framework VS drydock

I did implement PSR7 in the develop branch though.

https://github.com/Divergence/framework/blob/6af1b6b0e56b25c...

Where people go wrong is trying to do this field normalization logic inside their query builder. That leads to all sorts of problems. The query builder shouldn't know what is going on with your ORM at all. It's just gluing strings together to form a query.

You can see my dead simple query builder here: https://github.com/Divergence/framework/tree/develop/src/IO/...

I honestly used to not have one as it's so simple but decided it was a good abstraction to assimilate from other frameworks. My query builder does not attempt to sanitize anything. The ActiveRecord class takes care of that through the data mapping conversion functions for all the basic HTTP CRUD functionality.

If you believe you can find a security hole in my ORM I'm all ears.

I wrote my own like 10 years ago https://github.com/Divergence/framework/blob/release/src/Models/ActiveRecord.php

I have been using SQLite in Go projects for a few years now. During early stages of development I always start with SQLite as the main database, then when the project matures, I usually add support for PostgreSQL.

(I usually make a Store interface which is application specific and doesn't even assume there is an SQL database underneath. Then I make "driver" packages for each storage system - be it PostgreSQL, SQLite, flat files, timeseries etc. I have only one set of unit tests that is then run against all drivers. And when I have a caching layer, I also run all the unit tests with or without caching. The cache is usually just an adapter that wraps a Store type. I maintain separate schemas and drivers for each "driver" because I have found that this is actually faster and easier than trying to make generic SQL drivers for instance.)

However, I always keep the SQLite support and it is usually the default when you start up the application without explicitly specifying a database. This means that it is easy for other developers to do ad-hoc experiments or even create integration tests without having to fire up a database, which even when you are able to do it quickly, still takes time and effort. In production you usually want to point to a PostgreSQL (or other) database. Usually, but not always.

I also use it extensively in unit tests (often creating and destroying in-memory databases hundreds of times during just a couple of seconds of tests). I run all my tests on every build while developing and then speed matters a lot. When testing with PostgreSQL I usually set a build tag that specifies that I want to run the tests against PostgreSQL as well. I always want to run all the database tests - I don't always need to run them against PostgreSQL

(Actually, I made a quick hack called Drydock which takes care of creating a PostgreSQL instance and creates one database per test. This is experimental, but I've gotten a lot of use out of it: https://github.com/borud/drydock)

The reason I do this is that it results in much quicker turnaround during the initial phase when the data model may go through several complete rewrites. The lack of friction is significant.

SQLite has actually surprised me. I use it in a project where I routinely have tens of millions of rows in the biggest table. And it still performs well enough at well north of 100M rows. I wouldn't recommend it in production, but for a surprising number of systems you could if you wanted to.

The transpiled SQLite is very interesting to me for two reasons. It makes cross compiling a lot less complex. I make extensive use of Go and SQLite on embedded ARM platforms and then you either have to choose between compiling on the target platform or mess around with C libraries. It also eliminates the need to do two stage Docker builds (which cuts down building Docker images from 50+ seconds to perhaps 4-5 seconds).

The transpiled version is slower by quite a lot. I haven't done a systematic benchmark, but I noticed that a server that stores 30-40 datapoints per second went from 0.5% average CPU load to about 2% average CPU load. I'm not terribly worried about it, but it does mean that when I increase the influx of data I'm most likely going to hit a wall sooner.

I'll be using the transpiled SQLite a lot more in the coming year and I'll be on the Gophers Slack so if anyone is interested in sharing experiences, discussing SQLite in Go, please don't be shy.

This isn't new. A lot of applications and libraries do this. And I think it is a good way to design things.

Usually the database I use to develop a SQL schema is Sqlite3, since it allows for really nice testing. Then I add PostgreSQL support (which requires more involved testing setup, but I have a library that makes this somewhat easier: https://github.com/borud/drydock). (SQLite being in C is a bit of a problem since it means I can't get a purely statically linked binary on all platforms - at least I haven't found a way to do that except on Linux. So if anyone has some opinions on alternatives in pure Go, I'm all ears)

In the Java days JDBC every single method implementing some operation would be a lot of boilerplate. JDBC wasn't a very good API. But in Go that is much less of a problem. In part because you have struct tags, and libraries like Sqlx. To that I also add some helper functions to deal with result/error combos. Turns out the majority of my interactions with SQL databases can be carried out in 1-3 lines of code - with a surprising number of cases just being a oneliner. (The performance hit from using Sqlx is in most cases so minimal it doesn't matter. If it matters to you: use Sqlx when modeling and evolving the persistence, and then optimize it out if you must. I think I've done that just once in about 100kLOC worth of code written over the last few years).

And best of all: I get to deal with the database as a database. I write SQL DDL statements to define the schema, and SQL to perform the transactions. I don't have to pretend it is a object model, so I can make full use of the SQL. (Well, actually, I try to make do as far as possible with trivial SQL, but that's a whole different discussion). The interface type takes care of exposing the persistence in a way that fits the application.

(Another thing I've started experimenting with a bit is to return channels or objects containing channels instead of arrays of things. But there is still some experimenting that needs to be done to find a pleasing design)