Sming library providing strongly typed JSON configuration database support.
Applications requiring complex non-volatile configuration data typically store this in JSON files. Although libraries such as ArduinoJson provide flexible and easy access to these, there are a few drawbacks:
- Once configuration exceeds a certain size it becomes unmanageable as a single JSON file
- Configuration data is typically managed on an ad-hoc basis using structures, and code is required to read/write these structures
- Code must deal with various common conditions such as default values, range checking, formatting, etc.
- Clients such as web applications or other external processors require further code to interpret configuration data
Design goals:
- Describe data once using a standard schema
- Generate API code from the schema which ensures consistent access, applies default values, checks value ranges, etc.
- Make use of available compile-time checks where possible (definitely for value types)
- Allow use of existing tools to enable easy access to data in other languages and on other platforms
- Allow partitioning of database into multiple stores to control use of RAM and enable more efficient updates for related data groups
- Reduce RAM usage as database size increases
- Maintain good read/write performance
- Pluggable system to support backing stores other than ArduinoJson. For example, binary formats
- Keep application interface independent of implementation detail
Usage:
- Database is described in a JSON schema.
- A python script (tools/dbgen.py) parses the schema and generates class structure for application use.
- Content of database can be streamed to/from web clients
The database structure is defined using a standard JSON schema. A good introduction is to take the Tour.
An initial schema can be created from an existing sample JSON configuration file using a generation tool such as https://github.com/saasquatch/json-schema-inferrer. Go to the online demo and paste in your JSON configuration. The resulting schema can then be edited and further customised.
Standardised web editors can also be generated using tools such as https://github.com/json-editor/json-editor. Go to the online demo and scroll down to Schema.
Note
Sming uses JSON schema for:
- Hardware configuration Sming/Components/Storage/schema.json
- IFS build scripts Sming/Components/IFS/tools/fsbuild/schema.json
- USB config Sming/Libraries/USB/schema.json
It's probably fair to consider it a standard part of the framework.
Configuration JSON can be validated against the .cfgdb schema files using check-jsonschema:
pip install check-jsonschema check-jsonschema --schemafile basic-config.cfgdb sample-config.json
Separate documentation can be generated from JSON Schema using various tools such as JSON Schema for Humans. For example:
python -m pip install json-schema-for-humans generate-schema-doc basic-config.cfgdb basic-config.md --config template_name=md
See the :sample:`Basic_Config` sample schema. The test application contains further examples.
- Root object is always a :cpp:class:`ConfigDB::Database`
- A database is always rooted in a directory
- An optional include array annotation can be added to specify additional header files required for custom types used in the schema.
- Contains one or more stores. The root (un-named) object is the primary store, with the filename _root.json.
- Immediate children of the root may have a store value attached to place them into a new store. This can have any value, typically true.
- A custom type can be defined for Property accessors using the ctype annotation. This means that with an IP address property, for example, you can use :cpp:class:`IpAddress` instead of :cpp:class:`String` because it can be constructed from a String. The database still stores the value internally as a regular String.
- Enumerated Properties can be defined for any type; ctype is used here to optionally define a custom enum class type for these values.
- Re-useable definitions can be created using the $ref schema keyword. Such definitions should be contained within the $defs section of the schema.
- Arrays of simple types (including Strings) or objects are supported.
- Unions can contain any one of a number of user-defined object types, and can be used in object arrays.
- Null values are not supported. If encountered in existing datasets then ConfigDB uses the default value for the property.
- Multiple property types, such as "type": ["boolean", "null"] are not supported. A type must be defined for all properties and must be a string value. This also applies to array items.
Properties may have alternative names to support reading legacy datasets. For example:
{
"$schema": "http://json-schema.org/draft-07/schema#",
"type": "object",
"properties": {
"trans_fin_interval": {
"type":"object",
"alias": "transfin_interval",
"properties":{
"type":"integer"
}
}
}
}
Existing JSON data using the old transfin_interval name will be accepted during loading. When changes are made the new (canonical) name of trans_fin_interval will be used.
If multiple aliases are required for a property, provide them as a list.
Items with number type are considered floating-point values. They are not stored internally as float or double but instead use a base-10 representation.
This provides more flexibility in how these values are used and allows applications to work with very large or small numbers without requiring any floating-point arithmetic.
See :cpp:class:`ConfigDB::number_t` and :cpp:class:`ConfigDB::Number` for details. There is also :cpp:class:`ConfigDB::const_number_t` to ease support for format conversion at compile time.
JsonSchema offers the enum keyword to restrict values to a set of known values. For example:
{
"$schema": "http://json-schema.org/draft-07/schema#",
"type": "object",
"properties": {
"color": {
"type": "string",
"enum": [
"red",
"green",
"blue"
]
}
}
}
ConfigDB treats these as an indexed map, so red has the index 0, green is 1 and blue 2. Indices are of type uint8_t. The example has an intrinsic minimum of 0 and maximum of 2. As with other numeric properties, attempting to set values outside this range are clipped.
The default is 0 (the first string in the list). If a default value is given in the schema, it must match an item in the enum array.
The corresponding setColor, getColor methods set or retrieve the value as a number. Adding "ctype": "Color" to the property will generate an enum class definition instead. This is the preferred approach.
The color value itself will be stored as a string with one of the given values. The integer and number types are also supported, which can be useful for generating constant lookup tables.
ConfigDB uses the array schema keyword to implement both simple arrays (containing integers, numbers or Strings) and object arrays.
Simple arrays are accessed via the :cpp:class:`ConfigDB::Array` class. All elements must be of the same type. A default value may be specified which is applied automatically for uninitialised stores. The :cpp:func:`ConfigDB::Object::loadArrayDefaults` method may also be used during updates to load these default definitions.
The :cpp:class:`ConfigDB::ObjectArray` type can be used for arrays of objects or unions. Default values are not currently supported for these.
These are defined using the oneOf schema keyword, which defines an array of option definitions.
Each definition can be defined using $ref. The name of the option will be taken from that definition, and can be overridden by adding a title keyword. Option definitions can also be given directly, in which case title is required.
The test application contains an example of this in the test-config-union.cfgdb schema. It is used in the Updates test module.
Like a regular C++ union, a :cpp:class:`ConfigDB::Union` object has one or more object types overlaid in the same storage space. The size of the object is therefore governed by the size of the largest type stored. A uint8_t property tag indicates which type is stored.
The code generator produces an asXXX method for each type of object which can be stored. The application is responsible for checking which type is present via :cpp:func:`ConfigDB::Union::getTag`; if the wrong method is called, a runtime assertion will be generated.
The corresponding Union Updater class has a :cpp:func:`ConfigDB::Union::setTag` method. This changes the stored object type and initialises it to default values. This is done even if the tag value doesn't change so can be used to 'reset' an object to defaults. The code generator produces a toXXX method which sets the tag and returns the appropriate object type.
JSON Schema describes ways to structure complex schemas.
Re-useable (shared) definitions are, by convention, placed under $defs. These are referenced using the $ref keyword with JSON pointer syntax.
For example:
"$ref": "#/$defs/MyObject"
Definitions from other schema may be used:
"$ref": "other-schema/$defs/MyObject"
The dbgen.py code generator is passed the names of all schema found in the current Sming project, which are loaded and parsed as a set using the base name of the .cfgdb schema (without file extension) as its identifier.
Note
The full URI resolution described by JSON Schema is not currently implemented. This would require $id annotations in all schema.
When using shared objects only the name of the related property can be changed. For example:
"font-color": {
"foreground": {
"$ref": "#/$defs/Color"
},
"background": {
"$ref": "#/$defs/Color"
}
}
This generates a C++ property fontColor using a FontColor object definition which itself contains two properties: foreground and background. The object definition for both is Color.
References to simple (non-object) property types are handled differently. A type is not defined but instead used as a base definition which can be modified. For example, we can provide a general Pin definition:
"$defs": {
"Pin": {
"type": "integer",
"minimum": 0,
"maximum": 63
}
}
And use it like this:
"properties": {
"input-pin": {
"$ref": "#/$defs/Pin",
"default": 13
},
"output-pin": {
"$ref": "#/$defs/Pin",
"default": 4
}
}
This is identical to the following:
"properties": {
"input-pin": {
"type": "integer",
"minimum": 0,
"maximum": 63,
"default": 13
}
"output-pin": {
"type": "integer",
"minimum": 0,
"maximum": 63,
"default": 4
}
}
This approach can make the schema more readable, reduce duplication and simplify maintainance.
This example generates a uint8_t property value. A different type may be specified for property accessors using the ctype annotation.
By default, stores are saved as JSON files to the local filesystem.
The code generator creates a default :cpp:class:`ConfigDB::Database` class. This can be overridden to customise loading/saving behaviour.
The :cpp:func:`ConfigDB::Database::getFormat` method is called to get the storage format for a given Store. A :cpp:class:`ConfigDB::Format` implementation provides various methods for serializing and de-serializing database and object content.
Currently only json is implemented - see :cpp:class:`ConfigDB::Json::format`. Each store is contained in a separate file. The name of the store forms the JSONPath prefix for any contained objects and values.
The :sample:`BasicConfig` sample demonstrates using the stream classes to read and write data from a web client.
Important
Any invalid data in a JSON update file will produce a debug warning, but will not cause processing to stop. This behaviour can be changed by implementing a custom :cpp:func:`ConfigDB::Database::handleFormatError` method.
The default streaming update (writing) behaviour is to overwrite only those values received. This allows selective updating of properties. For example:
{
"security": {
"api_secured": "false"
}
}
This updates the api_secured value in the database, leaving everything else unchanged.
Arrays are handled slightly differently. To overwrites the array with new values:
"x": [1, 2, 3, 4]
To clear the array:
"x": []
Indexed array operations
Array selectors can be used which operate in the same way as python list operations. So x[i] corresponds to a single element at index i, x[i:j] is a 'slice' starting at index i and ending with index (j-1). Negative numbers refer to offsets from the end of the array, so -1 is the last element.
When selecting a single array element x[5], the provided index must exist in the array or import will fail. When updating a range, index values equal to or greater than the array length will be treated as an append operation.
The following example operations demonstrate what happens with an initial JSON array x:
{
"x": [1, 2, 3, 4]
}
The result value shows the value for x after the update operation. The same operations are supported for arrays of other types, including objects.
Update single item:
{
"x[0]" : 8,
"result": [8, 2, 3, 4]
},
{
"x[2]" : 8,
"result": [1, 2, 8, 4]
},
{
"x[-1]" : 8,
"result": [1, 2, 3, 8]
}
Update multiple items
Note that the assigned value must be an array or the import will fail:
{
"x[0:2]" : [8, 9],
"result": [8, 9, 3, 4]
},
{
"x[1:1]": [8, 9],
"result": [1, 8, 9, 2, 3, 4]
},
"x[1:2]": [8, 9],
"result": [1, 8, 9, 3, 4]
},
{
"x[2:]": [8, 9],
"result": [1, 2, 8, 9]
}
Insert item:
{
"x[3:0]" : [8],
"result": [1, 2, 3, 8, 4]
},
{
"x[3:3]": [8],
"result": [1, 2, 3, 8, 4]
},
{
"x[-1:]" : [8, 9],
"result": [1, 2, 3, 8, 9]
}
Append item:
{
"x[]": [8, 9],
"result": [1, 2, 3, 4, 8, 9]
},
{
"x[]": 8,
"result": [1, 2, 3, 4, 8]
}
Append multiple items:
{
"x[]": [8, 9],
"result": [1, 2, 3, 4, 8, 9]
},
{
"x[10:]": [8, 9],
"result": [1, 2, 3, 4, 8, 9]
}
Object array selection
The x[name=value] syntax can be used to select one object from an array of objects. Here's the test data:
{
"x": [
{
"name": "object 1",
"value": 1
},
{
"name": "object 2",
"value": 2
}
]
}
And the selector can be used like this:
{
"x[name=object 1]": { "value": 8 },
"result": [
{
"name": "object 1",
"value": 8
},
{
"name": "object 2",
"value": 2
}
]
}
or:
{
"x[value=2]": { "value": 8 },
"x[value=1]": { "value": 1234 },
"result": [
{
"name": "object 1",
"value": 1234
},
{
"name": "object 2",
"value": 8
}
]
}
Limitations:
- Only the first matching object will be selected
- Only one object key can be matched
You can find more examples in the test application under resource/array-test.json.
Each .cfgdb file found in the project directory is compiled into a corresponding .h and .cpp file in out/ConfigDB. This directory is added to the #include path.
For example:
- basic-config.cfgdb is compiled into basic-config.h and basic-config.cpp
- The applications will #include <basic-config.h>
- This file contains defines the BasicConfig class which contains all accessible objects and array items
- Each object defined in the schema, such as network, gets a corresponding contained class such as ContainedNetwork, and an outer class such as Network.
- Both of these classes provide read-only access to the data via getXXX methods.
- Outer classes contain a :cpp:class:`ConfigDB::StoreRef`, whereas contained classes do not (they obtain the store from their parent object).
- Application code can instantiate the outer class directly BasicConfig::Network network(database);
- Child objects within classes are defined as read-only member variables, such as network.mqtt, which is a ContainedMqtt class instance.
- A third updater class type is also generated which adds setXXX and resetXXX methods for changing values. Child objects/arrays can be updated using their provided methods.
- Only one updater per store can be open at a time. This ensures consistent data updates.
Code can update database entries in several ways.
Using updater created on read-only class:
BasicConfig::Root::Security sec(database); if(auto update = sec.update()) { update.setApiSecured(true); }The update value is a BasicConfig::Root::Security::Updater instance.
Any changes made via the update are immediately reflected in sec as they share the same Store instance. The update() method can be called multiple times when used in this way. Changes are committed automatically when the last updater loses scope.
Directly instantiate updater class:
BasicConfig::Root::Security::Updater update(database); if(update) { update.setApiSecured(true); }Only one updater instance is permitted.
Asynchronous update:
BasicConfig::Root::Security sec(database); bool completed = sec.update([](auto update) { update.setApiSecured(true); });If there are no other updates in progress then the update happens immediately and completed is true. Otherwise the update is queued and false is returned. The update will be executed when the store is released.
During an update, applications can optionally call :cpp:func:`Updater::commit` to save changes at any time. Changes are only saved if the Store dirty flag is set. Calling :cpp:func:`Updater::clearDirty` will prevent auto-commit, provided further changes are not made.
.. doxygenclass:: ConfigDB::Database :members:
.. doxygenclass:: ConfigDB::Store :members:
.. doxygenclass:: ConfigDB::Object :members:
.. doxygenclass:: ConfigDB::Union :members:
.. doxygenclass:: ConfigDB::Array :members:
.. doxygenclass:: ConfigDB::ObjectArray :members:
.. doxygenclass:: ConfigDB::Format :members:
.. doxygenvariable:: ConfigDB::Json::format
.. doxygenclass:: ConfigDB::Number :members:
.. doxygenstruct:: ConfigDB::number_t :members:
.. doxygenstruct:: ConfigDB::const_number_t