Concept
better argument names:
// generated signature becomes
fn get_users(limit: i64, offset: i64) -> ...
// instead of
fn get_users(arg_0: i64, arg_1: i64) -> ...
The real win is just making arg_0, arg_1 not be the parameter names. That's a broader problem with your current codegen anyway, not just pagination.
Right now your macro generates this regardless of what the SQL is:
// generated for: SELECT * FROM users WHERE id = ? AND name = ? LIMIT ? OFFSET ?
pub fn get_users(arg_0: i64, arg_1: &str, arg_2: i64, arg_3: i64) -> ...
The binding types are correct but the names are useless. The fix is using what you already know from the AST:
// WHERE id = ? → you know the column is `id` → name it `id`
// WHERE name = ? → you know the column is `name` → name it `name`
// LIMIT ? → SQL keyword → name it `limit`
// OFFSET ? → SQL keyword → name it `offset`
pub fn get_users(id: i64, name: &str, limit: i64, offset: i64) -> ...
Implementation (unconfirmed)
You already have the column info in binding_patterns.rs when you traverse the WHERE clause. You're just not carrying the name through, only the type. So the change is roughly:
// current
pub struct Type {
pub base_type: BaseType,
pub nullable: bool,
pub contains_placeholder: bool,
}
// add name
pub struct BindingParam {
pub data_type: Type,
pub name: String, // "id", "name", "limit", "offset" etc
}Then in codegen instead of:
let arg_name = quote::format_ident!("arg_{}", i);You do:
let arg_name = quote::format_ident!("{}", param.name); // "id", "name" etcThe tricky cases are:
// WHERE age > ? AND age < ? → two params from same column
// name them `age_min` and `age_max`? or `age_0` and `age_1`?
// WHERE id = ? OR name = ? → harder to infer intent
// probably just fall back to `arg_0`, `arg_1`
Concept
better argument names:
The real win is just making
arg_0, arg_1not be the parameter names. That's a broader problem with your current codegen anyway, not just pagination.Right now your macro generates this regardless of what the SQL is:
The binding types are correct but the names are useless. The fix is using what you already know from the AST:
Implementation (unconfirmed)
You already have the column info in
binding_patterns.rswhen you traverse the WHERE clause. You're just not carrying the name through, only the type. So the change is roughly:Then in codegen instead of:
You do:
The tricky cases are: