@tonykaralis
Deleting a property off of a class in the core library would then result in the change not being noticed in the Web app
How so? You broke the contract between the core library and its dependency at the assembly level. That will be a source and binary break. If you have types in your Web app derived from core types they will either break when you remove that property or they won't because the property wasn't being used.
I'm not saying there is no need for JSON schema validation, but I tend to rely on the classes as defined in my code and the consumers of those classes. If you removed a property from a POCO and it didn't break anything at a source level, what was it doing for you in the database jsonb?
@barryhagan you do make a good point. I think the most important issue I am concerned about is someone changing the contract and not realising that this will break other apps. Some apps we are working on may not require a fix or feature until a month later. By that time we realised the breaking change too late. The validation was a quick and immediate way for a developer to know whether they are breaking other apps or not. I do too rely on the classes most of the time.
The schema validation is a tricky one, it requires storing the schema somewhere and then updating only when there arent breaking changes.
Also, there are certain properties which are never used in the 'client' apps, but still need to be persisted so the core lib can make use of them later for calculations. If we dropped one of those properties in the database with valid critical data, even though the app itself doesnt use it, the core does and wed be in a pickle.
Am still trying to find a simple way of providing this check.
It allows to catch breaking changes to the codebase
although it seems to be marked as minor, then imho it contains breaking changes (at least to dependencies)
it'll force us to use at least 4.6.1 .NET and JSON.NET in v12
With JSON.NET v12 we observed some performance degradation
and with the .NET we have right now 4.6
I'd feel more safer to make this v4.0, although pragmatically we need to do this upgrade for our users
Thoughts?
about 4.6.1 upgrade I think that we could even make a step forward to 4.6.2 as afair it's the version that microsoft supports
So big thanks to @baronfel being on the watch
But 4.1.0 is already released so they won't downgrade the references
So we need to leave with that (although I hate this upgrade process)
After we decide what to do with 4.1.0, I’ll also check the 5.0 pre release. I have foreseen more troubles there - so it's better to start preparation and give us and them the chance to discuss the potential issues.
Thoughts?
Although the problem will still remain
Probably some users might want to use a newer version - eg. when they are using it together with the EF or Dapper
I’ll prepare the PR for that
I’ll also check if we can drop using Npgsql.json.net package, that’s forcing the JSON.net upgrade
Although I really don't think comfortable with not following strictly semantic versioning...
It’s even more confusing for the users more than it is for us...
We could, in theory, go with 4.0 just to be prim and proper
but then it won’t be good from the marketing perspective
As we already officially started a discussion on the plans for the v4
Note that there is already an issue raised by a user who attempted upgrading to Npgsql 4.1.0. JasperFx/marten#1359
I think that we need to do a better job of being more proactive in the version checking (eg. as @baronfel did for us). I was thinking about some nightly build with the latest prerelease of npgsql
However close we work with Npgsql versions, any breaking changes will take time to resolve for us
Nightly build with pre-release versions of Npgsql would certainly help us to uncover issues sooner than later
In general, I’m trying to find some pragmatic approach
😉