Forest 2.0 is a refactoring of the forest system using components that have been put into their own libraries: transact, walrus and collect. It is documented Here.
Forest is an observable state composed of nested Leaf instances. Leaves have values
which are any referencable object, managed by @wonderlandlabs/collect wrappers.
Each Forest has a Root Leaf instance. that instance and the leaf, expose a value property
that is actually a reference to the leaf's store - a Collect instance. You can change the
Forest('s leaf's) value.
Change leaf state is immediate;
const item = new Forest({$value: {x: 1, y: 3}});
item.subscribe((v) => console.log('--- value is ', v));
// -- value is {x: 1, y: 3}
item.value = {x: 2, y: 4};
// -- value is {x: 2, y: 4}
console.log('item.value', item.value);
// item.value = {x: 2, y: 4};
console.log('item.root.,value', item.root.value);
// item.root.value = {x: 2, y: 4};Each Leaf (and the Forest instance) has two properties identifying the nature of its contents;
it uses the @wonderlandlabs/walrus system:
| type | form | family |
|---|---|---|
| undefined | void | void |
| null | void | void |
| string | scalar | scalar |
| number | scalar | scalar |
| boolean | scalar | scalar |
| symbol | scalar | scalar |
| array | array | container |
| map | container | container |
| object | container | container |
| set | container | container |
| function | function | function |
typeis not equivalent to typeof;objectis only equivalent to true object -- it does not include the more specific container type (Map, Array, etc.) nor does it include null.formmerges all "empty" type into a catchall "void" and groups complex grouping type (Map, Object, Set) into a general "container" category.familyonly has four buckets; all the multiple-value type are grouped under "container"
If you do not define test, leaves can contain any value and the value can be changed from one type to another without any constraint. This can cause difficulty when for instance you attach child leaves (more later) to a base leaf and the leaf type is changed form a container form to a scalar one.
in your configuration you can (optional) limit:
- which type(s) you want to allow the Leaf instance to allow (a string, or an array of strings)
- one or more test you want the Leaf instance to allow (a function or an array of functions)
- you can for brevity set
test: truewhich will snapshot the type of the initial value (after filtering, see below) and requires all following values to equal that type.
There are always those edge cases where validation doesn't "fit" within the type constraints. In that case, it is best to write your own validator. Your validator will be passed the candidate value and the leaf:
const isInteger = (value) => {
if (!Number.isInteger(value)) throw new Error('number must be an integer');
}
const isWhole = (value) => {
if (value < 0) {
throw new Error('numbers must be positive or zero')
}
};
const isPrime = (value) => {
if (value <= 1) return;
for (let divisor = 2; divisor < value; ++divisor) {
if (!(value / divisor)) {
throw new Error(`${value} is not a prime number. (divisible by ${divisor})`);
}
}
}
const primeNumber = new Forest( {$value: 0, type: 'number', test: [isInteger, isWhole, isPrime]});
primeNumber.value = 7;
try {
primeNumber.value = 8;
} catch (err) {
console.log(err.mesage)
}
// 8 is not a prime number. (divisible by 2)
try {
primeNumber.value = -8;
} catch (err) {
console.log(err.mesage)
}
// numbers must be positive or zero
try {
primeNumber.value = 1.5;
} catch (err) {
console.log(err.mesage)
}
// number must be an integer
try {
primeNumber.value = 'a bad value'
} catch (err) {
console.log(err.mesage)
}
// cannot add value of type string to leaf root (type number)
primeNumber.value = 11;Type constraints, if present, are processed before custom validators. If one custom validators fails, the other validators are not used; likewise only values that pass the type criteria are tested with custom validators.
The second value passed to custom validators is the leaf itself.
Any validator in a list of test that fails will preempt subsequent test. Put another way, you can assume that any value passed to a function in a list of validators has passed all previous test. A list of test is essentially a long "and" clause.
Validations apply only on changes to their leaf; they are not run on the initial value. If a vaildation test returns errors, the change is rejected. No update will be broadcast to subscribers, an error will be thrown, and the value will be what it was before the change was requested.
Filters are (optional) "cleanup" functions that cleanse or transform candidate values. This can include:
- rounding numeric values
- removing "junk" characters or padding around string values
- limiting arrays to a maximum size
- asserting default properties on object type
Filters occur before validation. If they throw, they will (like a failed validation) "roll back" the value's assertion.
Filters will be applied when the initial value is asserted in the Leaf (and therefore the forest) constructor; if they throw on the initial value the entire Forests' construction will fail.
Every Leaf instance has a .do property, that contains actions that can be run on the
Leaf. The Forest has a shortcut to its root's .do property: myForest.do === myForest.root.do.
This property is a blend of setters and custom actions.
Object and Map type Leaf instances have .do properties that update the keyed values.
The setters are based on the original keys of the Leaf instance's values. if the type of the
leaf changes, these setters will be removed from the do property until the leaf's type is
changed to an object or Map. So, if an object type is changed to a Map type, it will still have setters
but they will change the map's keys.
Setter methods are in the form of set_(name); keys with un-stringable names will not have setters -- but
will not be errors. (note - this is a breaking change from earlier versions which camelcased setters. this is
to respect identical keys with different casings.)
If you manually provide a method with the same name, it will override the default setter.
(hint: you can always call leaf.set(key value) or myForest.set(key, value).
Setter actions call myleaf.set(key, value) and like that method, they can throw if invalid values are submitted.
Setting keys mapped to children will update the entire value of the child leaf. (see below).
Custom actions are user-created actions.
- They may or may not return a value.
- They should (see below) be synchronous.
- They are wrapped in a transaction; if there is an untrapped error, \ all change that occurred in the action will be reverted.
- You can call setter actions and custom actions from inside an action.
const point = new Forest({
$value: { q: 10, y: 20 }, actions: {
double(leaf) {
// @ts-ignore
for (const [key, value] of leaf.store.iter) {
leaf.set(key, 2 * value);
}
},
addToEach(leaf, add) {
for (const [key, value] of leaf.store.iter) {
leaf.set(key, value + add);
}
},
addAndDouble(leaf, add) {
leaf.do.addToEach(add);
leaf.do.double();
}
},
highestValue(leaf) {
return Math.max(...leaf.store.values);
}
});
point.addToEach(100); // note
console.log('offset leaf:', point.value);
// offset leaf: {q: 110, y: 120}
console.log('highest:', point.do.highestValue());
// highest: 120
point.addAndDouble(5);
console.log('final leaf:', point.value);
// final leaf: {q: 230, y: 250}Action arguments are passed into the handler defined in the constructor actions property. The leaf itself is passed as the implicit first argument; it can be examined, called, changed, and you can call actions within actions.
subscribers will only be sent an updated value
- when all actions have completed
- and no errors have been thrown
An untrapped error will revert the entire action's activity -- and that of any actions it calls. so if you do a number of operations and one of them fails your state will be reset to before any of the activity within that action was taken.
Within the action, if you examine the leaf's value, it will be immediately updated with the (pending) values your actions have submitted.
The value of Forest is that all changes are submitted in a transactional cycle. At any time, there may be multiple pending transactions, and when one of them fails, it, and any pending transactions, will be reverted, and the state of the Forest will be that which was in place before the reverted transactions were submitted.
Catching errors inside transctionally bound activity can contain the collapsing of pending transactions.
However, while there are pending transactions, they will be exposed as leaf values immediately. These pending values will not be broadcast to subscribers, but they will be shown as the leaf's value.
This means you have the most immediate values at all times when you inspect the leaves, and your subscribers are guaranteed to only receive values that have passed the filter/validation process.
Say you have an append method that adds the arguments to a list. However it throws when the appended value is not a number. If you do the appending inside an action and it fails, no values will end up being (permanantly) added. However, if you trap the adding of values, you can resume adding the other numbers -- or at least you can keep the ones you addded up to that point.
the relevant test:
const makeList = () => new Forest({
$value: [],
actions: {
append(leaf, added) {
if (typeof added !== 'number') {
throw new Error('non-numeric value passed to append');
}
leaf.value = [...leaf.value, added]
},
appendMany(leaf, list: any[]) {
for (const val of list) {
leaf.do.append(val);
}
},
appendManyOrStop(leaf, list: any[]) {
for (const val of list) {
try {
leaf.do.append(val);
} catch (err) {
//@ts-ignore
console.log('appendManyOrStop ending:',val, err.message);
return;
}
}
},
appendManyIfGood(leaf, list: any[]) {
for (const val of list) {
try {
leaf.do.append(val);
} catch (err) {
//@ts-ignore
console.log('appendManyOrStop skipping:',val, err.message);
// note - will continue with other values
}
}
}
}
});
/*
error: non-numeric value passed to append
appendManyOrStop ending: six non-numeric value passed to append
appendManyOrStop skipping: eight non-numeric value passed to append
*/
it('should purge ALL changes from an action if there is an uncaught failure', () => {
const point = makePoint(0, 0, 0, {
actions: {
setXYZ(leaf: leafI, x: any, y: any, z: any) {
leaf.set('x', x);
leaf.set('y', y);
leaf.set('z', z);
}
}
});
point.do.setXYZ(10, 20, 30);
expect(point.value).toEqual({ x: 10, y: 20, z: 30 });
let e;
try {
point.do.setXYZ(40, 50, 'sixty');
} catch (err: any) {
e = err;
}
expect(e?.message).toMatch(/cannot add value of type string to leaf root:.* \(type number\)/);
expect(point.value).toEqual({ x: 10, y: 20, z: 30 });
});As you can see -- not trapping errors puts you in "all or nothing mode." But, if you trap errors, you can choose how to handle the failure and contain the problems, retaining all the effort done up to that point.
You have the option of defining child leaves; these are leaves whose entire values are mapped to keys in the root leaf.
This gives you the option of:
- adding actions to the child Leaf instances
- adding validators to sub-parts of the leaf
- ensuring sub-parts of a leaf are type-locked
Defining a deep schema is optional. It also requires you to keep the parent value to be a container type. (consider type-locking it, as described in "validators" above).
Child leaves allow you to, for instance, reuse patterns of convenience such as defining structures for forms, including error messages, labels, options, etc.
Setting a root value with a key that is mapped to a child will transmit the value downwards to that child. similarly, changing a child's value will update the parent appropriately.
You can nest children to whatever depth you want.
Children are intended to be stable structural elements -- collections, etc. If you want to create structures whose element roster changes regularly, contain those elements in a Leaf but do not make the changing elements themselves children.
(future development will allow for better handling of dynamic children. )
Children are attached to their parents by a reference on the child that has the child id and the key it links to.
From the Child, there is an (optional) parentId that describes the child instance's parent. That being said,
there is no field in the child that defines exactly where in the parent the child is attached. To reduce
redundancy, this can be found only by exploring the parent's .childKeys collection.
Sometimes you want to get a computed value from a Leaf. You can do this by defining a selector. Selectors are functions that take the value of a leaf (and optional arguments) and pass out a value computed from the leaf.
Actions -- see above --- can return a value. However they also occur in a transactional buffer, and that can slow down execution. Selectors when called as executed without any insulation.
Selectors are intended to be immutable and not change the value of a Leaf. As such, the entire Forest is temporarily frozen during the execution of a selector: no actions or setters can be performed until a selector completes. (the reverse is not true -- actions can call selectors.)
Care is taken to "unfreeze" a leaf if a selector throws, but if for some reason a leaf is locked up,
calling leaf.unfreeze() will clear out its frozen status.