Question

Given an OO language in which the naming convention for object properties is camelCased, and an example object like this:

{
    id: 667,
    firstName: "Vladimir",
    lastName: "Horowitz",
    canPlayPiano: true
}

How should I model this structure in a PostgreSQL table? Options:

1. unquoted camelCase column names
2. quoted camelCase column names
3. unquoted (lowercase) names with underscores

Each have drawbacks:

1. Unquoted identifiers fold to lowercase. You can create a table with a canPlayPiano column, but the mixed case never reaches the database. The column will show up as canplaypiano in psql, pgAdmin, explain results, error messages, everwhere.

2. Quoted identifiers keep their case, but once you create them like that you will always have to quote them. If you create a table with a "canPlayPiano" column, a SELECT canPlayPiano ... will fail. This adds noise to all SQL statements.

3. Lowercase names with underscores are unambiguous but they don’t map to the names the application language is using. You will have to remember to use different names for storage (can_play_piano) and code (canPlayPiano). It also prevents certain types of code automation where properties and database columns need to be named the same.

Whatever I do, some part is going to feel awkward. I’ve been using option 3, but keep hoping for a better solution. The case folding and the need for quotes is coming from the SQL standard (PostgreSQL’s adaptation of the standard). I know how it works; I’m more interested in best practices than explanations about how PostgreSQL handles identifiers.

Answer 1

Given that PostgreSQL uses case-insensitive identifiers with underscores, should you change all your identifiers in your application to do the same? Clearly not. So why do you think the reverse is a reasonable choice?

The convention in PostgreSQL has come about through a mix of standards compliance and long-term experience of its users. Stick with it.

If translating between column-names and identifiers gets tedious, have the computer do it – they’re good at things like that. I’m guessing almost all of the 9-million database abstraction libraries out there can do that. If you have a dynamic language it’ll take you all of two lines of code to swap column-names to identifiers in CamelCase.