useUnderscoreReferenceColumns=true (the wheels new default) produces columns the belongsTo/hasMany foreignKey default never matches

[bpamiri]

Summary

useUnderscoreReferenceColumns (framework default false, true in the wheels new template) changes the columns the migrator generates — t.references("user") produces user_id instead of userid — but the model layer's association default foreign key is not aware of the setting. It is unconditional string concatenation:

// vendor/wheels/model/sql.cfc:1334-1339
if (!Len(local.classAssociations[local.name].foreignKey)) {
    ...
    local.classAssociations[local.name].foreignKey =
        local.associatedClass.$classData().modelName
        & Replace(local.associatedClass.$classData().keys, ",", ",#...modelName#", "all");

So belongsTo(name="tenant") defaults to tenantid, while the migrator on a new app has created tenant_id.

The result is that a stock wheels new app which declares an association without an explicit foreignKey and then uses include= throws at runtime:

key [tenantid] doesn't exist

Grepping the framework confirms the setting never reaches the model layer — useUnderscoreReferenceColumns appears only in events/init/orm.cfm (where it is defined), migrator/Migration.cfc, migrator/TableDefinition.cfc, their specs, and the reference docs. There is no code path under vendor/wheels/model/ that consults it.

The docs assert the opposite

vendor/wheels/migrator/CLAUDE.md:51:

The wheels new template at cli/lucli/templates/app/config/settings.cfm opts new apps into true so they match Wheels model belongsTo defaults out of the box.

That sentence is the trap: it tells the reader the two halves line up, so nobody adds an explicit foreignKey. They line up only for the column name in the database; the association default still derives tenantid.

Reproduction

On a stock wheels new app (so set(useUnderscoreReferenceColumns=true)):

// migration
t = createTable(name="memberships");
t.references(columnNames="user");    // creates user_id
t.references(columnNames="tenant");  // creates tenant_id
t.create();

// app/models/Membership.cfc
component extends="Model" {
    function config() {
        belongsTo(name="user");
        belongsTo(name="tenant");
    }
}

model("Membership").findAll(include="tenant");   // throws: key [tenantid] doesn't exist

Plain finds work; the failure only appears once an association is traversed (include=, or the association accessor), which is why it can sit latent for a long time. In our case it shipped to a live screen — a tenant-picker page for users belonging to more than one tenant — and was only found weeks later when an unrelated task happened to add the first include= on that model.

Suggested fix

Options, roughly in order of preference:

1. Make the association default honor the setting. In the !Len(foreignKey) branch of sql.cfc, insert _ between the model name and each key when $get("useUnderscoreReferenceColumns") is true. That makes migrator output and association defaults genuinely agree, which is what the docs already promise. It is a behavior change for any existing app that has the flag on and relies on the current (broken-for-them) default — but such an app cannot be working today, precisely because the default doesn't match its schema.
2. If changing the default is considered too invasive, correct migrator/CLAUDE.md:51 and the references() reference doc to state plainly that belongsTo/hasMany still default to modelnameid and that apps with the flag enabled must pass foreignKey="<name>_id" explicitly.
3. Either way, consider a dev-mode assertion: when an association's derived default foreign key does not exist among the target table's columns, throw a descriptive error at association-resolution time rather than surfacing key [tenantid] doesn't exist from deep inside the join builder. That message currently gives no hint that the fix is a foreignKey argument.

Option 1 plus option 3 would have prevented this class of bug outright.

Environment

Wheels 4.0.5 (vendored), Lucee 7, PostgreSQL-wire (YugabyteDB). The same derivation is present at current dev HEAD.

Happy to open a PR for whichever option you prefer — option 2 is trivial, and I can add specs for option 1 covering both flag states.

/cc @claude