The all-in-one local AI studio for your desktop: chat, image and video generation and a coding agent in one free, open source app. Windows and Linux. No Docker, no terminal, no cloud required.
TypeScript agentai-chatai-studioall-in-onecoding-agentdesktop-appfluximage-generationlinuxllmlocal-ailocal-llmopen-sourceprivacyself-hostedstable-diffusionuncensoreduncensored-aivideo-generationwindows
/**
* A14 third review: the module that decides whether the backend ANSWERED did
* not recognise the one refusal the app actually produces.
*
* `backendCall` outside the desktop build routes through the HTTP bridge's
* endpoint table and throws `Unknown backend command: <name>` for anything not
* in it. None of the fragments matched that, so every web and remote-bridge
* refresh treated the refusal as "no answer" and asked the same dead question
* again, forever. The two tests that were meant to cover this used invented
* wordings and passed on strings nothing ever throws.
*
* So the literal from src/api/backend.ts is pinned here, from the source file
* itself, and not retyped from memory.
*
* Run: npx vitest run src/lib/__tests__/engine-command-availability.test.ts
*/
import { describe, it, expect } from 'vitest'
import { readFileSync } from 'node:fs'
import { resolve, dirname } from 'node:path'
import { fileURLToPath } from 'node:url'
import { commandIsUnavailable, commandFailureText } from '../engine-command-availability'
const backendSrc = readFileSync(
resolve(dirname(fileURLToPath(import.meta.url)), '../../api/backend.ts'),
'utf8',
)
describe('the refusal the HTTP bridge really throws', () => {
it('is still worded the way this module expects', () => {
// The source guard. If someone rewords the throw, this fails here rather
// than in a web build six weeks later.
expect(backendSrc).toContain('Unknown backend command: ${command}')
})
it('counts as an answer', () => {
expect(commandIsUnavailable(new Error('Unknown backend command: list_bundled_models'))).toBe(true)
})
it('counts as an answer as a bare string too', () => {
// Tauri hands rejections across as strings, not as Errors.
expect(commandIsUnavailable('Unknown backend command: swap_bundled_model')).toBe(true)
expect(commandFailureText('Unknown backend command: swap_bundled_model'))
.toContain('Unknown backend command')
})
// NEGATIVE CONTROL: everything unrecognised must stay "no answer", because
// the cost of retrying is one extra command and the cost of not retrying is
// a dead engine for the rest of the session.
it('and a call that never got through still counts as nothing learned', () => {
for (const msg of [
'invoke timed out after 5000ms',
'Failed to fetch',
'NetworkError when attempting to fetch resource',
'The backend command took too long',
]) {
expect(commandIsUnavailable(new Error(msg)), msg).toBe(false)
}
})
// NEGATIVE CONTROL: the older refusals are still recognised.
it('keeps recognising the wordings it already knew', () => {
for (const msg of [
'command not_a_command not found',
'Unknown command foo',
'not running in Tauri',
'not implemented',
]) {
expect(commandIsUnavailable(new Error(msg)), msg).toBe(true)
}
})
})