feat: ADR-0036 Phase 1 — validate advanced-mode INSERT literals + show the value

Capture literal VALUES at parse onto Command::SqlInsert (no grammar change,
no reparse); validate them against column types before the still-verbatim
insert (reusing impl_value_for for DSL-parity wording); read them in the
error enricher so a constraint error names the real value. Execution,
auto-fill, and command identity unchanged. Adds run_sql_insert_with_literals
(runtime path); run_sql_insert stays the no-capture raw entry.

Proven: malformed date 2025/01/15 now refused in advanced-mode SQL; replayed
UNIQUE shows the real value. Tests +3 (expression runs, multi-row, natural
order) + 2 flipped/strengthened. 1930 pass / 0 fail / 0 skip; clippy clean.
This commit is contained in:
claude@clouddev1
2026-05-26 21:58:25 +00:00
parent dc9a4759ce
commit 1d5534b2bd
8 changed files with 312 additions and 22 deletions
+69
View File
@@ -696,6 +696,10 @@ enum Request {
listed_columns: Vec<String>,
row_source: String,
returning: bool,
/// Captured literal `VALUES` (per row, per position; `None` =
/// expression) for app-level type validation before the verbatim
/// insert (ADR-0036 Phase 1).
literal_rows: Vec<Vec<Option<Value>>>,
reply: oneshot::Sender<Result<InsertResult, DbError>>,
},
/// Run a grammar-validated SQL `UPDATE` (ADR-0033 §2). The
@@ -1444,6 +1448,12 @@ impl Database {
/// `sql` is the grammar-validated statement text; `source` is
/// the literal submitted line for `history.log`; `target_table`
/// is the parsed target whose CSV is re-persisted.
/// Run a grammar-validated SQL `INSERT` with **no** captured literals
/// (no app-level value validation — the verbatim ADR-0033 path). Used
/// by worker-level callers that build the statement directly. The
/// runtime, which has a parsed command, uses
/// [`Self::run_sql_insert_with_literals`] instead so the literals are
/// validated (ADR-0036 Phase 1).
pub async fn run_sql_insert(
&self,
sql: String,
@@ -1452,6 +1462,34 @@ impl Database {
listed_columns: Vec<String>,
row_source: String,
returning: bool,
) -> Result<InsertResult, DbError> {
self.run_sql_insert_with_literals(
sql,
source,
target_table,
listed_columns,
row_source,
returning,
Vec::new(),
)
.await
}
/// As [`Self::run_sql_insert`], plus the literal `VALUES` captured at
/// parse (per row, per position; `None` = expression) so the worker
/// can validate each literal against its column type before the
/// (still verbatim) insert and the error layer can name the offending
/// value (ADR-0036 Phase 1).
#[allow(clippy::too_many_arguments)]
pub async fn run_sql_insert_with_literals(
&self,
sql: String,
source: Option<String>,
target_table: String,
listed_columns: Vec<String>,
row_source: String,
returning: bool,
literal_rows: Vec<Vec<Option<Value>>>,
) -> Result<InsertResult, DbError> {
let (reply, recv) = oneshot::channel();
self.send(Request::RunSqlInsert {
@@ -1461,6 +1499,7 @@ impl Database {
listed_columns,
row_source,
returning,
literal_rows,
reply,
})
.await?;
@@ -2434,6 +2473,7 @@ fn handle_request(
listed_columns,
row_source,
returning,
literal_rows,
reply,
} => {
snapshot_then(snap, batch, conn, source.as_deref(), reply, || do_sql_insert(
@@ -2445,6 +2485,7 @@ fn handle_request(
&listed_columns,
&row_source,
returning,
&literal_rows,
));
}
Request::RunSqlUpdate {
@@ -8388,10 +8429,38 @@ fn do_sql_insert(
listed_columns: &[String],
row_source: &str,
returning: bool,
literal_rows: &[Vec<Option<Value>>],
) -> Result<InsertResult, DbError> {
debug!(sql = %sql, table = %target_table, returning, "sql_insert");
let canonical_table = require_canonical_table(conn, target_table)?;
let target_table = canonical_table.as_str();
// ADR-0036 Phase 1: validate captured literal VALUES against their
// column types BEFORE the (still verbatim) insert runs — sharing the
// DSL's per-type validators (`impl_value_for`) for identical wording.
// Only literal positions are checked; expression positions (`None`)
// are left to the engine. Column mapping mirrors the engine's: an
// explicit column list maps by position; natural order maps to the
// schema's columns in definition order. An out-of-range position
// (arity mismatch) is left for the engine / parse-time diagnostic.
// Execution below is unchanged (no binding, no auto-fill change).
if literal_rows.iter().any(|r| r.iter().any(Option::is_some)) {
let schema = read_schema(conn, target_table)?;
let columns: Vec<&str> = if listed_columns.is_empty() {
schema.columns.iter().map(|c| c.name.as_str()).collect()
} else {
listed_columns.iter().map(String::as_str).collect()
};
for row in literal_rows {
for (idx, slot) in row.iter().enumerate() {
if let Some(value) = slot
&& let Some(col) = columns.get(idx)
{
impl_value_for(&schema, col, value)?;
}
}
}
}
// The `shortid` auto-fill rewrite reconstructs only `INSERT …
// VALUES …` and would drop any trailing clause — `ON CONFLICT …`
// (3h) and/or `RETURNING …` (3g). `row_source` is the clean