razeri-plugin-api 0.1.1

// Copyright (C) 2026 Gabriel Shea
// Licensed under either of the Apache License, Version 2.0 or the MIT
// license, at your option.

//! Canonical dialect resolver with alias support and near-miss suggestions.
//!
//! All service crates (CLI, API, MCP, AI, pipeline) should call
//! [`resolve_dialect_by_name`] instead of maintaining their own match blocks.

use crate::PlatformDialect;
use crate::dialect::PlatformRegistry;
use std::sync::Arc;
use thiserror::Error;

/// Error returned when a dialect name cannot be resolved.
#[derive(Debug, Error)]
pub enum ResolveError {
    /// The name did not match any registered dialect or known alias.
    #[error(
        "unknown dialect: '{name}'. {suggestions_text}Registered: {registered}",
        suggestions_text = if suggestions.is_empty() {
            String::new()
        } else {
            format!("Did you mean: {}? ", suggestions.join(", "))
        },
        registered = registered.join(", ")
    )]
    UnknownDialect {
        /// The name that was looked up.
        name: String,
        /// Near-miss suggestions (by edit distance).
        suggestions: Vec<String>,
        /// All registered dialect names.
        registered: Vec<String>,
    },
}

/// Alias table mapping user-facing names to canonical dialect names.
///
/// The canonical name must match `DialectInfo::name` of a registered dialect.
/// Order does not matter — the first match on the left column wins.
const ALIAS_TABLE: &[(&str, &str)] = &[
    // PostgreSQL
    ("postgres", "postgres"),
    ("postgresql", "postgres"),
    ("pg", "postgres"),
    // Snowflake
    ("snowflake", "snowflake"),
    ("sf", "snowflake"),
    // SQL Server / T-SQL
    ("sqlserver", "sqlserver"),
    ("tsql", "sqlserver"),
    ("mssql", "sqlserver"),
    ("sql_server", "sqlserver"),
    // Databricks / Spark
    ("databricks", "databricks"),
    ("dbx", "databricks"),
    ("spark", "databricks"),
    // DuckDB
    ("duckdb", "duckdb"),
    ("duck", "duckdb"),
    // ClickHouse
    ("clickhouse", "clickhouse"),
    ("ch", "clickhouse"),
    // MariaDB / MySQL
    ("mariadb", "mariadb"),
    ("mariadb_columnstore", "mariadb"),
    ("mysql", "mysql"),
    // CitusDB
    ("citusdb", "citus"),
    ("citus", "citus"),
    // Agnostic (Razeri's universal cross-database dialect)
    ("agnostic", "agnostic"),
    // SQLite (extension dialect)
    ("sqlite", "sqlite"),
    // Google BigQuery
    ("bigquery", "bigquery"),
    ("big_query", "bigquery"),
    ("bq", "bigquery"),
    // Amazon Redshift
    ("redshift", "redshift"),
    ("amazon_redshift", "redshift"),
    ("rs", "redshift"),
];

/// Resolve a dialect by user-facing name string.
///
/// Performs case-insensitive matching with alias support. Falls back to the
/// [`PlatformRegistry`] for any name not in the alias table (e.g. third-party
/// extension plugins).
///
/// # Errors
///
/// Returns [`ResolveError::UnknownDialect`] with near-miss suggestions if the
/// name is not recognized.
///
/// # Examples
///
/// ```no_run
/// use razeri_plugin_api::ResolveError;
/// use razeri_plugin_api::resolve::resolve_dialect_by_name;
///
/// # fn main() -> Result<(), ResolveError> {
/// let dialect = resolve_dialect_by_name("pg")?;
/// println!("{}", dialect.info().name);
///
/// let dialect = resolve_dialect_by_name("SNOWFLAKE")?;
/// # Ok(())
/// # }
/// ```
pub fn resolve_dialect_by_name(name: &str) -> Result<Arc<dyn PlatformDialect>, ResolveError> {
    let registry = PlatformRegistry::with_auto_discovered();
    resolve_dialect_by_name_with_registry(name, &registry)
}

/// Resolve using an explicit registry (useful for testing and composition root).
///
/// # Errors
///
/// Returns an error if the operation fails.
pub fn resolve_dialect_by_name_with_registry(
    name: &str,
    registry: &PlatformRegistry,
) -> Result<Arc<dyn PlatformDialect>, ResolveError> {
    let normalized = name.trim().to_lowercase();

    // 1. Try alias table → canonical name → registry
    if let Some(canonical) = alias_lookup(&normalized)
        && let Some(dialect) = registry.get_by_name(canonical)
    {
        return Ok(Arc::clone(dialect));
    }

    // 2. Try direct registry lookup (covers third-party extensions not in alias table)
    if let Some(dialect) = registry.get_by_name(&normalized) {
        return Ok(Arc::clone(dialect));
    }

    // 3. Build error with suggestions
    let all_names = registered_names(registry);
    let suggestions = near_misses(&normalized, &all_names);
    Err(ResolveError::UnknownDialect { name: name.to_string(), suggestions, registered: all_names })
}

/// **The accepted spellings, as `(spelling, canonical name)` pairs.**
///
/// A design names its platform in whatever spelling its author reaches for, and
/// this is the table that turns one into the other. A platform whose own name
/// is missing from it is refused at the CLI door however well its dialect
/// renders.
#[must_use]
pub const fn alias_spellings() -> &'static [(&'static str, &'static str)] {
    ALIAS_TABLE
}

/// List all names accepted by the resolver (alias table + registered names).
#[must_use]
pub fn all_accepted_names() -> Vec<String> {
    let registry = PlatformRegistry::with_auto_discovered();
    let mut names: Vec<String> =
        ALIAS_TABLE.iter().map(|(alias, _)| (*alias).to_string()).collect();
    for info in registry.list() {
        let n = info.name.to_string();
        if !names.contains(&n) {
            names.push(n);
        }
    }
    names.sort();
    names.dedup();
    names
}

// ── Internal helpers ────────────────────────────────────────────────────

/// Look up an alias (already lowercased) in the alias table.
fn alias_lookup(normalized: &str) -> Option<&'static str> {
    ALIAS_TABLE.iter().find(|(alias, _)| *alias == normalized).map(|(_, canonical)| *canonical)
}

/// Collect all registered dialect names from the registry.
fn registered_names(registry: &PlatformRegistry) -> Vec<String> {
    let mut names: Vec<String> = registry.list().iter().map(|info| info.name.to_string()).collect();
    names.sort();
    names
}

/// Find near-miss suggestions using Levenshtein distance ≤ 3.
fn near_misses(input: &str, candidates: &[String]) -> Vec<String> {
    let mut scored: Vec<(&str, usize)> = candidates
        .iter()
        .filter_map(|c| {
            let dist = levenshtein(input, c);
            if dist <= 3 { Some((c.as_str(), dist)) } else { None }
        })
        .collect();
    scored.sort_by_key(|(_, d)| *d);
    scored.into_iter().map(|(name, _)| name.to_string()).collect()
}

/// Simple Levenshtein distance (no allocation beyond the DP row).
fn levenshtein(a: &str, b: &str) -> usize {
    let a_bytes = a.as_bytes();
    let b_bytes = b.as_bytes();
    let m = a_bytes.len();
    let n = b_bytes.len();

    let mut prev: Vec<usize> = (0..=n).collect();
    let mut curr = vec![0; n + 1];

    for i in 1..=m {
        curr[0] = i;
        for j in 1..=n {
            let cost = usize::from(a_bytes[i - 1] != b_bytes[j - 1]);
            curr[j] = (prev[j] + 1).min(curr[j - 1] + 1).min(prev[j - 1] + cost);
        }
        std::mem::swap(&mut prev, &mut curr);
    }
    prev[n]
}

// ── Also include alias lookup in the ALIAS_TABLE entries ────────────────
/// All known aliases (for documentation and help messages).
#[must_use]
pub fn all_aliases() -> &'static [(&'static str, &'static str)] {
    ALIAS_TABLE
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::dialect::{DialectInfo, PlatformDialect, PlatformRegistry};
    use razeri_plugin_contract::DataType;
    use razeri_plugin_contract::DialectFamily;
    use razeri_plugin_contract::{
        DdlRenderer, DialectCapabilities, DmlRenderer, ExpressionRenderer, FormulaRenderer,
        ProcedureRenderer, ReplicationRenderer, SqlRenderer, StorageRenderer, TaskRenderer,
    };

    /// Minimal test dialect.
    struct StubDialect {
        info: DialectInfo,
    }

    impl SqlRenderer for StubDialect {
        fn render_data_type(&self, _dt: &DataType) -> String {
            String::new()
        }
        fn quote_identifier(&self, name: &str) -> String {
            format!("\"{name}\"")
        }
        fn qualify_name(&self, _schema: Option<&str>, name: &str) -> String {
            self.quote_identifier(name)
        }
        fn render_current_timestamp(&self) -> String {
            String::new()
        }
    }
    impl ExpressionRenderer for StubDialect {}
    impl DdlRenderer for StubDialect {}
    impl DmlRenderer for StubDialect {}
    impl ProcedureRenderer for StubDialect {}
    impl FormulaRenderer for StubDialect {}
    impl TaskRenderer for StubDialect {}
    impl DialectCapabilities for StubDialect {}
    impl ReplicationRenderer for StubDialect {}
    impl StorageRenderer for StubDialect {}
    impl PlatformDialect for StubDialect {
        fn info(&self) -> DialectInfo {
            self.info.clone()
        }
    }

    fn test_registry() -> PlatformRegistry {
        let mut reg = PlatformRegistry::new();
        let entries: &[(&str, &str, DialectFamily, &str, bool, bool, bool)] = &[
            ("postgres", "PostgreSQL", DialectFamily::PostgreSql, "PostgreSQL", false, false, true),
            ("snowflake", "Snowflake", DialectFamily::Snowflake, "Snowflake", true, true, false),
            ("sqlserver", "SQL Server", DialectFamily::TSql, "T-SQL", false, false, false),
            ("databricks", "Databricks", DialectFamily::SparkSql, "Spark SQL", true, true, false),
            (
                "clickhouse",
                "ClickHouse",
                DialectFamily::ClickHouse,
                "ClickHouse SQL",
                false,
                true,
                false,
            ),
            ("duckdb", "DuckDB", DialectFamily::PostgreSql, "PostgreSQL", false, true, true),
            ("citus", "Citus", DialectFamily::PostgreSql, "PostgreSQL", false, false, true),
            ("mariadb", "MariaDB", DialectFamily::MariaDB, "MariaDB", false, true, false),
            ("mysql", "MySQL", DialectFamily::MariaDB, "MariaDB", false, false, false),
            ("agnostic", "Agnostic", DialectFamily::Razeri, "ANSI", false, false, false),
            ("sqlite", "SQLite", DialectFamily::MariaDB, "SQLite", false, false, false),
        ];
        for &(name, display, family, fam_str, cloud, olap, pg_compat) in entries {
            reg.register(
                name.to_string(),
                Arc::new(StubDialect {
                    info: DialectInfo {
                        name,
                        display_name: display,
                        dialect_family: family,
                        family: fam_str,
                        is_cloud_native: cloud,
                        is_olap: olap,
                        is_postgres_compatible: pg_compat,
                    },
                }),
            );
        }
        reg
    }

    #[test]
    fn alias_lookup_finds_canonical_names() {
        assert_eq!(alias_lookup("pg"), Some("postgres"));
        assert_eq!(alias_lookup("postgresql"), Some("postgres"));
        assert_eq!(alias_lookup("tsql"), Some("sqlserver"));
        assert_eq!(alias_lookup("mssql"), Some("sqlserver"));
        assert_eq!(alias_lookup("sf"), Some("snowflake"));
        assert_eq!(alias_lookup("ch"), Some("clickhouse"));
        assert_eq!(alias_lookup("duck"), Some("duckdb"));
        assert_eq!(alias_lookup("dbx"), Some("databricks"));
        assert_eq!(alias_lookup("spark"), Some("databricks"));
        assert_eq!(alias_lookup("citusdb"), Some("citus"));
        assert_eq!(alias_lookup("sqlite"), Some("sqlite"));
        assert_eq!(alias_lookup("agnostic"), Some("agnostic"));
    }

    #[test]
    fn alias_lookup_with_unknown_returns_none() {
        assert_eq!(alias_lookup("nosuchdb"), None);
    }

    #[test]
    fn levenshtein_distance_basic() {
        assert_eq!(levenshtein("postgres", "postgres"), 0);
        assert_eq!(levenshtein("postgre", "postgres"), 1);
        assert_eq!(levenshtein("postgrse", "postgres"), 2);
        assert_eq!(levenshtein("", "abc"), 3);
    }

    #[test]
    fn resolve_with_canonical_name_succeeds() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("postgres", &reg).is_ok());
    }

    #[test]
    fn resolve_with_alias_succeeds() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("pg", &reg).is_ok());
    }

    #[test]
    fn resolve_is_case_insensitive() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("SNOWFLAKE", &reg).is_ok());
    }

    #[test]
    fn resolve_sqlite_extension_succeeds() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("sqlite", &reg).is_ok());
    }

    #[test]
    fn resolve_databricks_succeeds() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("databricks", &reg).is_ok());
    }

    #[test]
    fn resolve_mssql_alias_succeeds() {
        let reg = test_registry();
        assert!(resolve_dialect_by_name_with_registry("mssql", &reg).is_ok());
    }

    #[test]
    fn resolve_unknown_returns_error_with_suggestions() {
        let reg = test_registry();
        let result = resolve_dialect_by_name_with_registry("postres", &reg);
        let Err(err) = result else { panic!("expected error") };
        let msg = err.to_string();
        assert!(msg.contains("unknown dialect"), "error should mention unknown dialect: {msg}");
        assert!(msg.contains("postgres"), "error should suggest postgres: {msg}");
    }

    #[test]
    fn resolve_completely_unknown_returns_registered_list() {
        let reg = test_registry();
        let result = resolve_dialect_by_name_with_registry("zzzznotadb", &reg);
        let Err(err) = result else { panic!("expected error") };
        let msg = err.to_string();
        assert!(msg.contains("Registered:"), "error should list registered: {msg}");
    }

    #[test]
    fn all_accepted_names_includes_aliases_and_registered() {
        let names = all_accepted_names();
        assert!(names.contains(&"pg".to_string()));
        assert!(names.contains(&"postgres".to_string()));
        assert!(names.contains(&"snowflake".to_string()));
    }

    /// The published design schema's `DatabasePlatform` enum must speak the
    /// SAME vocabulary the resolver accepts.
    ///
    /// A design's `design.platforms` / `design.default_platform` strings are
    /// resolved — everywhere they are resolved at all — through
    /// [`resolve_dialect_by_name`], whose left-hand [`ALIAS_TABLE`] column is
    /// the operative accepted set. When the schema's enum is a *subset* of it,
    /// designs that parse and render cleanly fail their own published schema;
    /// when it is a *superset*, the schema advertises a spelling the resolver
    /// refuses. Set equality is the only honest relation, so this test pins it
    /// in the crate that owns the table — the schema cannot drift away from
    /// the resolver, and the resolver cannot grow a spelling the schema never
    /// learned.
    ///
    /// It is *set* equality, not ordered equality, because [`ALIAS_TABLE`]'s
    /// own contract says its order does not matter — an alias moved between
    /// two families is not a drift, and a test that failed on it would be
    /// lying.
    #[test]
    fn design_schema_platform_enum_matches_the_alias_table() {
        let schema_path = std::path::Path::new(env!("CARGO_MANIFEST_DIR"))
            .join("../../schemas/razeri-design.schema.json");
        let text = std::fs::read_to_string(&schema_path)
            .unwrap_or_else(|e| panic!("read {}: {e}", schema_path.display()));
        let schema: serde_json::Value =
            serde_json::from_str(&text).unwrap_or_else(|e| panic!("parse the design schema: {e}"));

        let enum_values = schema["$defs"]["DatabasePlatform"]["enum"]
            .as_array()
            .expect("$defs.DatabasePlatform.enum is an array in the published design schema");
        let mut published: Vec<String> = enum_values
            .iter()
            .map(|v| v.as_str().expect("every DatabasePlatform enum value is a string").to_string())
            .collect();
        published.sort();
        published.dedup();

        let mut accepted: Vec<String> =
            ALIAS_TABLE.iter().map(|(alias, _)| (*alias).to_string()).collect();
        accepted.sort();
        accepted.dedup();

        let advertised_but_refused: Vec<&String> =
            published.iter().filter(|n| !accepted.contains(n)).collect();
        let accepted_but_unpublished: Vec<&String> =
            accepted.iter().filter(|n| !published.contains(n)).collect();

        assert!(
            advertised_but_refused.is_empty() && accepted_but_unpublished.is_empty(),
            "schemas/razeri-design.schema.json $defs.DatabasePlatform has drifted from the \
             canonical resolver's ALIAS_TABLE.\n  advertised by the schema but REFUSED by the \
             resolver: {advertised_but_refused:?}\n  accepted by the resolver but MISSING from \
             the schema: {accepted_but_unpublished:?}\nThe alias table is the authority: add the \
             spelling to the schema's enum, or take it out of the alias table — do not let the \
             two vocabularies differ."
        );
    }
}