEST_DB_NAME")

To load JSON data into a test DB of your choice, follow the instruction below. The best place is in your "conftest.py" file if you are using pytest.

To import and use JSONLoad and optionally, DEFAULT_DUMP_DATA_FOLDER_NAME, include the following import statement in your test script:

from viewmodel.json_load import JSONLoad, DEFAULT_DUMP_DATA_FOLDER_NAME

Optionally, override the DEFAULT_DUMP_DATA_FOLDER_NAME with another in your script:

DEFAULT_DUMP_DATA_FOLDER_NAME = 'my_alternate_folder_name'

Provide a test DB name (here in a separate variable called TEST_DB) and create a test fixture that uses JSONLoad to call the method restore_db_from_json:

TEST_DB = 'my_test_db_name'

@pytest.fixture(scope='session', autouse=True)
def restore_db_from_json():
    JSONLoad(db_name=TEST_DB).restore_db_from_json()

Then be sure to connect to your test DB:

res = ObjDict(dbname=TEST_DB, dbserver=None)
viewModelDB.baseDB.connect(res)

JSONLoad Method Signatures

# not sure what the idea here is...see .rst file history? these signature need a way of being written, then cleaned up
__init__(host_name: str = 'localhost', port_number: int = 27017, db_name: str = None)
insert_one(collection_name: str = None, data: dict = None)
insert_many(collection_name: str = None, data: List = None)
drop_db(db_name: str)
drop_collection(collection_name: str)
read_json_data_file(path_to_file: str, file_name: str)
load_data(collection_name: str, path_to_file: str, file_name: str)
get_default_dumped_data_path()
load_all(json_data_path: str = None)
restore_db_from_json()

Data Relationships and Joins

The term 'relational database' comes from the concept that data contained in separate tables (or collections) is related.

Data Relationship Types

Many-to-One

These are classic 'dry' (Dont Repeat Yourself) solutions. Several records (or rows or documents) in a table/collection will use the same information. For example, consider the Students are each signed up for a degree. Consider a model where are several Students for each degree, but each Student is in only one degree program at one time, even if that degree program is itself a double degree. For each degree, there could be further information such as the degree description, number of years within the degree, head teacher for the degree and an information URL. Each Student document could contain all of this information about their degree program, but many students documents would repeat the same information. The solution is having a separate document for each degree and linking the students to their degree.

See the tests tutorialtest_tutorial::class GenerateCourseData

on (name, city code, state) from a separate city table will mean that information for each city is not repeated for each address with the same city. From the perspective of the address, the relationship is 'one-to-one' because for each address there is only one city. The 'many-to-one' is that many addresses may reference each city.

If our view is based on a single address, then retrieving the 'join' of the information for the address together with the information for the city still leaves a single 'row' in the resulting view.

In database design, to implement a 'many-to-one', each entry from the many tables, contains a key to the city table. Read an address, the use the 'key to the city' to read data from the city table.

One-to-Many

From a technical perspective, this is simply the same as 'many-to-one', but viewed from the opposite perspective. However, the devil is in the detail, and having the opposite perspective has implications that can mean the correct implementation is very different. Looking at the previous cities and addresses, the 'one-to-many' view from the city perspective is to consider all addresses with the city.

If our view is based on a single city, then retrieving the 'join' would result in rows for each address. So while the one-to-many is the many-to-one from the opposite perspective, the view changes entirely and in nature depending on which perspective.

In database design, the cross-reference key is still the 'key to the city' within the address table. Read the city key (as 'our city key'). Then using the key field find all addresses with their 'key to the city' value matching the key in 'our city key'.

One of Many Selector

This is a real-world application of the 'many-to-one' join, where the table of possible 'ones' effective represents one of a finite set of choices which may be chosen from a 'drop-down list box'. ViewModel has a specific Field Type, the 'EnumForeignField'. Note that to display choices for editing the entire table of choices is required. There are no strict formulae as to when the number of choices or total data of the choices table is too large but generally the system must have the capacity to consider having the entire table in memory acceptable.

Many-to-Many

Consider now database with not just addresses and cities, but also people. Each person might have a relationship to several addresses. However, rather than this being a 'one-to-many' relationships, like the Cities -> Addresses, where viewed from the other perspective, Addresses -> Cities, for each address, there would be only one city, this time for each address there may be multiple people.

In database design, this usually represents more of a challenge. If we start with people, we cannot look for addresses with a 'person key' field that matches since our person, since each address will need to match potentially several (or many) people. The matching person cannot be stored as a single value in our table. With SQL and even sometimes with NoSQL, the solution is to have a separate table of relationships. If we read this table for all entries matching our person we can find an entry for each relationship to an address for that person. This solves the problem because we can have more relationships than we have either people or addresses, so one entry per table will not work without a particular table that can have an entry for each relationship.

NoSQL like Mongo provides another alternative, which is keeping a list of relationships inside one (or even both) of the tables. Since an entry in the table can be a list, we could keep a list of addresses in the people table. Read a person, and we have a list of addresses. Read an address, and we can read all people with our address in their address list. The principle is still the same, but there is this implementation choice.

Relationship Specific Data

In some cases, there can be data specific to a relationship. Consider the following people, addresses and then relationships:

People:  Bob, Tom, Alice
Addresses: RedHouse, Office1, Office2, GreenHouse
Relationships:
    Bob: RedHouse is 'home', Office1 is 'work'
    Alice: RedHouse is 'home' and 'office'
    Tom: GreenHouse is Home, RedHouse is 'work1' and Office2 'work2'

The relationships between the people can each have their labels, just as the relationships between people can. In fact, each relationship can have a label from each perspective. Consider people relationships where Bob could be 'husband' to Alice, but the same relationship from the other perspective could be 'wife'.

So for Bob, we may have to have not only added 'RedHouse' and created a relationship, we also have to manage a label for the relationship.

Joins

In SQL, a join is a read, or update, of data from more than one table. The join uses the relationship between tables to select rows of data that combine information from multiple tables. Each table in the join is effectively a source of data.

ViewModel support data from multiple sources, but currently this has only been used to support joins from relationship tables and tables that are part of the relationship.

Inserts With Joins

When a new document is inserted for any source within a ViewModel, fields within the current view can be automatically updated to reference the new _id generated. These fields should be listed in the _sources[<source updated>].join_links list. This list is the field names to be updated.

How It Works

The Rows Structure

The actual data is kept in a view list called dbRows_, which reflects the actual data being held in the underlying database. For each row of the view, there is one entry in dbRows_.

The List of Elements of 'dbRows_'

Each entry is of type 'objdict' and the elements of the objdict were originally the values of the fields in the view, but a new layer has been added, so that 'objdict' entries at the top level represent the data from a single source.

From:

[ {'name':'Jane','course':'computing'}]

To:

[ {'students': {'name':'Jane','course':'computing'}}]

The two-tiered structure, keyed by the 'table/collection' which is the data source, better provides for data from multiple sources.

Data is not added directly to these rows but through the 'viewmodel_row' wrappers. So if a ViewModel row has a view_field (say 'last_name') which is not present in the row, setting the name would add a new field to the appropriate ObjDict within the row, but also an an entry to an additional 'changes' copy of the row, which holds new values not yet committed to the database.

The 'rows' and 'changes' are the bridges between what is in the database files, and what is held in memory.

The DBSource Descriptor

See the DBSource class documentation, but this class describes the sources of data that are held within the dbRows.

Each 'row' has a set of a least one 'source'. Source types can be MongoDB table, MongoDB document, memory, (and soon) another view.

Each source requires a method to load from the source, and update to the source. 'getrows' methods currently takes a 'load filter' and uses that to load all sources, but a structure is required to more flexible to handle all sources.

Update methods again handle all source types.

It is suggested that a useful revision would be to have 'getrows' that calls a src_getrows for each source and update call src_update() for each source.

New getRows

A new getrows would take a filter dictionary or list as valid parameters. Each entry would need a lead and a lazy. Run 'leads' in sequence until lead returns a non zero list. List is applied for that source, all other sources are empty, but have 'lazy' load available.

Once a lead returns true, the scr_getrows_table() would apply a dictionary;