fix: ai prevent python async + max db schema length

backend/windmill-worker/src/graphql_executor.rs

@@ -80,7 +80,9 @@ pub async fn do_graphql(
     }));
 
     if let Some(token) = &api.bearer_token {
-        request = request.bearer_auth(token.as_str());
+        if token.len() > 0 {
+            request = request.bearer_auth(token.as_str());
+        }
     }
 
     if let Some(headers) = &api.custom_headers {

frontend/src/lib/components/DBSchemaExplorer.svelte

@@ -93,7 +93,7 @@
 				</ToggleButtonGroup>
 			{/if}
 			{#if dbSchema.lang === 'graphql'}
-				<GraphqlSchemaViewer code={formatGraphqlSchema(dbSchema)} class="h-full" />
+				<GraphqlSchemaViewer code={formatGraphqlSchema(dbSchema.schema)} class="h-full" />
 			{:else}
 				<ObjectViewer json={formatSchema(dbSchema)} pureViewer collapseLevel={1} />
 			{/if}

frontend/src/lib/components/apps/components/display/dbtable/utils.ts

@@ -1,7 +1,13 @@
 import { JobService, Preview } from '$lib/gen'
 import type { DBSchema, DBSchemas, GraphqlSchema, SQLSchema } from '$lib/stores'
-import { buildClientSchema, getIntrospectionQuery, printSchema } from 'graphql'
+import {
+	buildClientSchema,
+	getIntrospectionQuery,
+	printSchema,
+	type IntrospectionQuery
+} from 'graphql'
 import { tryEvery } from '$lib/utils'
+import { stringifySchema } from '$lib/components/copilot/lib'
 
 export enum ColumnIdentity {
 	ByDefault = 'By Default',
@@ -442,21 +448,29 @@ export async function getDbSchemas(
 								const schema =
 									processingFn !== undefined ? processingFn(testResult.result) : testResult.result
 
-								dbSchemas[resourcePath] = {
+								const dbSchema = {
 									lang: resourceTypeToLang(resourceType) as SQLSchema['lang'],
 									schema,
 									publicOnly: !!schema.public || !!schema.PUBLIC || !!schema.dbo
 								}
+								dbSchemas[resourcePath] = {
+									...dbSchema,
+									stringified: stringifySchema(dbSchema)
+								}
 							} else {
 								if (typeof testResult.result !== 'object' || !('__schema' in testResult.result)) {
 									console.error('Invalid GraphQL schema')
 
 									errorCallback('Invalid GraphQL schema')
 								} else {
-									dbSchemas[resourcePath] = {
-										lang: 'graphql',
+									const dbSchema = {
+										lang: 'graphql' as GraphqlSchema['lang'],
 										schema: testResult.result
 									}
+									dbSchemas[resourcePath] = {
+										...dbSchema,
+										stringified: stringifySchema(dbSchema)
+									}
 								}
 							}
 						}
@@ -486,18 +500,20 @@ export async function getDbSchemas(
 	})
 }
 
-export function formatSchema(dbSchema: DBSchema) {
-	if (dbSchema.lang !== 'graphql' && dbSchema.publicOnly) {
+export function formatSchema(dbSchema: {
+	lang: SQLSchema['lang']
+	schema: SQLSchema['schema']
+	publicOnly: SQLSchema['publicOnly']
+}) {
+	if (dbSchema.publicOnly) {
 		return dbSchema.schema.public || dbSchema.schema.PUBLIC || dbSchema.schema.dbo || dbSchema
-	} else if (dbSchema.lang === 'mysql' && Object.keys(dbSchema.schema).length === 1) {
-		return dbSchema.schema[Object.keys(dbSchema.schema)[0]]
 	} else {
 		return dbSchema.schema
 	}
 }
 
-export function formatGraphqlSchema(dbSchema: GraphqlSchema): string {
-	return printSchema(buildClientSchema(dbSchema.schema))
+export function formatGraphqlSchema(schema: IntrospectionQuery): string {
+	return printSchema(buildClientSchema(schema))
 }
 
 export function getFieldType(type: string, databaseType: DbType) {

frontend/src/lib/components/copilot/ScriptGen.svelte

@@ -1,7 +1,7 @@
 <script lang="ts">
 	import { Button } from '../common'
 
-	import { SUPPORTED_LANGUAGES, copilot } from './lib'
+	import { MAX_SCHEMA_LENGTH, SUPPORTED_LANGUAGES, addThousandsSeparator, copilot } from './lib'
 	import type { SupportedLanguage } from '$lib/common'
 	import { sendUserToast } from '$lib/toast'
 	import type Editor from '../Editor.svelte'
@@ -19,10 +19,20 @@
 	import LoadingIcon from '../apps/svelte-select/lib/LoadingIcon.svelte'
 	import { sleep } from '$lib/utils'
 	import { autoPlacement } from '@floating-ui/core'
-	import { Ban, Bot, Check, ExternalLink, HistoryIcon, Wand2, X } from 'lucide-svelte'
+	import {
+		AlertTriangle,
+		Ban,
+		Bot,
+		Check,
+		ExternalLink,
+		HistoryIcon,
+		Wand2,
+		X
+	} from 'lucide-svelte'
 	import { fade } from 'svelte/transition'
 	import { isInitialCode } from '$lib/script_helpers'
 	import { twMerge } from 'tailwind-merge'
+	import Popover from '../Popover.svelte'
 
 	// props
 	export let iconOnly: boolean = false
@@ -406,13 +416,24 @@
 
 					{#if ['postgresql', 'mysql', 'snowflake', 'bigquery', 'mssql', 'graphql'].includes(lang) && dbSchema?.lang === lang}
 						<div class="flex flex-row items-center justify-between gap-2 w-96">
-							<div class="flex flex-row items-center">
+							<div class="flex flex-row items-center gap-1">
 								<p class="text-xs text-secondary">
 									Context: {lang === 'graphql' ? 'GraphQL' : 'DB'} schema
 								</p>
-								<Tooltip>
+								<Tooltip placement="top">
 									We pass the selected schema to GPT-4 Turbo for better script generation.
 								</Tooltip>
+								{#if dbSchema.stringified.length > MAX_SCHEMA_LENGTH}
+									<Popover notClickable placement="top">
+										<AlertTriangle size={16} class="text-yellow-500" />
+										<svelte:fragment slot="text">
+											The schema is about {addThousandsSeparator(dbSchema.stringified.length / 3.5)}
+											tokens. To avoid exceeding the model's context length, it will be truncated to
+											{addThousandsSeparator(MAX_SCHEMA_LENGTH / 3.5)}
+											tokens.
+										</svelte:fragment>
+									</Popover>
+								{/if}
 							</div>
 							{#if dbSchema.lang !== 'graphql' && (dbSchema.schema?.public || dbSchema.schema?.PUBLIC || dbSchema.schema?.dbo)}
 								<ToggleButtonGroup class="w-auto shrink-0" bind:selected={dbSchema.publicOnly}>

frontend/src/lib/components/copilot/lib.ts

@@ -2,7 +2,7 @@ import { OpenAI } from 'openai'
 import { OpenAPI, ResourceService, Script } from '../../gen'
 import type { Writable } from 'svelte/store'
 
-import type { DBSchema } from '$lib/stores'
+import type { DBSchema, GraphqlSchema, SQLSchema } from '$lib/stores'
 import { formatResourceTypes } from './utils'
 
 import { EDIT_CONFIG, FIX_CONFIG, GEN_CONFIG } from './prompts'
@@ -147,53 +147,72 @@ export async function addResourceTypes(scriptOptions: CopilotOptions, prompt: st
 	return prompt
 }
 
+export const MAX_SCHEMA_LENGTH = 100000 * 3.5
+
+export function addThousandsSeparator(n: number) {
+	return n.toFixed().replace(/\B(?=(\d{3})+(?!\d))/g, "'")
+}
+
+export function stringifySchema(
+	dbSchema: Omit<SQLSchema, 'stringified'> | Omit<GraphqlSchema, 'stringified'>
+) {
+	const { schema, lang } = dbSchema
+	if (lang === 'graphql') {
+		let graphqlSchema = printSchema(buildClientSchema(schema))
+		return graphqlSchema
+	} else {
+		let smallerSchema: {
+			[schemaKey: string]: {
+				[tableKey: string]: Array<[string, string, boolean, string?]>
+			}
+		} = {}
+		for (const schemaKey in schema) {
+			smallerSchema[schemaKey] = {}
+			for (const tableKey in schema[schemaKey]) {
+				smallerSchema[schemaKey][tableKey] = []
+				for (const colKey in schema[schemaKey][tableKey]) {
+					const col = schema[schemaKey][tableKey][colKey]
+					const p: [string, string, boolean, string?] = [colKey, col.type, col.required]
+					if (col.default) {
+						p.push(col.default)
+					}
+					smallerSchema[schemaKey][tableKey].push(p)
+				}
+			}
+		}
+
+		let finalSchema: typeof smallerSchema | (typeof smallerSchema)['schemaKey'] = smallerSchema
+		if (dbSchema.publicOnly) {
+			finalSchema =
+				smallerSchema.public || smallerSchema.PUBLIC || smallerSchema.dbo || smallerSchema
+		} else if (lang === 'mysql' && Object.keys(smallerSchema).length === 1) {
+			finalSchema = smallerSchema[Object.keys(smallerSchema)[0]]
+		}
+		return JSON.stringify(finalSchema)
+	}
+}
+
 function addDBSChema(scriptOptions: CopilotOptions, prompt: string) {
 	const { dbSchema, language } = scriptOptions
 	if (
 		dbSchema &&
 		['postgresql', 'mysql', 'snowflake', 'bigquery', 'mssql', 'graphql'].includes(language) && // make sure we are using a SQL/query language
 		language === dbSchema.lang // make sure we are using the same language as the schema
 	) {
-		const { schema, lang } = dbSchema
-		if (lang === 'graphql') {
-			const graphqlSchema = printSchema(buildClientSchema(schema))
-			prompt =
-				prompt +
-				'\nHere is the GraphQL schema: <schema>\n' +
-				JSON.stringify(graphqlSchema) +
-				'\n</schema>'
-		} else {
-			let smallerSchema: {
-				[schemaKey: string]: {
-					[tableKey: string]: Array<[string, string, boolean, string?]>
-				}
-			} = {}
-			for (const schemaKey in schema) {
-				smallerSchema[schemaKey] = {}
-				for (const tableKey in schema[schemaKey]) {
-					smallerSchema[schemaKey][tableKey] = []
-					for (const colKey in schema[schemaKey][tableKey]) {
-						const col = schema[schemaKey][tableKey][colKey]
-						const p: [string, string, boolean, string?] = [colKey, col.type, col.required]
-						if (col.default) {
-							p.push(col.default)
-						}
-						smallerSchema[schemaKey][tableKey].push(p)
-					}
-				}
+		let { stringified } = dbSchema
+		if (dbSchema.lang === 'graphql') {
+			if (stringified.length > MAX_SCHEMA_LENGTH) {
+				stringified = stringified.slice(0, MAX_SCHEMA_LENGTH) + '...'
 			}
-
-			let finalSchema: typeof smallerSchema | (typeof smallerSchema)['schemaKey'] = smallerSchema
-			if (dbSchema.publicOnly) {
-				finalSchema =
-					smallerSchema.public || smallerSchema.PUBLIC || smallerSchema.dbo || smallerSchema
-			} else if (lang === 'mysql' && Object.keys(smallerSchema).length === 1) {
-				finalSchema = smallerSchema[Object.keys(smallerSchema)[0]]
+			prompt = prompt + '\nHere is the GraphQL schema: <schema>\n' + stringified + '\n</schema>'
+		} else {
+			if (stringified.length > MAX_SCHEMA_LENGTH) {
+				stringified = stringified.slice(0, MAX_SCHEMA_LENGTH) + '...'
 			}
 			prompt =
 				prompt +
 				"\nHere's the database schema, each column is in the format [name, type, required, default?]: <dbschema>\n" +
-				JSON.stringify(finalSchema) +
+				stringified +
 				'\n</dbschema>'
 		}
 	}

frontend/src/lib/components/copilot/prompts/edit.yaml

@@ -18,6 +18,7 @@ prompts:
       ```
       <contextual_information>
       You have to write a function in python called "main". Specify the parameter types. Do not call the main function. You should generally return the result.
+      The "main" function cannot be async. If you need to use async code, you can use the asyncio library.
       You can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:
       <resourceTypes>
       {resourceTypes}

frontend/src/lib/components/copilot/prompts/editPrompt.ts

@@ -2,7 +2,7 @@ export const EDIT_PROMPT = {
   "system": "You are a helpful coding assistant for Windmill, a developer platform for running scripts. You modify code as instructed by the user. Each user message includes some contextual information which should guide your answer.\nOnly output code. Wrap the code in a code block. \nPut explanations directly in the code as comments.\nReturn the complete modified code.\n\nHere's how interactions have to look like:\nuser: {sample_question}\nassistant: ```language\n{code}\n```",
   "prompts": {
     "python3": {
-      "prompt": "Here's my python3 code: \n```python\n{code}\n```\n<contextual_information>\nYou have to write a function in python called \"main\". Specify the parameter types. Do not call the main function. You should generally return the result.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nYou need to define the type of the resources that are needed before the main function, but only include them if they are actually needed to achieve the function purpose.\nThe resource type name has to be exactly as specified (has to be IN LOEWRCASE). If the type name conflicts with any imported methods, you have to rename the imported method with the conflicting name.\n<contextual_information>\nMy instructions: {description}"
+      "prompt": "Here's my python3 code: \n```python\n{code}\n```\n<contextual_information>\nYou have to write a function in python called \"main\". Specify the parameter types. Do not call the main function. You should generally return the result.\nThe \"main\" function cannot be async. If you need to use async code, you can use the asyncio library.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nYou need to define the type of the resources that are needed before the main function, but only include them if they are actually needed to achieve the function purpose.\nThe resource type name has to be exactly as specified (has to be IN LOEWRCASE). If the type name conflicts with any imported methods, you have to rename the imported method with the conflicting name.\n<contextual_information>\nMy instructions: {description}"
     },
     "deno": {
       "prompt": "Here's my TypeScript code in a deno running environment:\n```typescript\n{code}\n```\n<contextual_information>\nWe have to export a \"main\" function like this: \"export async function main(...)\" and specify the parameter types but do not call it. You should generally return the result.\nIf needed, the standard fetch method is available globally, do not import it.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. Name the resource parameters like this: \"{resource_type}Resource\". \nThe resource type name has to be exactly as specified.\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nOnly define the type for resources that are actually needed to achieve the function purpose. If the type name conflicts with the imported object, rename the imported object NOT THE TYPE.\n</contextual_information>\nMy instructions: {description}"

frontend/src/lib/components/copilot/prompts/fix.yaml

@@ -20,6 +20,7 @@ prompts:
       ```
       <contextual_information>
       You have to write a function in python called "main". Specify the parameter types. Do not call the main function. You should generally return the result.
+      The "main" function cannot be async. If you need to use async code, you can use the asyncio library.
       You can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:
       <resourceTypes>
       {resourceTypes}

frontend/src/lib/components/copilot/prompts/fixPrompt.ts

@@ -2,7 +2,7 @@ export const FIX_PROMPT = {
   "system": "You are a helpful coding assistant for Windmill, a developer platform for running scripts. You fix the code shared by the user. Each user message includes some contextual information which should guide your answer.\nOnly output code. Wrap the code in a code block. \nExplain the error and the fix after generating the code inside an <explanation> tag.\nAlso put explanations directly in the code as comments.\nReturn the complete fixed code.\n\nHere's how interactions have to look like:\nuser: {sample_question}\nassistant: ```language\n{code}\n```\n<explanation>{explanation}</explanation>",
   "prompts": {
     "python3": {
-      "prompt": "Here's my python3 code: \n```python\n{code}\n```\n<contextual_information>\nYou have to write a function in python called \"main\". Specify the parameter types. Do not call the main function. You should generally return the result.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nYou need to define the type of the resources that are needed before the main function, but only include them if they are actually needed to achieve the function purpose.\nThe resource type name has to be exactly as specified (has to be IN LOEWRCASE). If the type name conflicts with any imported methods, you have to rename the imported method with the conflicting name.\n<contextual_information>\nI get the following error: {error}\nFix my code."
+      "prompt": "Here's my python3 code: \n```python\n{code}\n```\n<contextual_information>\nYou have to write a function in python called \"main\". Specify the parameter types. Do not call the main function. You should generally return the result.\nThe \"main\" function cannot be async. If you need to use async code, you can use the asyncio library.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nYou need to define the type of the resources that are needed before the main function, but only include them if they are actually needed to achieve the function purpose.\nThe resource type name has to be exactly as specified (has to be IN LOEWRCASE). If the type name conflicts with any imported methods, you have to rename the imported method with the conflicting name.\n<contextual_information>\nI get the following error: {error}\nFix my code."
     },
     "deno": {
       "prompt": "Here's my TypeScript code in a deno running environment:\n```typescript\n{code}\n```\n<contextual_information>\nWe have to export a \"main\" function like this: \"export async function main(...)\" and specify the parameter types but do not call it. You should generally return the result.\nIf needed, the standard fetch method is available globally, do not import it.\nYou can take as parameters resources which are dictionaries containing credentials or configuration information. Name the resource parameters like this: \"{resource_type}Resource\".\nThe following resource types are available:\n<resourceTypes>\n{resourceTypes}\n</resourceTypes>\nOnly define the type for resources that are actually needed to achieve the function purpose. The resource type name has to be exactly as specified. If the type name conflicts with the imported object, rename the imported object NOT THE TYPE.\n</contextual_information>\nI get the following error: {error}\nFix my code."

frontend/src/lib/components/copilot/prompts/gen.yaml

@@ -13,6 +13,7 @@ prompts:
     prompt: |-
       <contextual_information>
       You have to write a function in Python called "main". Specify the parameter types. Do not call the main function. You should generally return the result.
+      The "main" function cannot be async. If you need to use async code, you can use the asyncio library.
       You can take as parameters resources which are dictionaries containing credentials or configuration information. For Windmill to correctly detect the resources to be passed, the resource type name has to be exactly as specified in the following list:
       <resourceTypes>
       {resourceTypes}