[bug] v3: false-positive "duplicate attribute" when a struct attribute shares a name with a field attribute (regression in edadb203929) #28804

Describe the bug

Since commit edadb20 ("parser: validate attribute declarations", 2026-09-19), the new compiler rejects valid code whenever a struct-level attribute has the same name as a field-level attribute of that struct.

The duplicate check added to parse_field_attrs_with_kinds_mode() also compares the current attribute against p.pending_decl_attrs. While a struct body is parsed, p.pending_decl_attrs still holds the attributes of the enclosing struct declaration, so a field attribute such as @[comment: ...] is falsely reported as a duplicate of the struct-level @[comment: ...].

This pattern is used in V's own tests (vlib/db/pg/pg_orm_test.v, vlib/db/mysql/mysql_orm_test.v), and v -check on them now fails.

Reproduction Steps

main.v:

module main

@[comment: 'struct comment']
@[table: 'foo']
pub struct Foo {
pub:
	id string @[comment: 'UUID'; primary; sql_type: 'CHAR(36)']
}

fn main() {}

$ v -new-compiler run main.v
main.v:7:14: error: duplicate attribute `comment`
    5 | pub struct Foo {
    6 | pub:
    7 |     id string @[comment: 'UUID'; primary; sql_type: 'CHAR(36)']
      |                 ~~~~~~~
    8 | }

The same happens with @[json: 'x'] on the struct and on a field, and with the default compiler (v run main.v). One error is emitted per matching field attribute.

Existing in-tree reproduction:

$ v -check vlib/db/pg/pg_orm_test.v
vlib/db/pg/pg_orm_test.v:47:22: error: duplicate attribute `comment`

TestCommentAttribute has @[comment: 'This is a table comment'] on the struct and @[comment: 'real user name'] on the name field. vlib/db/mysql/mysql_orm_test.v:43 fails the same way.

Expected Behavior

The code compiles. Attributes on a declaration and attributes on its fields are independent scopes; a struct-level @[comment: ...] must not conflict with a field-level @[comment: ...].

Current Behavior

False duplicate attribute errors stop compilation. In a real project this reaches the diagnostic limit:

model/schema_iam/iam_api_key.v:9:19: error: duplicate attribute `comment`
model/schema_iam/iam_api_key.v:10:19: error: duplicate attribute `comment`
...
error: too many errors; stopping after 100 diagnostics

v -old-compiler main.v (V1 fallback) compiles the repro fine; v and v -new-compiler fail, so this is a new-compiler regression.

Possible Solution

The p.pending_decl_attrs comparison is needed for duplicates across separate declaration-level @[...] groups (see vlib/v/parser/tests/fn_attributes_duplicate_multiple.vv), but it must not run while parsing a declaration's fields. Please make that cross-group check apply to declaration attributes only (e.g. pass an explicit flag from parse_pending_decl_attrs()), and add a regression test for a struct with same-named struct-level and field-level attributes.

The offending code, added by edadb20, is in vlib/v/parser/parser.v:3859 at bc640f2:

if piece_name.len > 0
	&& (attrs.any(it.all_before(':').trim_space() == piece_name)
		|| p.pending_decl_attrs.any(it.all_before(':').trim_space() == piece_name)) {
	p.record_diagnostic_span('duplicate attribute `${piece_name}`', piece_start, piece_end)
}

V version

V 0.5.2 bc640f2 (master, 2026-09-20)

Environment details (OS name and version, etc.)

Linux x86_64, Deepin 25, kernel 6.18.48-amd64-desktop-rolling, gcc 12.3.0.