Skip to content

Commit

Permalink
AI: Generate PageObject, added types, shell improvement (#4319)
Browse files Browse the repository at this point in the history
* added types for ai

* fix: types complaint

* added gen page objects, improved types, improved shell

* fixed tests

---------

Co-authored-by: kobenguyent <[email protected]>
  • Loading branch information
DavertMik and kobenguyent authored Jun 19, 2024
1 parent d8f45eb commit e48051c
Show file tree
Hide file tree
Showing 7 changed files with 311 additions and 37 deletions.
107 changes: 103 additions & 4 deletions docs/ai.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@ So, instead of asking "write me a test" it can ask "write a test for **this** pa
CodeceptJS AI can do the following:

* 🏋️‍♀️ **assist writing tests** in `pause()` or interactive shell mode
* 📃 **generate page objects** in `pause()` or interactive shell mode
* 🚑 **self-heal failing tests** (can be used on CI)
* 💬 send arbitrary prompts to AI provider from any tested page attaching its HTML contents

Expand Down Expand Up @@ -260,15 +261,29 @@ By evaluating this information you will be able to check how effective AI can be
### Arbitrary GPT Prompts
What if you want to take ChatGPT on the journey of test automation and ask it questions while browsing pages?
What if you want to take AI on the journey of test automation and ask it questions while browsing pages?
This is possible with the new `AI` helper. Enable it in your config and it will automatically attach to Playwright, WebDriver, or another web helper you use. It includes the following methods:
This is possible with the new `AI` helper. Enable it in your config file in `helpers` section:
```js
// inside codecept.conf
helpers: {
// Playwright, Puppeteer, or WebDrver helper should be enabled too
Playwright: {
},

AI: {}
}
```
AI helper will be automatically attached to Playwright, WebDriver, or another web helper you use. It includes the following methods:
* `askGptOnPage` - sends GPT prompt attaching the HTML of the page. Large pages will be split into chunks, according to `chunkSize` config. You will receive responses for all chunks.
* `askGptOnPageFragment` - sends GPT prompt attaching the HTML of the specific element. This method is recommended over `askGptOnPage` as you can reduce the amount of data to be processed.
* `askGptGeneralPrompt` - sends GPT prompt without HTML.
* `askForPageObject` - creates PageObject for you, explained in next section.
OpenAI helper won't remove non-interactive elements, so it is recommended to manually control the size of the sent HTML.
`askGpt` methods won't remove non-interactive elements, so it is recommended to manually control the size of the sent HTML.
Here are some good use cases for this helper:
Expand All @@ -282,7 +297,84 @@ Here are some good use cases for this helper:
const pageDoc = await I.askGptOnPageFragment('Act as technical writer, describe what is this page for', '#container');
```
As of now, those use cases do not apply to test automation but maybe you can apply them to your testing setup.
As of now, those use cases do not apply to test automation but maybe you can apply them to your testing setup.
## Generate PageObjects
Last but not the least. AI helper can be used to quickly prototype PageObjects on pages browsed within interactive session.
![](/img/ai_page_object.png)
Enable AI helper as explained in previous section and launch shell:
```
npx codeceptjs shell --ai
```
Also this is availble from `pause()` if AI helper is enabled,
Ensure that browser is started in window mode, then browse the web pages on your site.
On a page you want to create PageObject execute `askForPageObject()` command. The only required parameter is the name of a page:
```js
I.askForPageObject('login')
```
This command sends request to AI provider should create valid CodeceptJS PageObject.
Run it few times or switch AI provider if response is not satisfactory to you.
> You can change the style of PageObject and locator preferences by adjusting prompt in a config file
When completed successfully, page object is saved to **output** directory and loaded into the shell as `page` variable so locators and methods can be checked on the fly.
If page object has `signInButton` locator you can quickly check it by typing:
```js
I.click(page.signInButton)
```
If page object has `clickForgotPassword` method you can execute it as:
```js
=> page.clickForgotPassword()
```
```shell
Page object for login is saved to .../output/loginPage-1718579784751.js
Page object registered for this session as `page` variable
Use `=>page.methodName()` in shell to run methods of page object
Use `click(page.locatorName)` to check locators of page object

I.=>page.clickSignUp()
I.click(page.signUpLink)
I.=> page.enterPassword('asdasd')
I.=> page.clickSignIn()
```
You can improve prompt by passing custom request as a second parameter:
```js
I.askForPageObject('login', 'implement signIn(username, password) method')
```
To generate page object for the part of a page, pass in root locator as third parameter.
```js
I.askForPageObject('login', '', '#auth')
```
In this case, all generated locators, will use `#auth` as their root element.
Don't aim for perfect PageObjects but find a good enough one, which you can use for writing your tests.
All created page objects are considered temporary, that's why saved to `output` directory.
Rename created PageObject to remove timestamp and move it from `output` to `pages` folder and include it into codecept.conf file:
```js
include: {
loginPage: "./pages/loginPage.js",
// ...
```
## Advanced Configuration
Expand Down Expand Up @@ -315,6 +407,7 @@ ai: {
prompts: {
writeStep: (html, input) => [{ role: 'user', content: 'As a test engineer...' }]
healStep: (html, { step, error, prevSteps }) => [{ role: 'user', content: 'As a test engineer...' }]
generatePageObject: (html, extraPrompt = '', rootLocator = null) => [{ role: 'user', content: 'As a test engineer...' }]
}
}
```
Expand Down Expand Up @@ -392,3 +485,9 @@ To debug AI features run tests with `DEBUG="codeceptjs:ai"` flag. This will prin
```
DEBUG="codeceptjs:ai" npx codeceptjs run --ai
```
or if you run it in shell mode:
```
DEBUG="codeceptjs:ai" npx codeceptjs shell --ai
```
2 changes: 2 additions & 0 deletions docs/pageobjects.md
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,8 @@ module.exports = function() {
## PageObject

> ✨ CodeceptJS can [generate PageObjects using AI](/ai#generate-pageobjects). It fetches all interactive elements from a page, generates locators and methods page and writes JS code. Generated page object can be tested on the fly within the same browser session.
If an application has different pages (login, admin, etc) you should use a page object.
CodeceptJS can generate a template for it with the following command:

Expand Down
48 changes: 47 additions & 1 deletion lib/ai.js
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,36 @@ const defaultPrompts = {
Here is HTML code of a page where the failure has happened: \n\n${html}`,
}];
},

generatePageObject: (html, extraPrompt = '', rootLocator = null) => [{
role: 'user',
content: `As a test automation engineer I am creating a Page Object for a web application using CodeceptJS.
Here is an sample page object:
const { I } = inject();
module.exports = {
// setting locators
element1: '#selector',
element2: '.selector',
element3: locate().withText('text'),
// seting methods
doSomethingOnPage(params) {
// ...
},
}
I want to generate a Page Object for the page I provide.
Write JavaScript code in similar manner to list all locators on the page.
Use locators in order of preference: by text (use locate().withText()), label, CSS, XPath.
Avoid TailwindCSS, Bootstrap or React style formatting classes in locators.
Add methods to to interact with page when needed.
${extraPrompt}
${rootLocator ? `All provided elements are inside '${rootLocator}'. Declare it as root variable and for every locator use locate(...).inside(root)` : ''}
Add only locators from this HTML: \n\n${html}`,
}],
};

class AiAssistant {
Expand Down Expand Up @@ -182,10 +212,26 @@ class AiAssistant {
return this.config.response(response);
}

/**
*
* @param {*} extraPrompt
* @param {*} locator
* @returns
*/
async generatePageObject(extraPrompt = null, locator = null) {
if (!this.isEnabled) return [];
if (!this.minifiedHtml) throw new Error('No HTML context provided');

const response = await this.createCompletion(this.prompts.generatePageObject(this.minifiedHtml, locator, extraPrompt));
if (!response) return [];

return this.config.response(response);
}

calculateTokens(messages) {
// we implement naive approach for calculating tokens with no extra requests
// this approach was tested via https://platform.openai.com/tokenizer
// we need it to display current usage tokens usage so users could analyze effectiveness of AI
// we need it to display current tokens usage so users could analyze effectiveness of AI

const inputString = messages.map(m => m.content).join(' ').trim();
const numWords = (inputString.match(/[^\s\-:=]+/g) || []).length;
Expand Down
91 changes: 84 additions & 7 deletions lib/helper/AI.js
Original file line number Diff line number Diff line change
@@ -1,15 +1,23 @@
const Helper = require('@codeceptjs/helper');
const ora = require('ora-classic');
const fs = require('fs');
const path = require('path');
const ai = require('../ai');
const standardActingHelpers = require('../plugin/standardActingHelpers');
const Container = require('../container');
const { splitByChunks, minifyHtml } = require('../html');
const { beautify } = require('../utils');
const output = require('../output');
const { registerVariable } = require('../pause');

/**
* AI Helper for CodeceptJS.
*
* This helper class provides integration with the AI GPT-3.5 or 4 language model for generating responses to questions or prompts within the context of web pages. It allows you to interact with the GPT-3.5 model to obtain intelligent responses based on HTML fragments or general prompts.
* This helper should be enabled with any web helpers like Playwright or Puppeteer or WebDrvier to ensure the HTML context is available.
*
* Use it only in development mode. It is recommended to run it only inside pause() mode.
*
* ## Configuration
*
* This helper should be configured in codecept.json or codecept.conf.js
Expand Down Expand Up @@ -66,9 +74,9 @@ class AI extends Helper {

if (htmlChunks.length > 1) messages.push({ role: 'user', content: 'If action is not possible on this page, do not propose anything, I will send another HTML fragment' });

const response = await this.aiAssistant.createCompletion(messages);
const response = await this._processAIRequest(messages);

console.log(response);
output.print(response);

responses.push(response);
}
Expand Down Expand Up @@ -96,15 +104,15 @@ class AI extends Helper {
{ role: 'user', content: `Within this HTML: ${minifyHtml(html)}` },
];

const response = await this.aiAssistant.createCompletion(messages);
const response = await this._processAIRequest(messages);

console.log(response);
output.print(response);

return response;
}

/**
* Send a general request to ChatGPT and return response.
* Send a general request to AI and return response.
* @param {string} prompt
* @returns {Promise<string>} - A Promise that resolves to the generated response from the GPT model.
*/
Expand All @@ -113,10 +121,79 @@ class AI extends Helper {
{ role: 'user', content: prompt },
];

const response = await this.aiAssistant.createCompletion(messages);
const response = await this._processAIRequest(messages);

output.print(response);

return response;
}

/**
* Generates PageObject for current page using AI.
*
* It saves the PageObject to the output directory. You can review the page object and adjust it as needed and move to pages directory.
* Prompt can be customized in a global config file.
*
* ```js
* // create page object for whole page
* I.askForPageObject('home');
*
* // create page object with extra prompt
* I.askForPageObject('home', 'implement signIn(username, password) method');
*
* // create page object for a specific element
* I.askForPageObject('home', null, '.detail');
* ```
*
* Asks for a page object based on the provided page name, locator, and extra prompt.
*
* @async
* @param {string} pageName - The name of the page to retrieve the object for.
* @param {string|null} [extraPrompt=null] - An optional extra prompt for additional context or information.
* @param {string|null} [locator=null] - An optional locator to find a specific element on the page.
* @returns {Promise<Object>} A promise that resolves to the requested page object.
*/
async askForPageObject(pageName, extraPrompt = null, locator = null) {
const html = locator ? await this.helper.grabHTMLFrom(locator) : await this.helper.grabSource();

const spinner = ora(' Processing AI request...').start();
await this.aiAssistant.setHtmlContext(html);
const response = await this.aiAssistant.generatePageObject(extraPrompt, locator);
spinner.stop();

if (!response[0]) {
output.error('No response from AI');
return '';
}

const code = beautify(response[0]);

console.log(response);
output.print('----- Generated PageObject ----');
output.print(code);
output.print('-------------------------------');

const fileName = path.join(output_dir, `${pageName}Page-${Date.now()}.js`);

output.print(output.styles.bold(`Page object for ${pageName} is saved to ${output.styles.bold(fileName)}`));
fs.writeFileSync(fileName, code);

try {
registerVariable('page', require(fileName));
output.success('Page object registered for this session as `page` variable');
output.print('Use `=>page.methodName()` in shell to run methods of page object');
output.print('Use `click(page.locatorName)` to check locators of page object');
} catch (err) {
output.error('Error while registering page object');
output.error(err.message);
}

return code;
}

async _processAIRequest(messages) {
const spinner = ora(' Processing AI request...').start();
const response = await this.aiAssistant.createCompletion(messages);
spinner.stop();
return response;
}
}
Expand Down
19 changes: 16 additions & 3 deletions lib/history.js
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,9 @@ const output = require('./output');
*/
class ReplHistory {
constructor() {
if (global.output_dir) {
this.historyFile = path.join(global.output_dir, 'cli-history');
}
this.commands = [];
}

Expand All @@ -21,16 +24,26 @@ class ReplHistory {
this.commands.pop();
}

load() {
if (!this.historyFile) return;
if (!fs.existsSync(this.historyFile)) {
return;
}

const history = fs.readFileSync(this.historyFile, 'utf-8');
return history.split('\n').reverse().filter(line => line.startsWith('I.')).map(line => line.slice(2));
}

save() {
if (!this.historyFile) return;
if (this.commands.length === 0) {
return;
}

const historyFile = path.join(global.output_dir, 'cli-history');
const commandSnippet = `\n\n<<< Recorded commands on ${new Date()}\n${this.commands.join('\n')}`;
fs.appendFileSync(historyFile, commandSnippet);
fs.appendFileSync(this.historyFile, commandSnippet);

output.print(colors.yellow(` Commands have been saved to ${historyFile}`));
output.print(colors.yellow(` Commands have been saved to ${this.historyFile}`));

this.commands = [];
}
Expand Down
Loading

0 comments on commit e48051c

Please sign in to comment.