[
(don't) call on me (anymore) - Steve Winwood
A first class validation library for the JVM
It has been decided to halt development on Valerie. In working with this project it has become clear that core valuable pieces it provides are:
- Combinable, expressive validation feedback.
- Applying composable validation while traversing a tree/graph.
The concluding line of thought is that this library should be ported to Java and be focused on those pieces while leaving the collection of predicates/matchers to other libraries. If that is done, it is not yet clear whether there would be enough substance left to justify having a library rather than capturing it in project local code.
This is ultimately being abandoned because there has now been a sustained period during which there has been ideas for ways to improve the library, but no driving need to do so.
Valerie has been in use in the system for which it was created for several years without the need for enhancement. The initial release provides flexible, albeit often awkward, building blocks; while there has been a background desire to provide some constructs which would allow for more eloquent definitions, it's never been strictly necessary and has never gathered enough mass to be given priority. Aside from the notable deficiency of lack of documentation, it's been extensible enough to evolve with the system without issue.
There have been multiple projects for which validation has been implemented and for which there was a thought that Valerie would be pulled in (after removal of Groovy as a runtime dependency). In all of these cases, however, that never ended up happening; there was a repeated pattern of waiting for the validation requirements to grow enough in complexity, but such a tipping point was never reached. This leads to the conclusion that the confluence of forces that led to the inception of Valerie are uncommon. The key factor was likely non-trivial validation requirements which were being addressed by an implementation which was not conducive to evolution and therefore entropy dragged it into a tangle. Most services tend to have fairly simple validation requirements, and those that have or accrete more complex needs can apply some well-established practices to keep the validation in order without requiring the assistance of an additional library.
It's likely that the reference consumer of Valerie will shift to an alternative moving forward. A runtime dependency on Groovy is likely to be avoided and for the aforementioned reasons porting Valerie does not seem worthwhile, though some of the approaches herein may be implemented locally as required. The target service has a fair amount of tooling which is driven by OpenAPI and therefore a solution which is driven by JSON Schema definitions within an OpenAPI file is likely.
Most validation libraries are defined in terms of an established business model and are therefore second class in relation to that model. This leads to validation being conceptually behind the binding of input to the model, rather than in between the input and the model; to put it another way, most validation libraries are focused on validating objects rather than less constrained input. By providing first class validation that is not bound to a defined model, Valerie allows for a wider range of validation rules for an accordingly wide range of possible inputs.
The more common second class approach mentioned above (defining validation in terms of your model) is a very good one as it provides many conveniences, is generally DRY'er, and can highlight the role of the model in the code. There are times, however, when this approach becomes overly-restrictive. Some examples may include:
- providing richer feedback to less defined or more widely varying user input
- allowing the model to evolve cleanly while migrating the API to that model
- separating the read/query model from the write/command model (CQRS) or validating partial representations/patches for domain objects
Valerie is available for download from jCenter. It is presently not published to Maven Central, but can be so if desired (file an Issue). The artifacts can be manually downloaded from jCenter/BinTray by clicking on the Download link/badge at the top of this README.
Make sure jCenter is configured as a repository:
repositories {
jcenter()
}Declare a compile time dependency on Valerie:
dependencies {
compile 'com.brightcove:valerie:latest.release'
}latest.release will automatically be resolved by Gradle but using
some form of fixed version or locking is recommended: the latest
released version can be quickly retrieved from the Download badge on
the top of this REAMDE
The separation of validation is likely to lead to more work and a more fragmented domain model, but in cases like those above, second class validation may start to get in the way rather than offering help. The ensuing implementation difficulties and/or divergent changes can lead to a loss of code structure and readibility. Valerie provides a decoupled alternative which allows for consistent and declarative code.
The goal of this project is not to improve upon or replace existing validation libraries but to solve a slightly different problem. If defining your validation in terms of your model works, then you should do so: if you find your model getting blurry or polluted due to validation concerns that don't fit in cleanly, or the general validation logic has devolved into a tangle due to the restrictions of your validation framework, then switching to first class validation may make sense.