The Positive Outcomes of Negative Testing

As software testers and automation engineers, we often think about the "Happy Path"- the path that the user will most likely take when they are using our application.  When we write our automated UI tests, we want to make sure that we are automating those Happy Paths, and when we write API automation, we want to verify that every endpoint returns a "200 OK" or similar successful response.

But it's important to think about negative testing, in both our manual and automated tests.  Here are a few reasons why:

Our automated tests might be passing for the wrong reasons.

When I first started writing automated UI tests in Javascript, I didn't understand the concept of the promise.  I just assumed that when I made a request to locate an element, it wouldn't return that element until it was actually located.  I was so excited when my tests started coming back with the green "Passed" result, until a co-worker suggested I try to make the test fail by asserting on a different value.  It passed again, because it was actually validating against the promise that existed, which was always returning "True".  That taught me a valuable lesson- never assume that your automated tests are working correctly just because they are passing.  Be sure to run some scenarios where your tests should fail, and make sure that they do so.  This way you can be sure that you are really testing what you think you are testing.

Negative testing can expose improperly handled errors that could impact a user.

In API testing, any client-related error should result in a 400-level response, rather than a 500-level server error.  If you are doing negative testing and you discover that a 403 response is now coming back as a 500, this could mean that the code is no longer handling that use case properly.  A 500 response from the server could keep the user from getting the appropriate information they need for fixing their error, or at worst, it could crash the application.

Negative testing can find security holes.

Just as important as making sure that a user can log in to an application is making sure that a user can't log into an application when they aren't supposed to.  If you only run a login test with a valid username and password, you are missing this crucial area!  I have seen a situation where a user could log in with anything as the password, a situation where a user could log in with a blank password, and a situation where if both the username and password were wrong the user could log in.

It's also crucial to verify that certain users don't have access to parts of an application.  Having a carefully tested and functional Admin page won't mean much if it turns out that any random user can get to it.

Negative testing keeps your database clean.

As I mentioned in my post two weeks ago on input validation, having good, valid data in your database will help keep your application healthy.  Data that doesn't conform to expectations can cause web pages to crash or fail to load, or cause information to be displayed incorrectly.  The more negative testing you can do on your inputs, the more you can ensure that you will only have good data.

For every input field I am responsible for testing, I like to know exactly which characters are allowed.  Then I can run a whole host of negative tests to make sure that entries with the forbidden characters are refused.

Sometimes users take the negative path.

It is so easy, especially with a new feature that is being rushed to meet a deadline, to forget to test those user paths where they will hit the "Cancel" or "Delete" button.  But users do this all the time; just think about times where you have thought about making an online purchase and then changed your mind and removed an item from your cart.  Imagine your frustration if you weren't able to remove something from your cart, or if a "Cancel" button didn't clear out a form to allow you to start again.  User experience in this area is just as crucial as the Happy Path.

Software testing is about looking for unexpected behaviors, so that we find them before a user does.  When negative testing is combined with Happy Path testing, we can ensure that our users will have no unpleasant surprises.

Three Ways to Test Output Validation

Last week, I wrote about the importance of input validation for the security, appearance, and performance of your application.  An astute reader commented that we should think about output validation as well.  I love it when people give me ideas for blog posts!

There are three main things to think about when testing outputs:

1. How is the output displayed?  

A perfect example of an output that you would want to check the appearance of is a phone number.  Hopefully when a user adds a phone number to your application's data store it is being saved without any parentheses, periods, or dashes.  But when you display that number to the user, you probably won't want to display it as 8008675309, because that's hard to read.  You'll want the number to be formatted in a way that the user would expect; for US users, the number would be displayed as 800-867-5309 or (800) 867-5309.

Another example would be a currency value.  If financial calculations are made and the result is displayed to the user, you wouldn't want the result displayed as $45.655, because no one makes payments in half-pennies.  The calculation should be rounded or truncated so that there are only two decimal places.

2. Will the result of a calculation be saved correctly in the database?

Imagine that you have an application that takes a value for x and a value for y from the user, adds them together, and stores them as z.  The data type for x, y, and z is set to tinyint in the database.  If you're doing a calculation with small numbers, such as when x is 10 and y is 20, this won't be a problem.  But what happens if x is 255- the upper limit of tinyint- and y is 1?  Now your calculated value for z is 256, which is more than can be stored in the tinyint field, and you will get a server error.

Similarly, you'll want to make sure that your calculation results don't go below zero in certain situations, such as an e-commerce app.  If your user has merchandise totaling $20, and a discount coupon for $25, you don't want to have your calculations show that you owe them $5!

3. Are the values being calculated correctly?

This is especially important for complicated financial applications. Let's imagine that we are testing a tax application for the Republic of Jackvonia.  The Jackvonia tax brackets are simple:

Income	Tax Rate
$0 - $25,000	1%
$25,001 - $50,000	3%
$50,001 - $75,000	5%
$75,001 - $100,000	7%
$100,001 and over	9%

There is only one type of tax deduction in Jackvonia, and that is the dependents deduction:

Number of Dependents	Deduction
1	$100
2	$200
3 or more	$300

The online tax calculator for Jackvonia residents has an income field, which can contain any dollar amount from 0 to one million dollars; and a dependents field, which can contain any whole number of dependents from 0 to 10.  The user enters those values and clicks the "Calculate" button, and then the amount of taxes owed appears.

If you were charged with testing the tax calculator, how would you test it?  Here's what I would do:

First, I would verify that a person with $0 income and 0 dependents would owe $0 in taxes.

Next, I would verify that it was not possible to owe a negative amount of taxes: if, for example, a person made $25,000 and had three dependents, they should owe $0 in taxes, not -$50.

Then I would verify that the tax rate was being applied correctly at the boundaries of each tax bracket.  So a person who made $1 and had 0 dependents should owe $.01, and a person who made $25,000 and had 0 dependents should owe $250.  Similarly, a person who made $25,001 and had 0 dependents should owe $750.03 in taxes.  I would continue that pattern through the other tax brackets, and would include a test with one million dollars, which is the upper limit of the income field.

Finally, I would test the dependents calculation. I would test with 1, 2, and 3 dependents in each tax bracket and verify that the $100, $200, or $300 tax deduction was being applied correctly. I would also do a test with 4, 5, and 10 dependents to make sure that the deduction was $300 in each case.

This is a lot of repetitive testing, so it would definitely be a good idea to automate it. Most automation frameworks allow a test to process a grid or table of data, so you could easily test all of the above scenarios and even add more for more thorough testing.

Output validation is so important because if your users can't trust your calculations, they won't use your application!  Remember to always begin with thinking about what you should test, and then design automation that verifies the correct functionality even in boundary cases.

Four Reasons You Should Test Input Validation (Even Though It's Boring)

When I first started in software testing, I found it fun to test text fields.  It was entertaining to discover what would happen when I put too many characters in a field.  But as I entered my fourth QA job and discovered that once again I had a contact form to test, my interest started to wane.  It's not all that interesting to input the maximum amount of characters, the minimum amount of characters, one too many characters, one too few characters, and so on for every text field in an application!

However, it was around this time that I realized that input validation is extremely important.  Whenever a user has the opportunity to add data in an application, there is the potential of malicious misuse or unexpected consequences.  Testing input validation is a critical activity for the following four reasons:

1. Security

Malicious users can exploit text fields to get information they shouldn't have.  They can do this in three ways:

• Cross-site scripting- an attacker enters a script into a text field.  If the text field does not have proper validation that strips out scripting characters, the value will be saved and the script will then execute automatically when an unsuspecting user navigates to the page.  The executed script can return information about the user's session id, or even pop up a form and prompt the user to enter their password, which then gets written to a location the attacker has access to.

• SQL injection- if a text field allows certain characters such as semicolons, it's possible that an attacker can enter values into the field which will fool the database into executing a SQL command and returning information such as the usernames and passwords of all the users on the site.  It's even possible for an attacker to erase a data table through SQL injection.

• Buffer overflow attack- if a variable is configured to have enough memory for a certain number of characters, but it's possible to enter a much larger number of characters into the associated text field, the memory can overflow into other locations.  When this happens, an attacker can exploit this to gain access to sensitive information or even manipulate the program.

2. Stability

When a user is able to input data that the application is not equipped to handle, the application can react in unexpected ways, such as crashing or refusing to save.  Here are a couple of examples:

• My Zip code begins with a 0.  I have encountered forms where I can't save my address because the application strips the leading 0 off of the Zip code and then tells me that my Zip code has only four digits.  

• I have a co-worker who has both a hyphen and an apostrophe in his last name.  He told me that entering his name frequently breaks the forms he is filling out.

3. Visual Consistency

When a field has too many characters in it, it can affect the way a page is displayed.  This can be easily seen when looking at any QA test environment.  For example, if a list of first names and last names is displayed on a page of contacts, you will often see that some astute tester has entered "Reallyreallyreallyreallyreallylongfirstname Reallyreallyreallyreallyreallylonglastname" as one of the contacts.  If a name like this causes the contact page to be excessively wide and need a horizontal scroll bar, then a real user in the production environment could potentially cause the page to render in this way.

4. Health of the Database

When fields are not validated correctly, all kinds of erroneous data can be saved to the database.  This can affect both how the application runs and how it behaves.  

The phone number field is an excellent example of how unhealthy data can affect an application.  I worked for a company where for years phone numbers were not validated properly.  When we were updating the application, we wanted to automatically format phone numbers so they would display attractively in this format:  (800)-555-1000.  But because there were values in the database like "Dad's number", there was no way to format them, therefore causing an error on the page.

Painstakingly validating input fields can be very tedious, but the above examples demonstrate why it is so important.  The good news is that there are ways to alleviate the boredom.  Automating validation checks can keep us from having to manually run the same tests repeatedly.  Monkey-testing tools can help flush out bugs.  And adding a sense of whimsy to testing can help keep things interesting.  I have all the lyrics to "Frosty the Snowman" saved in a text file.  Whenever I need to test the allowed length of a text field, I paste all or some of the lyrics into the field.  When a developer sees database entries with "Frosty the Snowman was a j", they know I have been there!

Easy Free Automation Part VIII: Accessibility Tests

Accessibility in the context of a software application means that as many people as possible can use the application easily.  When making an application accessible, we should consider users with limited vision or hearing, limited cognitive ability, and limited dexterity.  Accessibility also means that users from all over the world can use the application, even if their language is different from that of the developers who created it.

In this final post in my "Easy Free Automation" series, I'll be showing two easy ways to test for accessibility.  I'll be using Python and Selenium Webdriver.  You can download the simple test here.

To run the test, you will need to have Python and Selenium installed.  You can find instructions for installing Python in Easy Free Automation Part I: Unit Tests.  To install Selenium, open a command window and type pip install selenium.  You may also need to have Chromedriver installed.  You can find instructions for installing it here.

Once you have downloaded the test file and installed all the needed components, navigate to the test folder in the command line and type python3 easyFreeAccessibilityTest.py.  (If you don't have Python 3, or if you don't have two versions of Python installed, you may be able to type python instead of python3.) The test should run, the Chrome browser should open and close when the test is completed, and in the command line you should see these two log entries:
Alt text is present
Page is in German

Let's take a look at these two tests to see what they do.  The first test verifies that an image has an alt text.  Alt texts are used to provide a description of an image for any user who might not be able to see the image.  A screen-reading application will read the alt text aloud so the user will know what image is portrayed.

driver.get("https://www.amazon.com/s?k=goodnight+moon&ref=nb_sb_noss_1")
elem = driver.find_element_by_class_name("s-image")
val = elem.get_attribute('alt')
if val == 'Goodnight Moon':
print('Alt text is present')
else:
print('Alt text is missing or incorrect')

In the first line, we are navigating to an Amazon.com web page where we are searching for the children's book "Goodnight Moon".  In the next line, we are locating the book image.  In the third line, we are getting the 'alt' attribute of the web element and assigning it to the variable 'val'.  If there is no alt text, this variable will remain null.

Finally we are using an if statement to assert that the alt text is correct.  If the alt text is not the title of the book, we will get a message that the text is missing.

The second test verifies that we are able to change the language of the Audi website to German.

driver.get("https://www.audi.com/en.html")
driver.find_element_by_link_text("DE").click()
try:
elem = driver.find_element_by_link_text("Kontakt")
if elem:
print('Page is in German')
except:
print('Page is not in German')

In the first line, we navigate to the Audi website.  In the second line, we find the button that will change the language to German, and we click it.  Then we look for the element with the link text of "Kontakt".  If we find the element, we can conclude that we are on the German version of the page.  If we do not find the element, the page has not been changed to German.  The reason I am using a try-except block here is that if the element with the link text is not located, an error will be thrown.  I'm catching the error so that an appropriate error message can be logged and the test can end properly.

There are other ways to verify things like alt texts and page translations.  There are CSS scanning tools that will verify the presence of alt texts and rate how well a page can be read by a screen reader.  There are services that will check your internationalization of your site with native speakers of many different languages.  But if you are looking for an easy, free way to check these things, this simple test script provides you with a way to get started.

For the last eight weeks, we've looked at easy, free ways to automate each area of the Automation Test Wheel.  I hope you have found these posts informative!  If you missed any of the posts, I hope you'll go back and take a look.  Remember also that each week has a code sample that can be found at my Github page.  Happy automating!

Easy Free Automation Part VII: Load Tests

Load testing is a key part of checking the health of your application.  Just because you get a timely response when you make an HTTP request in your test environment doesn't mean that the application will respond appropriately when 100,000 users are making the same request in your production environment.  With load testing, you can simulate different scenarios as you make HTTP calls to determine how your application will behave under real-world conditions.

There are a wide variety of load testing tools available, but many of them require a subscription.  Both paid and free tools can often be confusing to use or difficult to set up.  For load testing that is free, easy to install, and fairly easy to set up, I recommend K6.

As with every installment of this "Easy Free Automation" series, I've created an automation script that you can download here.  In order to run the load test script, you'll need to install K6, which is easy to do with these instructions.

Once you have installed K6 and downloaded your test script, open a command window and navigate to the location where you downloaded the script.  Then type k6 run loadTestScript.js.  The test should run and display a number of metrics as the result.

Let's take a look at what this script is doing.  I'm making four different requests to the Swagger Pet Store.  (For more information about the Swagger Pet Store, take a look at this blog post and the posts that follow it.)  I've kept my requests very simple to make it easier to read the test script: I'm adding a pet with just a pet name, I'm retrieving the pet, I'm updating the pet by changing the name, and I'm deleting the pet.

import http from "k6/http";
import { check, sleep } from "k6";

In the first two lines of the script, I'm importing the modules needed for the script: the http module that allows me to make http requests, and the check and sleep modules that allow me to do assertions and put in a wait in between requests. 

export let options = {
  vus: 1,
  duration: "5s"
};

In this section, I'm setting the options for running my load tests.  The word "vus" stands for "virtual users", and "duration" describes how long in seconds the test should run.

var id = Math.floor(Math.random() * 10000);
console.log("ID used: "+id);

Here I'm coming up with a random id to use for the pet, which will get passed through one complete test iteration. 

var url = "http://petstore.swagger.io/v2/pet";
var payload = JSON.stringify({ id: id, name: "Grumpy Cat" });
var params =  { headers: { "Content-Type": "application/json" } }
let postRes = http.post(url, payload, params);

This is how the POST request is set up.  First I set the URL, then the payload, then the headers; then I do the POST request. 

check(postRes, {
    "post status is 200": (r) => r.status == 200,
    "post transaction time is OK": (r) => r.timings.duration < 200
  });
sleep(1);

Once the POST has executed, and the result has been assigned to the postRes variable, I check to make sure that the status returned was 200, and that the transaction time was less than 200 milliseconds.  Finally, I have the script sleep for one second. 

Now let's take a look at the load test output:

INFO[0005] ID used: 1067

This is the id created for the test, which I set up to be logged in line 12 of the script.  At the beginning of each iteration of the script, a new id will be created and logged. 

✓ put status is 200
✓ put transaction time is OK
✓ delete status is 200
✓ delete transaction time is OK
✓ post status is 200
✓ post transaction time is OK
✓ get status is 200
✓ get transaction time is OK

Here are the results of my assertions.  All the POSTs, GETs, PUTs, and DELETEs were successful.

http_req_duration..........: avg=27.56ms min=23.16ms med=26.68ms max=34.69ms p(90)=31.66ms p(95)=33.18ms  

This section shows metrics about the duration of each request.  The average request duration was 27.56 milliseconds, and the maximum request time was 34.69 milliseconds.

iterations................. : 1       0.199987/s
vus........................ : 1         min=1 max=1
vus_max.................... : 1    min=1 max=1

This section shows how many complete iterations were run during the test, and what the frequency was; how many virtual users there were; and how many virtual users there were in maximum.

Obviously, this wasn't much of a load test, because we only used one user and it only ran for five seconds!  Let's make a change to the script and see how our results change.  First we'll leave the number of virtual users at 1, but we'll set the test to run for a full minute.  Change line 6 of the script to duration: "1m", and run the test again with the k6 run loadTestScript.js command. 

http_req_duration..........: avg=26.13ms min=22.3ms  med=25.86ms max=37.45ms p(90)=27.57ms  p(95)=33.56ms

The results look very similar to our first test, which isn't surprising, since we are still using just one virtual user. 

iterations.................: 14      0.233333/s

Because we had a longer test time, we went through several more iterations, at the rate of .23 per second.

Now let's see what happens when we use 10 virtual users.  Change line 5 of the test to: vus: 10, and run the test again.

✓ delete transaction time is OK
✓ post status is 200
✓ post transaction time is OK
✗ get status is 200
     ↳  83% — ✓ 117 / ✗ 23
✓ get transaction time is OK
✓ put status is 200
✓ put transaction time is OK
✗ delete status is 200
     ↳  77% — ✓ 108 / ✗ 31

We are now seeing the impact of adding load to the test; some of our GET requests and DELETE requests failed.

http_req_duration..........: avg=27.8ms  min=21.17ms med=26.67ms max=63.74ms p(90)=33.08ms p(95)=34.98ms

Note also that our maximum duration was much longer than our duration in the previous two test runs.

This is by no means a complete load test; it's just an introduction to what can be done with the K6 tool.  It's possible to set up the test to have realistic ramp-up and ramp-down times, where there's less load at the beginning and end of the test and more load in the middle.  You can also create your own custom metrics to make it easier to analyze the results of each request type.  If you ever find yourself needing to some quick free load testing, K6 may be the tool for you.

Next week, I'll close out the "Easy Free Automation" series with a look at accessibility tests!