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
+9
View File
@@ -422,6 +422,15 @@ pub enum Command {
/// `DataResult` when true; otherwise it surfaces the
/// affected-row count (+ auto-show) as before.
returning: bool,
/// Captured literal values per `VALUES` row, per position
/// (ADR-0036 Phase 1). `Some(v)` for a bare literal (incl. a
/// signed number); `None` for an expression position (nothing
/// static to validate). Empty when the row source is a
/// `SELECT`/`WITH` query. The worker validates these against the
/// column types before the (still verbatim) insert; the error
/// enricher reads them to show the offending value. Execution
/// itself is unchanged — these are *not* bound.
literal_rows: Vec<Vec<Option<Value>>>,
},
/// A SQL `UPDATE` validated by the walker (ADR-0033 §2,
/// advanced mode). Grammar-as-text: the worker executes `sql`
+91
View File
@@ -950,15 +950,106 @@ fn build_sql_insert(path: &MatchedPath, source: &str) -> Result<Command, Validat
// so the validated line runs verbatim (grammar-as-text,
// ADR-0030 §4) — no keyword reconstruction.
let sql = source.trim().to_string();
// Capture literal values per `VALUES` row for app-level type
// validation + error enrichment (ADR-0036 Phase 1). Only for a
// `VALUES` source (a `SELECT`/`WITH` source has no `values` keyword,
// so this stays empty). Bounded to the row-source region by the same
// `tail_start` the row_source slice used.
let values_start = path
.items
.iter()
.find(|i| matches!(i.kind, MatchedKind::Word("values")))
.map(|i| i.span.0);
let literal_rows = values_start.map_or_else(Vec::new, |vs| {
capture_literal_rows(path, vs, tail_start.unwrap_or(source.len()))
});
Ok(Command::SqlInsert {
sql,
target_table,
listed_columns,
row_source,
returning: path_has_returning(path),
literal_rows,
})
}
/// Capture the literal values of each `VALUES` tuple from the matched
/// path (ADR-0036 Phase 1). Each position is `Some(Value)` for a bare
/// literal (incl. a signed number — the leading sign is folded into the
/// number) and `None` for an expression position (a `func(x)`, `a+1`,
/// subquery, column ref — nothing static to validate). Works purely from
/// the tokens the walker already matched (no reparse); rows and positions
/// are delimited by tuple parens and depth-1 commas. `values_start` is the
/// byte offset of the `values` keyword; only items in `[values_start,
/// tail_end)` are considered (so any trailing `ON CONFLICT`/`RETURNING`
/// clause is excluded).
fn capture_literal_rows(
path: &MatchedPath,
values_start: usize,
tail_end: usize,
) -> Vec<Vec<Option<Value>>> {
let mut rows: Vec<Vec<Option<Value>>> = Vec::new();
let mut depth: i32 = 0;
let mut cur_row: Vec<Option<Value>> = Vec::new();
let mut pos: Vec<&MatchedItem> = Vec::new();
for item in &path.items {
if item.span.0 < values_start || item.span.0 >= tail_end {
continue;
}
match &item.kind {
MatchedKind::Word("values") => {}
MatchedKind::Punct('(') => {
depth += 1;
if depth == 1 {
cur_row = Vec::new();
pos.clear();
} else {
pos.push(item);
}
}
MatchedKind::Punct(')') => {
if depth == 1 {
cur_row.push(classify_value_position(&pos));
pos.clear();
rows.push(std::mem::take(&mut cur_row));
} else if depth > 1 {
pos.push(item);
}
depth -= 1;
}
MatchedKind::Punct(',') if depth == 1 => {
cur_row.push(classify_value_position(&pos));
pos.clear();
}
_ if depth >= 1 => pos.push(item),
_ => {}
}
}
rows
}
/// Classify one `VALUES` position's matched tokens into `Some(Value)` (a
/// bare literal) or `None` (an expression). A single literal token, or a
/// sign followed by a number, is a literal; anything else is an
/// expression (ADR-0036 §1).
fn classify_value_position(tokens: &[&MatchedItem]) -> Option<Value> {
match tokens {
[one] => item_to_value(one),
[sign, num]
if matches!(sign.kind, MatchedKind::Punct('-') | MatchedKind::Punct('+'))
&& matches!(num.kind, MatchedKind::NumberLit) =>
{
let text = if matches!(sign.kind, MatchedKind::Punct('-')) {
format!("-{}", num.text)
} else {
num.text.clone()
};
Some(Value::Number(text))
}
_ => None,
}
}
/// Whether the matched path contains a `RETURNING` clause
/// (ADR-0033 §5, sub-phase 3g). Located by the `returning` *Word
/// token* in the path — path-based, so a string literal can't be
+23 -1
View File
@@ -1570,6 +1570,19 @@ fn user_value_for_column(command: &Command, column: &str) -> Option<crate::dsl::
.iter()
.find(|(c, _)| c == column)
.map(|(_, v)| v.clone()),
// ADR-0036 Phase 1: a single-row literal SQL INSERT retains its
// captured literals, so a constraint error can name the real
// value. Explicit column list only here (natural order needs the
// schema); multi-row is skipped (which row conflicted is
// ambiguous) — both degrade to the neutral "that value".
Command::SqlInsert {
listed_columns,
literal_rows,
..
} if literal_rows.len() == 1 && !listed_columns.is_empty() => {
let idx = listed_columns.iter().position(|c| c == column)?;
literal_rows[0].get(idx).cloned().flatten()
}
_ => None,
}
}
@@ -2236,8 +2249,17 @@ async fn execute_command_typed(
listed_columns,
row_source,
returning,
literal_rows,
} => database
.run_sql_insert(sql, src, target_table, listed_columns, row_source, returning)
.run_sql_insert_with_literals(
sql,
src,
target_table,
listed_columns,
row_source,
returning,
literal_rows,
)
.await
.map(CommandOutcome::Insert),
// A SQL `UPDATE` (advanced mode; ADR-0033 §2). Grammar-as-