fix: Skip embed form standalone validation#474
Merged
Conversation
Nested forms backed by an embedded attribute/argument were validated
twice: once against the embed's own create/update action, and again
inline when the parent action cast the merged params. The standalone
pass bypasses parent-level logic (e.g. a `cast_input/2` override on
the embed that collapses blank input to `nil`), so an all-blank embed
sub-form produced spurious "is required" errors even when the parent
would have accepted the value as nil.
After validating, clear `source.errors` and mark the source valid for
forms with `embed?: true`. `carry_over_errors/2` replays any real
errors the parent's validation surfaces with a matching path. Forms
with `transform_params` set are skipped — unions rewrite params into
an `%Ash.Union{}` before handing them up, so their standalone
validation is the only line of defense.
`clear_embed_source_errors/2` used `!opts[:transform_params]` to detect union sub-forms. `transform_params` is also a public form option, so a caller setting it on a non-union embed (undocumented but reachable) would silently disable the exemption. Set `union?: true` in `unions/3` and check that instead.
Split the embed-defers-validation test into two: * A structural test asserting that the sub-form's `source.valid?` is `true` and field errors flow only through `carry_over_errors/2` from the parent. This demonstrates the patch's direct effect without relying on any consumer-side `cast_input` override. * A user-facing test demonstrating what the patch enables: an "all-or-nothing" embedded value (whose `cast_input` collapses blank input to `nil`) accepting blank input on a parent with a nullable embed attribute. Add a moduledoc to the `AshPhoenix.Test.Address` fixture explaining the `cast_input` override as the canonical "all-or-nothing embedded value" pattern rather than test-specific scaffolding.
sevenseacat
changed the title
zachdaniel
approved these changes
May 14, 2026
Contributor
|
🚀 Thank you for your contribution! 🚀 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
As discussed on Discord, there's a bit of a gap in how AshPhoenix treats forms for embedded resources vs. related resources.
Embedded sub-forms are currently validated standalone against the embed's own
:create/:updateaction, in addition to the parent action's inline cast of the same input. The standalone pass is redundant for embeds, and blocks legitimate patterns such as collapsing entirely-blank input to nil (instead of storing an embed of empty data).How this was observed
cast_inputoverrides are silently ignored.I think it's relatively common to have "all-or-nothing" embedded values (postal address, anything with "fill completely or leave empty" UX), via overriding
cast_input/2on the embed to collapse blank input to{:ok, nil}. The parent'sallow_nil?: trueattribute then accepts the value as nil.Pre-patch, the sub-form's strict standalone validation overrides this choice and the form is invalid for blank input.
Sub-form errors duplicate parent-cast errors.
After
carry_over_errors/2injects parent errors into the sub-form, both sets coexist insource.errors. Downstream renderers see two visible errors per missing field, each tagged with different breadcrumbs.Why unions are excluded
Unions rewrite their params into an
%Ash.Union{}struct before handing the value to the parent, so the parent can't catch input errors at the cast layer — the standalone validation is the only line of defense for unions. The patch tags union sub-forms withunion?: trueinAshPhoenix.Form.Auto.unions/3and excludes them.manage_relationshipsub-forms don't haveembed?: trueset, so the patch's clause doesn't fire for them.Contributor checklist
Leave anything that you believe does not apply unchecked.