WIP(schema-wasm): support schema split into multiple files (#4787)

* Implement multi-file schema handling in PSL

This commit implements multi-file schema handling in the Prisma Schema Language.

At a high level, instead of accepting a single string, `psl::validate_multi_file()` is an alternative to `psl::validate()` that accepts something morally equivalent to:

```json
{
  "./prisma/schema/a.prisma": "datasource db { ... }",
  "./prisma/schema/nested/b.prisma": "model Test { ... }"
}
```

There are tests for PSL validation with multiple schema files, but most of the rest of engines still consumes the single file version of `psl::validate()`. The implementation and the return type are shared between `psl::validate_multi_file()` and `psl::validate()`, so the change is completely transparent, other than the expectation of passing in a list of (file_name, file_contents) instead of a single string. The `psl::validate()` entry point should behave exactly the same as `psl::multi_schema()` with a single file named `schema.prisma`. In particular, it has the exact same return type.

Implementation
==============

This is achieved by extending `Span` to contain, in addition to a start and end offset, a `FileId`. The `FileId` is a unique identifier for a file and its parsed `SchemaAst` inside `ParserDatabase`. The identifier types for AST items in `ParserDatabase` are also extended to contain the `FileId`, so that they can be uniquely referred to in the context of the (multi-file) schema. After the analysis phase (the `parser_database` crate), consumers of the analyzed schema become multi-file aware completely transparently, no change is necessary in the other engines.

The only changes that will be required at scattered points across the codebase are the `psl::validate()` call sites that will need to receive a `Vec<Box<Path>, SourceFile>` instead of a single `SourceFile`. This PR does _not_ deal with that, but it makes where these call sites are obvious by what entry points they use: `psl::validate()`, `psl::parse_schema()` and the various `*_assert_single()` methods on `ParserDatabase`.

The PR contains tests confirming that schema analysis, validation and displaying diagnostics across multiple files works as expected.

Status of this PR
=================

This is going to be directly mergeable after review, and it will not affect the current schema handling behaviour when dealing with a single schema file.

Next steps
==========

- Replace all calls to `psl::validate()` with calls to `psl::validate_multi_file()`.
- The `*_assert_single()` calls should be progressively replaced with their multi-file counterparts across engines.
- The language server should start sending multiple files to prisma-schema-wasm in all calls. This is not in the spirit of the language server spec, but that is the most immediate solution. We'll have to make `range_to_span()` in `prisma-fmt` multi-schema aware by taking a FileId param.

Links
=====

Relevant issue: prisma/prisma#2377

Also see the [internal design doc](https://www.notion.so/prismaio/Multi-file-Schema-24d68fe8664048ad86252fe446caac24?d=68ef128f25974e619671a9855f65f44d#2889a038e68c4fe1ac9afe3cd34978bd).

* WIP(schema-wasm): Support schema split into multiple files

* Reformat support (psl crate)

* Add multifile reformatting tests

* Clippy

* feat(prisma-fmt): addd support for mergeSchemas, expose functions to prisma-fmt-wasm

* chore(prisma-fmt): removed unused function

* chore: fix typo

Co-authored-by: Serhii Tatarintsev <tatarintsev@prisma.io>

* feat(prisma-fmt): apply validation to merge_schemas

* chore(prisma-fmt): update unit test

* chore: fix bad merge

* chore: fix tests

---------

Co-authored-by: Tom Houlé <tom@tomhoule.com>
Co-authored-by: Alberto Schiabel <jkomyno@users.noreply.github.com>
Co-authored-by: jkomyno <skiabo97@gmail.com>

prisma-fmt/src/code_actions/multi_schema.rs

@@ -147,8 +147,7 @@ pub(super) fn add_schema_to_schemas(
             )
         }
         None => {
-            let has_properties = datasource.provider_defined()
-                || datasource.url_defined()
+            let has_properties = datasource.provider_defined() | datasource.url_defined()
                 || datasource.direct_url_defined()
                 || datasource.shadow_url_defined()
                 || datasource.relation_mode_defined()

prisma-fmt/src/get_config.rs

@@ -1,14 +1,14 @@
-use psl::Diagnostics;
+use psl::{Diagnostics, ValidatedSchema};
 use serde::Deserialize;
 use serde_json::json;
 use std::collections::HashMap;
 
-use crate::validate::SCHEMA_PARSER_ERROR_CODE;
+use crate::{schema_file_input::SchemaFileInput, validate::SCHEMA_PARSER_ERROR_CODE};
 
 #[derive(Deserialize, Debug)]
 #[serde(rename_all = "camelCase")]
 struct GetConfigParams {
-    prisma_schema: String,
+    prisma_schema: SchemaFileInput,
     #[serde(default)]
     ignore_env_var_errors: bool,
     #[serde(default)]
@@ -43,29 +43,38 @@ pub(crate) fn get_config(params: &str) -> Result<String, String> {
 }
 
 fn get_config_impl(params: GetConfigParams) -> Result<serde_json::Value, GetConfigError> {
-    let wrap_get_config_err = |errors: Diagnostics| -> GetConfigError {
-        use std::fmt::Write as _;
-
-        let mut full_error = errors.to_pretty_string("schema.prisma", &params.prisma_schema);
-        write!(full_error, "\nValidation Error Count: {}", errors.errors().len()).unwrap();
-
-        GetConfigError {
-            // this mirrors user_facing_errors::common::SchemaParserError
-            error_code: Some(SCHEMA_PARSER_ERROR_CODE),
-            message: full_error,
-        }
-    };
-
-    let mut config = psl::parse_configuration(&params.prisma_schema).map_err(wrap_get_config_err)?;
+    let mut schema = psl::validate_multi_file(params.prisma_schema.into());
+    if schema.diagnostics.has_errors() {
+        return Err(create_get_config_error(&schema, &schema.diagnostics));
+    }
 
     if !params.ignore_env_var_errors {
         let overrides: Vec<(_, _)> = params.datasource_overrides.into_iter().collect();
-        config
+        schema
+            .configuration
             .resolve_datasource_urls_prisma_fmt(&overrides, |key| params.env.get(key).map(String::from))
-            .map_err(wrap_get_config_err)?;
+            .map_err(|diagnostics| create_get_config_error(&schema, &diagnostics))?;
     }
 
-    Ok(psl::get_config(&config))
+    Ok(psl::get_config(&schema.configuration))
+}
+
+fn create_get_config_error(schema: &ValidatedSchema, diagnostics: &Diagnostics) -> GetConfigError {
+    use std::fmt::Write as _;
+
+    let mut rendered_diagnostics = schema.render_diagnostics(diagnostics);
+    write!(
+        rendered_diagnostics,
+        "\nValidation Error Count: {}",
+        diagnostics.errors().len()
+    )
+    .unwrap();
+
+    GetConfigError {
+        // this mirrors user_facing_errors::common::SchemaParserError
+        error_code: Some(SCHEMA_PARSER_ERROR_CODE),
+        message: rendered_diagnostics,
+    }
 }
 
 #[cfg(test)]

prisma-fmt/src/lib.rs

@@ -3,8 +3,10 @@ mod code_actions;
 mod get_config;
 mod get_dmmf;
 mod lint;
+mod merge_schemas;
 mod native;
 mod preview;
+mod schema_file_input;
 mod text_document_completion;
 mod validate;
 
@@ -89,6 +91,14 @@ pub fn validate(validate_params: String) -> Result<(), String> {
     validate::validate(&validate_params)
 }
 
+/// Given a list of Prisma schema files (and their locations), returns the merged schema.
+/// This is useful for `@prisma/client` generation, where the client needs a single - potentially large - schema,
+/// while still allowing the user to split their schema copies into multiple files.
+/// Internally, it uses `[validate]`.
+pub fn merge_schemas(params: String) -> Result<String, String> {
+    merge_schemas::merge_schemas(&params)
+}
+
 pub fn native_types(schema: String) -> String {
     native::run(&schema)
 }

prisma-fmt/src/merge_schemas.rs

@@ -0,0 +1,127 @@
+use psl::reformat_validated_schema_into_single;
+use serde::Deserialize;
+
+use crate::schema_file_input::SchemaFileInput;
+
+#[derive(Debug, Deserialize)]
+pub struct MergeSchemasParams {
+    schema: SchemaFileInput,
+}
+
+pub(crate) fn merge_schemas(params: &str) -> Result<String, String> {
+    let params: MergeSchemasParams = match serde_json::from_str(params) {
+        Ok(params) => params,
+        Err(serde_err) => {
+            panic!("Failed to deserialize MergeSchemasParams: {serde_err}");
+        }
+    };
+
+    let validated_schema = crate::validate::run(params.schema, false)?;
+
+    let indent_width = 2usize;
+    let merged_schema = reformat_validated_schema_into_single(validated_schema, indent_width).unwrap();
+
+    Ok(merged_schema)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use expect_test::expect;
+    use serde_json::json;
+
+    #[test]
+    fn merge_two_valid_schemas_succeeds() {
+        let schema = vec![
+            (
+                "b.prisma",
+                r#"
+                model B {
+                    id String @id
+                    a A?
+                }
+            "#,
+            ),
+            (
+                "a.prisma",
+                r#"
+                datasource db {
+                    provider = "postgresql"
+                    url = env("DBURL")
+                }
+
+                model A {
+                    id String @id
+                    b_id String @unique
+                    b B @relation(fields: [b_id], references: [id])
+                }
+            "#,
+            ),
+        ];
+
+        let request = json!({
+            "schema": schema,
+        });
+
+        let expected = expect![[r#"
+            model B {
+              id String @id
+              a  A?
+            }
+
+            datasource db {
+              provider = "postgresql"
+              url      = env("DBURL")
+            }
+
+            model A {
+              id   String @id
+              b_id String @unique
+              b    B      @relation(fields: [b_id], references: [id])
+            }
+        "#]];
+
+        let response = merge_schemas(&request.to_string()).unwrap();
+        expected.assert_eq(&response);
+    }
+
+    #[test]
+    fn merge_two_invalid_schemas_panics() {
+        let schema = vec![
+            (
+                "b.prisma",
+                r#"
+                model B {
+                    id String @id
+                    a A?
+                }
+            "#,
+            ),
+            (
+                "a.prisma",
+                r#"
+                datasource db {
+                    provider = "postgresql"
+                    url = env("DBURL")
+                }
+
+                model A {
+                    id String @id
+                    b_id String @unique
+                }
+            "#,
+            ),
+        ];
+
+        let request = json!({
+            "schema": schema,
+        });
+
+        let expected = expect![[
+            r#"{"error_code":"P1012","message":"\u001b[1;91merror\u001b[0m: \u001b[1mError validating field `a` in model `B`: The relation field `a` on model `B` is missing an opposite relation field on the model `A`. Either run `prisma format` or add it manually.\u001b[0m\n  \u001b[1;94m-->\u001b[0m  \u001b[4mb.prisma:4\u001b[0m\n\u001b[1;94m   | \u001b[0m\n\u001b[1;94m 3 | \u001b[0m                    id String @id\n\u001b[1;94m 4 | \u001b[0m                    \u001b[1;91ma A?\u001b[0m\n\u001b[1;94m 5 | \u001b[0m                }\n\u001b[1;94m   | \u001b[0m\n\nValidation Error Count: 1"}"#
+        ]];
+
+        let response = merge_schemas(&request.to_string()).unwrap_err();
+        expected.assert_eq(&response);
+    }
+}

prisma-fmt/src/schema_file_input.rs

@@ -0,0 +1,26 @@
+use psl::SourceFile;
+use serde::Deserialize;
+
+/// Struct for supporting multiple files
+/// in a backward-compatible way: can either accept
+/// a single file contents or vector of (filePath, content) tuples.
+/// Can be converted to the input for `psl::validate_multi_file` from
+/// any of the variants.
+#[derive(Deserialize, Debug)]
+#[serde(untagged)]
+pub(crate) enum SchemaFileInput {
+    Single(String),
+    Multiple(Vec<(String, String)>),
+}
+
+impl From<SchemaFileInput> for Vec<(String, SourceFile)> {
+    fn from(value: SchemaFileInput) -> Self {
+        match value {
+            SchemaFileInput::Single(content) => vec![("schema.prisma".to_owned(), content.into())],
+            SchemaFileInput::Multiple(file_list) => file_list
+                .into_iter()
+                .map(|(filename, content)| (filename, content.into()))
+                .collect(),
+        }
+    }
+}

prisma-fmt/src/validate.rs

@@ -1,14 +1,17 @@
+use psl::ValidatedSchema;
 use serde::Deserialize;
 use serde_json::json;
 use std::fmt::Write as _;
 
+use crate::schema_file_input::SchemaFileInput;
+
 // this mirrors user_facing_errors::common::SchemaParserError
 pub(crate) static SCHEMA_PARSER_ERROR_CODE: &str = "P1012";
 
 #[derive(Deserialize, Debug)]
 #[serde(rename_all = "camelCase")]
 struct ValidateParams {
-    prisma_schema: String,
+    prisma_schema: SchemaFileInput,
     #[serde(default)]
     no_color: bool,
 }
@@ -21,21 +24,22 @@ pub(crate) fn validate(params: &str) -> Result<(), String> {
         }
     };
 
-    run(&params.prisma_schema, params.no_color)
+    run(params.prisma_schema, params.no_color)?;
+    Ok(())
 }
 
-pub fn run(input_schema: &str, no_color: bool) -> Result<(), String> {
-    let validate_schema = psl::validate(input_schema.into());
+pub fn run(input_schema: SchemaFileInput, no_color: bool) -> Result<ValidatedSchema, String> {
+    let validate_schema = psl::validate_multi_file(input_schema.into());
     let diagnostics = &validate_schema.diagnostics;
 
     if !diagnostics.has_errors() {
-        return Ok(());
+        return Ok(validate_schema);
     }
 
     // always colorise output regardless of the environment, which is important for Wasm
     colored::control::set_override(!no_color);
 
-    let mut formatted_error = diagnostics.to_pretty_string("schema.prisma", input_schema);
+    let mut formatted_error = validate_schema.render_own_diagnostics();
     write!(
         formatted_error,
         "\nValidation Error Count: {}",
@@ -109,6 +113,83 @@ mod tests {
         validate(&request.to_string()).unwrap();
     }
 
+    #[test]
+    fn validate_multiple_files() {
+        let schema = vec![
+            (
+                "a.prisma",
+                r#"
+                datasource thedb {
+                    provider = "postgresql"
+                    url = env("DBURL")
+                }
+
+                model A {
+                    id String @id
+                    b_id String @unique
+                    b B @relation(fields: [b_id], references: [id])
+                }
+            "#,
+            ),
+            (
+                "b.prisma",
+                r#"
+                model B {
+                    id String @id
+                    a A?
+                }
+            "#,
+            ),
+        ];
+
+        let request = json!({
+            "prismaSchema": schema,
+        });
+
+        validate(&request.to_string()).unwrap();
+    }
+
+    #[test]
+    fn validate_multiple_files_error() {
+        let schema = vec![
+            (
+                "a.prisma",
+                r#"
+                datasource thedb {
+                    provider = "postgresql"
+                    url = env("DBURL")
+                }
+
+                model A {
+                    id String @id
+                    b_id String @unique
+                    b B @relation(fields: [b_id], references: [id])
+                }
+            "#,
+            ),
+            (
+                "b.prisma",
+                r#"
+                model B {
+                    id String @id
+                    a A
+                }
+            "#,
+            ),
+        ];
+
+        let request = json!({
+            "prismaSchema": schema,
+        });
+
+        let expected = expect![[
+            r#"{"error_code":"P1012","message":"\u001b[1;91merror\u001b[0m: \u001b[1mError parsing attribute \"@relation\": The relation field `a` on Model `B` is required. This is no longer valid because it's not possible to enforce this constraint on the database level. Please change the field type from `A` to `A?` to fix this.\u001b[0m\n  \u001b[1;94m-->\u001b[0m  \u001b[4mb.prisma:4\u001b[0m\n\u001b[1;94m   | \u001b[0m\n\u001b[1;94m 3 | \u001b[0m                    id String @id\n\u001b[1;94m 4 | \u001b[0m                    \u001b[1;91ma A\u001b[0m\n\u001b[1;94m 5 | \u001b[0m                }\n\u001b[1;94m   | \u001b[0m\n\nValidation Error Count: 1"}"#
+        ]];
+
+        let response = validate(&request.to_string()).unwrap_err();
+        expected.assert_eq(&response);
+    }
+
     #[test]
     fn validate_using_both_relation_mode_and_referential_integrity() {
         let schema = r#"