Rework setup of the Operation Hooks test suite for memory connector
so that the "unoptimized" variant disables not only atomic CRUD
operations, but also geo queries.
This way we can test both "near" querying implementations:
- When the connectors supports "near" queries natively.
- When "near" queries are executed at LoopBack side in-memory.
- On include = true or include = 1 the lib crashes with
"TypeError: includes.forEach is not a function".
- checking for boolean and number type and return empty array.
- fix: include symbol and function in normalization
- fix: review changes; adding debug stmt + early exit for tests
Before this change, when a property was configured with a default value
at LoopBack side and the database was returned a record with a missing
value for such property, then we would supply use the configured
default.
This behavior is problematic for reasons explained in #1692.
In this commit, we are introducing a new model-level setting called
`applyDefaultsOnReads`, which is enabled by default for backwards
compatibility.
When this setting is set to `false`, operations like `find` and
`findOrCreate` will NOT apply default property values on data returned
by the database (connector).
Please note that most of the other CRUD methods did not apply default
values on database data as long as the connector provided native
implementation of the operation, that aspect is not changing.
Also note that default values are applied only on properties with
`undefined` values. The value `null` does not trigger application of
default values. This is important because SQL connectors return
`null` for properties with no value set.
The new version of our config enabled function-paren-newline rule,
this commit fixes the codebase to use more consistent handling
of newlines when calling functions.
Fix the regression introduced by 9af79cf51a where updateAttributes
and replaceAttributes was incorrectly handling the response returned
by the database.
This commit restores the old behaviour where `context.data` is updated
only by a connector-provided function and reverts an incorrect change
of a test file made as part of the faulty fix.
Besides the fixes, this patch also renames `data` to `dbResponse` and
add comments explaining the structure of the response object for more
clarity.
Before this change, many Operation Hooks were not correctly propagating
changes made to `ctx.data` via reassigning ctx.data to a new object.
This change modifies existing tests to account for this different
scenario, adds few more tests for scenarios that were not covered by
tests before and finally fixes the problem discovered.
Use case:
1. Configure a datasource with lazyConnect = true
2. Do NOT start the DB
3. Start the app
4. Send first request and it fails to connnect to the DB
5. Start the DB
5. Requests are now served correctly
Add API allowing consumers (e.g. LoopBack) to remove a Model from all
juggler registries:
- ModelBuilder's models
- ModelBuilder's definitions
- Connector registry of models
Rework the code building model constructors to leverage `Function` class
and dynamically emit a constructor function named after the model.
Before this change, all model classes were called "ModelConstructor",
which made debugging difficult.
After this change, a model class for model "User" is called "User.
Because not all valid model names are also valid JavaScript identifiers,
we implement a simple sanitization technique (replacing characters like
"-", "." and ":" with underscore "_") and fall back to legacy
"ModelConstructor" if the model name is still not a valid JS identifier.
The uniq function does currently not work when the database is mongodb.
In the case of mongodb, the function will receive an array of bson
object of bson type ObjectID. The indexOf function will return a
different index, even if the mongodb ID is the same, as it is wrapped
in the ObjectID. This commit first transforms any ObjectID in the array
to a string representation. We can then use indexOf to check for
uniqueness.