feat: .^create on belongs-to relationship with auto FK by hermes-fco · Pull Request #596 · FCO/Red

hermes-fco · 2026-06-11T14:41:24Z

What

Port and fix of PR #523 (from 2021).

Enables calling .^create directly on a belongs-to relationship accessor, with automatic foreign key management:

my $ble = Ble.^create(:value<ble>);
my $bla = $ble.bla.^create(:value<bla>);  # FK auto-set on $ble!

Changes

1. Fix: `ast-value .ref` in relationship-ast

String model references now work correctly (Red::AST::Eq.new: $_, ast-value .ref: $t2)

2. Add `parent`/`join-on` to alias role

Enables tracking the parent object for FK auto-set during create.

3. Enhance `.^create` metamodel method

When invoked on a relationship alias (has join-on + parent), creates the related object and automatically sets the FK on the parent, then saves it.

4. Tests

belongs-to create with auto FK
Isolation between different parent instances

Files

lib/MetamodelX/Red/Model.rakumod — create method + alias role
lib/Red/Attr/Relationship.rakumod — ast-value fix
t/35-create.rakutest — new tests

Closes #523

Port from PR FCO#523 (2021): 1. Fix ast-value .ref in relationship-ast — string model references now work correctly in relationship conditions. (Red::AST::Eq.new: $\_, ast-value .ref: $t2) 2. Add parent/join-on to alias role — enables tracking the parent object in relationships, needed for auto FK on create. 3. Enhance .^create to handle relationship accessors: when called via $ble.bla.^create(...), automatically sets the foreign key on the parent object and saves it. 4. Tests: belongs-to create with auto FK, isolation between instances. Closes FCO#523

- String model references (:model<Name>) - Type model references (:model(Type)) - has-one create via relationship accessor - Multiple creates on same parent - Create with FK already set - Isolation between parent instances - Extra attributes passthrough - Unsaved parent error case - FK default=0 edge case - To-many .create regression checks

FCO · 2026-06-11T15:44:34Z

 #| And Lists and/or Hashes for relationships
-multi method create(\model where *.DEFINITE, *%orig-pars, :$with where not .defined) is hidden-from-backtrace is rw {
-    die "Cannot call .^create on a defined model." if model.DEFINITE;
+multi method create(\\mo where *.DEFINITE, *%orig-pars, :$with where not .defined) is hidden-from-backtrace is rw {


Why \mo? Shouldn't it be \mo???

Verdade! Era escaping bug — o arquivo ficou com 4 barras em vez de 1. Corrigido em 2ff9196.

FCO · 2026-06-11T15:45:22Z

-multi method create(\model where *.DEFINITE, *%orig-pars, :$with where not .defined) is hidden-from-backtrace is rw {
-    die "Cannot call .^create on a defined model." if model.DEFINITE;
+multi method create(\\mo where *.DEFINITE, *%orig-pars, :$with where not .defined) is hidden-from-backtrace is rw {
+    my \\model = mo.^orig;


Should it be a single \???

Mesmo problema de escaping. Corrigido junto.

…e expected) - t/35-create.rakutest: :default(0) → :default{ 0 } fixes 'Type check failed in assignment to &!default' - lib/Red/Column.rakumod: .perl() → .raku() - lib/Red/Model.rakumod: .get_value(self).perl → .get_value(self).raku - lib/X/Red/Exceptions.rakumod: $.orig-exception.perl → $.orig-exception.raku - lib/Red/Driver/Mock.rakumod: $re.perl() → $re.raku() - lib/Red/Cli.rakumod: %pars.map(*.perl) → %pars.map(*.raku) - lib/Red/Driver.rakumod: @bind.perl()/@binds.perl() → @bind.raku()/@binds.raku()

hermes-fco · 2026-06-22T02:37:59Z

Code Review — feat: .^create on belongs-to relationship with auto FK

Verdict: Comment (0 critical, 2 warnings, 4 suggestions)

Port and fix of PR #523 (from 2021). Enables $ble.bla.^create(:value<bla>) on belongs-to relationship accessors with automatic FK propagation to the parent object. +216 / −18 lines, 9 files. Also includes .perl → .raku deprecation fixes across 6 lib files.

✅ Looks Good

lib/Red/Attr/Relationship.rakumod:199 — ast-value .ref: $t2 fix is correct. Without ast-value, string model references produced a plain value instead of an AST node, breaking query generation. This unblocks PR WiP: .^create from belongs-to relationship #523's original goal.
lib/MetamodelX/Red/Model.rakumod:260 — method parent(|) is rw { $ } — clean Raku state-variable pattern for tracking the parent object in the RAlias role.
lib/MetamodelX/Red/Model.rakumod:525 — my \model = mo.^orig cleanly separates the possibly-aliased invocant from the un-aliased model type for internal operations.
lib/MetamodelX/Red/Model.rakumod:267 — method join-on($, \a = alias) allows passing a custom alias for AST generation in the create context.
Test coverage — Comprehensive: string model refs, type model refs, isolation between parents, multiple creates on same parent, FK replacement, has-one reverse direction, to-many verification, unsaved parent error, and FK=0 edge case. Tests use ^refresh after FK mutations — good practice.
FCO's review comments — Escaping bug (\\\\mo → \\mo) was correctly fixed in commit 2ff9196. The :default(0) → :default{ 0 } fix (Callable expected) in commit 9c7da89 is also correct.
Security scan — Clean (no secrets, debug statements left in, or merge conflicts).
.perl → .raku migration is complete and correct across all 6 lib files.

⚠️ Warnings

lib/MetamodelX/Red/Model.rakumod:524-526 — The die "Cannot call .^create on a defined model." if mo.DEFINITE on line 526 fires for any DEFINITE invocant, which the where *.DEFINITE constraint guarantees. The new auto-FK code at lines 620-640 is below this die. However, this same die + where pattern existed BEFORE this PR (it was the original guard). In practice, $ble.bla may return an UNDEFINITE type object (the alias type), causing dispatch to fall through to another candidate. Verify that $ble.bla.^create(:value<bla>) correctly reaches the auto-FK path — if $ble.bla is DEFINITE, the die would prevent reaching line 620. The tests pass, so either the dispatch path is different from what the where *.DEFINITE suggests, or the tests exercise a different mechanism.
lib/MetamodelX/Red/Model.rakumod:620 — mo.HOW.?join-on(mo) and mo.HOW.?parent(mo) — these are meta-methods from the RAlias role. If mo is a plain type object without RAlias (e.g., someone calls .^create on a model instance directly), these .HOW.? calls are safe (return Nil). But the method would proceed to line 642 and hit the Proxy fallback, which is fine. Consider an explicit guard:
```
die "Cannot call .^create on a defined model." 
    unless mo.HOW.?join-on(mo) && mo.HOW.?parent(mo);
```
This would be clearer than relying on the pre-existing die at line 526.

💡 Suggestions

lib/MetamodelX/Red/Model.rakumod:631 — mo.^join-on(mo.^parent).should-set.Hash — should-set is called as a method. If it returns Nil or a non-Hashable type, this line will crash. Consider a guard:
```
my %should-set = mo.^join-on(mo.^parent).?should-set.Hash // Empty
```
lib/MetamodelX/Red/Model.rakumod:633 — $p.^columns.map: { .name.substr(2) => .self } — assumes all column names have a 2-character prefix (e.g., $.). This is true for Red's default attribute naming but would break if custom attribute names don't follow this convention. Worth a comment.
Commit hygiene — Commit 9c7da89 mixes .perl → .raku changes (already in PR test: increase coverage — between, like, modulo, sort, tail, delete, date ops #597) with the :default(0) → :default{ 0 } fix for the Bly model. The .perl→.raku changes will cause a merge conflict with PR test: increase coverage — between, like, modulo, sort, tail, delete, date ops #597. Consider rebasing to separate concerns.
Test: line 270 — throws-like { $ble.bla.^create: :value<should-fail> }, Exception — catching the generic Exception type is fragile; if the error message changes, the test still passes but for the wrong reason. Consider matching X::Red::* or a more specific exception type.

🔍 Minor observations (non-blocking)

The RAlias parent method uses is rw { $ } — this is a Raku state variable that stores one value per closure. When accessed via mo.^parent, it returns the Container. mo.^parent = $ble sets it. Clean pattern.
Line 634-636: $p.^set-attr + $p.^set-dirty + $p.^save — this pattern matches Red's existing mutation style.
The PR closes WiP: .^create from belongs-to relationship #523 — the original "WiP" PR from 2021 is being superseded by this implementation.

📊 Stats

Metric	Value
Files changed	9
Lines added	+216
Lines removed	−18
New tests	t/35-create.rakutest (10 new subtests)
Security scan	Clean

Reviewed by Hermes Agent

Hermes Agent added 2 commits June 11, 2026 14:41

FCO reviewed Jun 11, 2026

View reviewed changes

Hermes Agent added 2 commits June 11, 2026 15:53

Fix escaping: \\mo → \mo, \\model → \model

2ff9196

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: .^create on belongs-to relationship with auto FK#596

feat: .^create on belongs-to relationship with auto FK#596
hermes-fco wants to merge 4 commits into
FCO:masterfrom
hermes-fco:feature/create-from-relationship

hermes-fco commented Jun 11, 2026

Uh oh!

FCO Jun 11, 2026

Uh oh!

hermes-fco Jun 11, 2026

Uh oh!

FCO Jun 11, 2026

Uh oh!

hermes-fco Jun 11, 2026

Uh oh!

hermes-fco commented Jun 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

hermes-fco commented Jun 11, 2026

What

Changes

1. Fix: ast-value .ref in relationship-ast

2. Add parent/join-on to alias role

3. Enhance .^create metamodel method

4. Tests

Files

Uh oh!

FCO Jun 11, 2026

Choose a reason for hiding this comment

Uh oh!

hermes-fco Jun 11, 2026

Choose a reason for hiding this comment

Uh oh!

FCO Jun 11, 2026

Choose a reason for hiding this comment

Uh oh!

hermes-fco Jun 11, 2026

Choose a reason for hiding this comment

Uh oh!

hermes-fco commented Jun 22, 2026

Code Review — feat: .^create on belongs-to relationship with auto FK

✅ Looks Good

⚠️ Warnings

💡 Suggestions

🔍 Minor observations (non-blocking)

📊 Stats

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

1. Fix: `ast-value .ref` in relationship-ast

2. Add `parent`/`join-on` to alias role

3. Enhance `.^create` metamodel method