After a decade or extra the place Single-Web page-Functions generated by
JavaScript frameworks have
develop into the norm, we see that server-side rendered HTML is turning into
well-liked once more, additionally due to libraries equivalent to HTMX or Turbo. Writing a wealthy net UI in a
historically server-side language like Go or Java is no longer simply potential,
however a really enticing proposition.
We then face the issue of find out how to write automated assessments for the HTML
elements of our net purposes. Whereas the JavaScript world has developed highly effective and refined methods to check the UI,
ranging in measurement from unit-level to integration to end-to-end, in different
languages we shouldn’t have such a richness of instruments obtainable.
When writing an online software in Go or Java, HTML is usually generated
by means of templates, which comprise small fragments of logic. It’s actually
potential to check them not directly by means of end-to-end assessments, however these assessments
are gradual and costly.
We will as an alternative write unit assessments that use CSS selectors to probe the
presence and proper content material of particular HTML components inside a doc.
Parameterizing these assessments makes it straightforward so as to add new assessments and to obviously
point out what particulars every check is verifying. This strategy works with any
language that has entry to an HTML parsing library that helps CSS
selectors; examples are offered in Go and Java.
Degree 1: checking for sound HTML
The primary factor we need to examine is that the HTML we produce is
mainly sound. I do not imply to examine that HTML is legitimate in response to the
W3C; it could be cool to do it, but it surely’s higher to start out with a lot less complicated and sooner checks.
As an example, we would like our assessments to
break if the template generates one thing like
<div>foo</p>
Let’s examine find out how to do it in levels: we begin with the next check that
tries to compile the template. In Go we use the usual html/template
package deal.
Go
func Test_wellFormedHtml(t *testing.T) { templ := template.Should(template.ParseFiles("index.tmpl")) _ = templ }
In Java, we use jmustache
as a result of it is quite simple to make use of; Freemarker or
Velocity are different widespread selections.
Java
@Take a look at void indexIsSoundHtml() { var template = Mustache.compiler().compile( new InputStreamReader( getClass().getResourceAsStream("/index.tmpl"))); }
If we run this check, it can fail, as a result of the index.tmpl
file does
not exist. So we create it, with the above damaged HTML. Now the check ought to move.
Then we create a mannequin for the template to make use of. The appliance manages a todo-list, and
we are able to create a minimal mannequin for demonstration functions.
Go
func Test_wellFormedHtml(t *testing.T) {
templ := template.Should(template.ParseFiles("index.tmpl"))
mannequin := todo.NewList()
_ = templ
_ = mannequin
}
Java
@Take a look at
void indexIsSoundHtml() {
var template = Mustache.compiler().compile(
new InputStreamReader(
getClass().getResourceAsStream("/index.tmpl")));
var mannequin = new TodoList();
}
Now we render the template, saving the leads to a bytes buffer (Go) or as a String
(Java).
Go
func Test_wellFormedHtml(t *testing.T) {
templ := template.Should(template.ParseFiles("index.tmpl"))
mannequin := todo.NewList()
var buf bytes.Buffer
err := templ.Execute(&buf, mannequin)
if err != nil {
panic(err)
}
}
Java
@Take a look at
void indexIsSoundHtml() {
var template = Mustache.compiler().compile(
new InputStreamReader(
getClass().getResourceAsStream("/index.tmpl")));
var mannequin = new TodoList();
var html = template.execute(mannequin);
}
At this level, we need to parse the HTML and we count on to see an
error, as a result of in our damaged HTML there’s a div
aspect that
is closed by a p
aspect. There’s an HTML parser within the Go
normal library, however it’s too lenient: if we run it on our damaged HTML, we do not get an
error. Fortunately, the Go normal library additionally has an XML parser that may be
configured to parse HTML (due to this Stack Overflow reply)
Go
func Test_wellFormedHtml(t *testing.T) {
templ := template.Should(template.ParseFiles("index.tmpl"))
mannequin := todo.NewList()
// render the template right into a buffer
var buf bytes.Buffer
err := templ.Execute(&buf, mannequin)
if err != nil {
panic(err)
}
// examine that the template will be parsed as (lenient) XML
decoder := xml.NewDecoder(bytes.NewReader(buf.Bytes()))
decoder.Strict = false
decoder.AutoClose = xml.HTMLAutoClose
decoder.Entity = xml.HTMLEntity
for {
_, err := decoder.Token()
swap err {
case io.EOF:
return // We're executed, it is legitimate!
case nil:
// do nothing
default:
t.Fatalf("Error parsing html: %s", err)
}
}
}
This code configures the HTML parser to have the best stage of leniency
for HTML, after which parses the HTML token by token. Certainly, we see the error
message we wished:
--- FAIL: Test_wellFormedHtml (0.00s) index_template_test.go:61: Error parsing html: XML syntax error on line 4: surprising finish aspect </p>
In Java, a flexible library to make use of is jsoup:
Java
@Take a look at
void indexIsSoundHtml() {
var template = Mustache.compiler().compile(
new InputStreamReader(
getClass().getResourceAsStream("/index.tmpl")));
var mannequin = new TodoList();
var html = template.execute(mannequin);
var parser = Parser.htmlParser().setTrackErrors(10);
Jsoup.parse(html, "", parser);
assertThat(parser.getErrors()).isEmpty();
}
And we see it fail:
java.lang.AssertionError: Anticipating empty however was:<[<1:13>: Unexpected EndTag token [</p>] when in state [InBody],
Success! Now if we copy over the contents of the TodoMVC
template to our index.tmpl
file, the check passes.
The check, nevertheless, is simply too verbose: we extract two helper capabilities, in
order to make the intention of the check clearer, and we get
Go
func Test_wellFormedHtml(t *testing.T) { mannequin := todo.NewList() buf := renderTemplate("index.tmpl", mannequin) assertWellFormedHtml(t, buf) }
Java
@Take a look at void indexIsSoundHtml() { var mannequin = new TodoList(); var html = renderTemplate("/index.tmpl", mannequin); assertSoundHtml(html); }
Degree 2: testing HTML construction
What else ought to we check?
We all know that the appears to be like of a web page can solely be examined, finally, by a
human taking a look at how it’s rendered in a browser. Nonetheless, there may be typically
logic in templates, and we would like to have the ability to check that logic.
One may be tempted to check the rendered HTML with string equality,
however this method fails in apply, as a result of templates comprise plenty of
particulars that make string equality assertions impractical. The assertions
develop into very verbose, and when studying the assertion, it turns into troublesome
to know what it’s that we’re attempting to show.
What we’d like
is a way to say that some elements of the rendered HTML
correspond to what we count on, and to ignore all the small print we do not
care about. A method to do that is by working queries with the CSS selector language:
it’s a highly effective language that permits us to pick out the
components that we care about from the entire HTML doc. As soon as we now have
chosen these components, we (1) depend that the variety of aspect returned
is what we count on, and (2) that they comprise the textual content or different content material
that we count on.
The UI that we’re presupposed to generate appears to be like like this:
There are a number of particulars which can be rendered dynamically:
- The variety of gadgets and their textual content content material change, clearly
- The type of the todo-item adjustments when it is accomplished (e.g., the
second) - The “2 gadgets left” textual content will change with the variety of non-completed
gadgets - One of many three buttons “All”, “Energetic”, “Accomplished” might be
highlighted, relying on the present url; as an illustration if we resolve that the
url that exhibits solely the “Energetic” gadgets is/lively
, then when the present url
is/lively
, the “Energetic” button ought to be surrounded by a skinny crimson
rectangle - The “Clear accomplished” button ought to solely be seen if any merchandise is
accomplished
Every of this considerations will be examined with the assistance of CSS selectors.
This can be a snippet from the TodoMVC template (barely simplified). I
haven’t but added the dynamic bits, so what we see right here is static
content material, offered for example:
index.tmpl
<part class="todoapp"> <ul class="todo-list"> <!-- These are right here simply to point out the construction of the listing gadgets --> <!-- Checklist gadgets ought to get the category `accomplished` when marked as accomplished --> <li class="accomplished"> ② <div class="view"> <enter class="toggle" kind="checkbox" checked> <label>Style JavaScript</label> ① <button class="destroy"></button> </div> </li> <li> <div class="view"> <enter class="toggle" kind="checkbox"> <label>Purchase a unicorn</label> ① <button class="destroy"></button> </div> </li> </ul> <footer class="footer"> <!-- This ought to be `0 gadgets left` by default --> <span class="todo-count"><sturdy>0</sturdy> merchandise left</span> ⓷ <ul class="filters"> <li> <a class="chosen" href="#/">All</a> ④ </li> <li> <a href="#/lively">Energetic</a> </li> <li> <a href="#/accomplished">Accomplished</a> </li> </ul> <!-- Hidden if no accomplished gadgets are left ↓ --> <button class="clear-completed">Clear accomplished</button> ⑤ </footer> </part>
By wanting on the static model of the template, we are able to deduce which
CSS selectors can be utilized to determine the related components for the 5 dynamic
options listed above:
function | CSS selector | |
---|---|---|
① | All of the gadgets | ul.todo-list li |
② | Accomplished gadgets | ul.todo-list li.accomplished |
⓷ | Gadgets left | span.todo-count |
④ | Highlighted navigation hyperlink | ul.filters a.chosen |
⑤ | Clear accomplished button | button.clear-completed |
We will use these selectors to focus our assessments on simply the issues we need to check.
Testing HTML content material
The primary check will search for all of the gadgets, and show that the info
arrange by the check is rendered appropriately.
func Test_todoItemsAreShown(t *testing.T) { mannequin := todo.NewList() mannequin.Add("Foo") mannequin.Add("Bar") buf := renderTemplate(mannequin) // assert there are two <li> components contained in the <ul class="todo-list"> // assert the primary <li> textual content is "Foo" // assert the second <li> textual content is "Bar" }
We want a method to question the HTML doc with our CSS selector;
library for Go is goquery, that implements an API impressed by jQuery.
In Java, we maintain utilizing the identical library we used to check for sound HTML, specifically
jsoup. Our check turns into:
Go
func Test_todoItemsAreShown(t *testing.T) { mannequin := todo.NewList() mannequin.Add("Foo") mannequin.Add("Bar") buf := renderTemplate("index.tmpl", mannequin) // parse the HTML with goquery doc, err := goquery.NewDocumentFromReader(bytes.NewReader(buf.Bytes())) if err != nil { // if parsing fails, we cease the check right here with t.FatalF t.Fatalf("Error rendering template %s", err) } // assert there are two <li> components contained in the <ul class="todo-list"> choice := doc.Discover("ul.todo-list li") assert.Equal(t, 2, choice.Size()) // assert the primary <li> textual content is "Foo" assert.Equal(t, "Foo", textual content(choice.Nodes[0])) // assert the second <li> textual content is "Bar" assert.Equal(t, "Bar", textual content(choice.Nodes[1])) } func textual content(node *html.Node) string { // Somewhat mess as a result of the truth that goquery has // a .Textual content() technique on Choice however not on html.Node sel := goquery.Choice{Nodes: []*html.Node{node}} return strings.TrimSpace(sel.Textual content()) }
Java
@Take a look at void todoItemsAreShown() throws IOException { var mannequin = new TodoList(); mannequin.add("Foo"); mannequin.add("Bar"); var html = renderTemplate("/index.tmpl", mannequin); // parse the HTML with jsoup Doc doc = Jsoup.parse(html, ""); // assert there are two <li> components contained in the <ul class="todo-list"> var choice = doc.choose("ul.todo-list li"); assertThat(choice).hasSize(2); // assert the primary <li> textual content is "Foo" assertThat(choice.get(0).textual content()).isEqualTo("Foo"); // assert the second <li> textual content is "Bar" assertThat(choice.get(1).textual content()).isEqualTo("Bar"); }
If we nonetheless have not modified the template to populate the listing from the
mannequin, this check will fail, as a result of the static template
todo gadgets have completely different textual content:
Go
--- FAIL: Test_todoItemsAreShown (0.00s) index_template_test.go:44: First listing merchandise: need Foo, obtained Style JavaScript index_template_test.go:49: Second listing merchandise: need Bar, obtained Purchase a unicorn
Java
IndexTemplateTest > todoItemsAreShown() FAILED org.opentest4j.AssertionFailedError: Anticipating: <"Style JavaScript"> to be equal to: <"Foo"> however was not.
We repair it by making the template use the mannequin knowledge:
Go
<ul class="todo-list"> {{ vary .Gadgets }} <li> <div class="view"> <enter class="toggle" kind="checkbox"> <label>{{ .Title }}</label> <button class="destroy"></button> </div> </li> {{ finish }} </ul>
Java – jmustache
<ul class="todo-list"> {{ #allItems }} <li> <div class="view"> <enter class="toggle" kind="checkbox"> <label>{{ title }}</label> <button class="destroy"></button> </div> </li> {{ /allItems }} </ul>
Take a look at each content material and soundness on the identical time
Our check works, however it’s a bit verbose, particularly the Go model. If we’ll have extra
assessments, they’ll develop into repetitive and troublesome to learn, so we make it extra concise by extracting a helper operate for parsing the html. We additionally take away the
feedback, because the code ought to be clear sufficient
Go
func Test_todoItemsAreShown(t *testing.T) { mannequin := todo.NewList() mannequin.Add("Foo") mannequin.Add("Bar") buf := renderTemplate("index.tmpl", mannequin) doc := parseHtml(t, buf) choice := doc.Discover("ul.todo-list li") assert.Equal(t, 2, choice.Size()) assert.Equal(t, "Foo", textual content(choice.Nodes[0])) assert.Equal(t, "Bar", textual content(choice.Nodes[1])) } func parseHtml(t *testing.T, buf bytes.Buffer) *goquery.Doc { doc, err := goquery.NewDocumentFromReader(bytes.NewReader(buf.Bytes())) if err != nil { // if parsing fails, we cease the check right here with t.FatalF t.Fatalf("Error rendering template %s", err) } return doc }
Java
@Take a look at void todoItemsAreShown() throws IOException { var mannequin = new TodoList(); mannequin.add("Foo"); mannequin.add("Bar"); var html = renderTemplate("/index.tmpl", mannequin); var doc = parseHtml(html); var choice = doc.choose("ul.todo-list li"); assertThat(choice).hasSize(2); assertThat(choice.get(0).textual content()).isEqualTo("Foo"); assertThat(choice.get(1).textual content()).isEqualTo("Bar"); } personal static Doc parseHtml(String html) { return Jsoup.parse(html, ""); }
Significantly better! At the least for my part. Now that we extracted the parseHtml
helper, it is
a good suggestion to examine for sound HTML within the helper:
Go
func parseHtml(t *testing.T, buf bytes.Buffer) *goquery.Doc {
assertWellFormedHtml(t, buf)
doc, err := goquery.NewDocumentFromReader(bytes.NewReader(buf.Bytes()))
if err != nil {
// if parsing fails, we cease the check right here with t.FatalF
t.Fatalf("Error rendering template %s", err)
}
return doc
}
Java
personal static Doc parseHtml(String html) { var parser = Parser.htmlParser().setTrackErrors(10); var doc = Jsoup.parse(html, "", parser); assertThat(parser.getErrors()).isEmpty(); return doc; }
And with this, we are able to eliminate the primary check that we wrote, as we at the moment are testing for sound HTML on a regular basis.
The second check
Now we’re in place for testing extra rendering logic. The
second dynamic function in our listing is “Checklist gadgets ought to get the category
accomplished
when marked as accomplished”. We will write a check for this:
Go
func Test_completedItemsGetCompletedClass(t *testing.T) { mannequin := todo.NewList() mannequin.Add("Foo") mannequin.AddCompleted("Bar") buf := renderTemplate("index.tmpl", mannequin) doc := parseHtml(t, buf) choice := doc.Discover("ul.todo-list li.accomplished") assert.Equal(t, 1, choice.Measurement()) assert.Equal(t, "Bar", textual content(choice.Nodes[0])) }
Java
@Take a look at void completedItemsGetCompletedClass() { var mannequin = new TodoList(); mannequin.add("Foo"); mannequin.addCompleted("Bar"); var html = renderTemplate("/index.tmpl", mannequin); Doc doc = Jsoup.parse(html, ""); var choice = doc.choose("ul.todo-list li.accomplished"); assertThat(choice).hasSize(1); assertThat(choice.textual content()).isEqualTo("Bar"); }
And this check will be made inexperienced by including this little bit of logic to the
template:
Go
<ul class="todo-list">
{{ vary .Gadgets }}
<li class="{{ if .IsCompleted }}accomplished{{ finish }}">
<div class="view">
<enter class="toggle" kind="checkbox">
<label>{{ .Title }}</label>
<button class="destroy"></button>
</div>
</li>
{{ finish }}
</ul>
Java – jmustache
<ul class="todo-list">
{{ #allItems }}
<li class="{{ #isCompleted }}accomplished{{ /isCompleted }}">
<div class="view">
<enter class="toggle" kind="checkbox">
<label>{{ title }}</label>
<button class="destroy"></button>
</div>
</li>
{{ /allItems }}
</ul>
So little by little, we are able to check and add the assorted dynamic options
that our template ought to have.
Make it straightforward so as to add new assessments
The primary of the 20 suggestions from the wonderful discuss by Russ Cox on Go
Testing is “Make it straightforward so as to add new check instances“. Certainly, in Go there
is a bent to make most assessments parameterized, for this very cause.
Then again, whereas Java has
good help
for parameterized assessments with JUnit 5, they are not used as a lot.
Since our present two assessments have the identical construction, we
may issue them right into a single parameterized check.
A check case for us will include:
- A reputation (in order that we are able to produce clear error messages when the check
fails) - A mannequin (in our case a
todo.Checklist
) - A CSS selector
- A listing of textual content matches that we look forward to finding after we run the CSS
selector on the rendered HTML.
So that is the info construction for our check instances:
Go
var testCases = []struct { title string mannequin *todo.Checklist selector string matches []string }{ { title: "all todo gadgets are proven", mannequin: todo.NewList(). Add("Foo"). Add("Bar"), selector: "ul.todo-list li", matches: []string{"Foo", "Bar"}, }, { title: "accomplished gadgets get the 'accomplished' class", mannequin: todo.NewList(). Add("Foo"). AddCompleted("Bar"), selector: "ul.todo-list li.accomplished", matches: []string{"Bar"}, }, }
Java
report TestCase(String title, TodoList mannequin, String selector, Checklist<String> matches) { @Override public String toString() { return title; } } public static TestCase[] indexTestCases() { return new TestCase[]{ new TestCase( "all todo gadgets are proven", new TodoList() .add("Foo") .add("Bar"), "ul.todo-list li", Checklist.of("Foo", "Bar")), new TestCase( "accomplished gadgets get the 'accomplished' class", new TodoList() .add("Foo") .addCompleted("Bar"), "ul.todo-list li.accomplished", Checklist.of("Bar")), }; }
And that is our parameterized check:
Go
func Test_indexTemplate(t *testing.T) { for _, check := vary testCases { t.Run(check.title, func(t *testing.T) { buf := renderTemplate("index.tmpl", check.mannequin) assertWellFormedHtml(t, buf) doc := parseHtml(t, buf) choice := doc.Discover(check.selector) require.Equal(t, len(check.matches), len(choice.Nodes), "surprising # of matches") for i, node := vary choice.Nodes { assert.Equal(t, check.matches[i], textual content(node)) } }) } }
Java
@ParameterizedTest @MethodSource("indexTestCases") void testIndexTemplate(TestCase check) { var html = renderTemplate("/index.tmpl", check.mannequin); var doc = parseHtml(html); var choice = doc.choose(check.selector); assertThat(choice).hasSize(check.matches.measurement()); for (int i = 0; i < check.matches.measurement(); i++) { assertThat(choice.get(i).textual content()).isEqualTo(check.matches.get(i)); } }
We will now run our parameterized check and see it move:
Go
$ go check -v === RUN Test_indexTemplate === RUN Test_indexTemplate/all_todo_items_are_shown === RUN Test_indexTemplate/completed_items_get_the_'accomplished'_class --- PASS: Test_indexTemplate (0.00s) --- PASS: Test_indexTemplate/all_todo_items_are_shown (0.00s) --- PASS: Test_indexTemplate/completed_items_get_the_'accomplished'_class (0.00s) PASS okay tdd-html-templates 0.608s
Java
$ ./gradlew check > Activity :check IndexTemplateTest > testIndexTemplate(TestCase) > [1] all todo gadgets are proven PASSED IndexTemplateTest > testIndexTemplate(TestCase) > [2] accomplished gadgets get the 'accomplished' class PASSED
Be aware how, by giving a reputation to our check instances, we get very readable check output, each on the terminal and within the IDE:
Having rewritten our two outdated assessments in desk kind, it is now tremendous straightforward so as to add
one other. That is the check for the “x gadgets left” textual content:
Go
{ title: "gadgets left", mannequin: todo.NewList(). Add("One"). Add("Two"). AddCompleted("Three"), selector: "span.todo-count", matches: []string{"2 gadgets left"}, },
Java
new TestCase( "gadgets left", new TodoList() .add("One") .add("Two") .addCompleted("Three"), "span.todo-count", Checklist.of("2 gadgets left")),
And the corresponding change within the html template is:
Go
<span class="todo-count"><sturdy>{{len .ActiveItems}}</sturdy> gadgets left</span>
Java – jmustache
<span class="todo-count"><sturdy>{{activeItemsCount}}</sturdy> gadgets left</span>
The above change within the template requires a supporting technique within the mannequin:
Go
kind Merchandise struct {
Title string
IsCompleted bool
}
kind Checklist struct {
Gadgets []*Merchandise
}
func (l *Checklist) ActiveItems() []*Merchandise {
var consequence []*Merchandise
for _, merchandise := vary l.Gadgets {
if !merchandise.IsCompleted {
consequence = append(consequence, merchandise)
}
}
return consequence
}
Java
public class TodoList {
personal last Checklist<TodoItem> gadgets = new ArrayList<>();
// ...
public lengthy activeItemsCount() {
return gadgets.stream().filter(TodoItem::isActive).depend();
}
}
We have invested just a little effort in our testing infrastructure, in order that including new
check instances is simpler. Within the subsequent part, we’ll see that the necessities
for the subsequent check instances will push us to refine our check infrastructure additional.
Making the desk extra expressive, on the expense of the check code
We’ll now check the “All”, “Energetic” and “Accomplished” navigation hyperlinks at
the underside of the UI (see the image above),
and these rely on which url we’re visiting, which is
one thing that our template has no method to discover out.
At present, all we move to our template is our mannequin, which is a todo-list.
It isn’t appropriate so as to add the at the moment visited url to the mannequin, as a result of that’s
consumer navigation state, not software state.
So we have to move extra data to the template past the mannequin. A straightforward method
is to move a map, which we assemble in our
renderTemplate
operate:
Go
func renderTemplate(mannequin *todo.Checklist, path string) bytes.Buffer { templ := template.Should(template.ParseFiles("index.tmpl")) var buf bytes.Buffer knowledge := map[string]any{ "mannequin": mannequin, "path": path, } err := templ.Execute(&buf, knowledge) if err != nil { panic(err) } return buf }
Java
personal String renderTemplate(String templateName, TodoList mannequin, String path) { var template = Mustache.compiler().compile( new InputStreamReader( getClass().getResourceAsStream(templateName))); var knowledge = Map.of( "mannequin", mannequin, "path", path ); return template.execute(knowledge); }
And correspondingly our check instances desk has another subject:
Go
var testCases = []struct { title string mannequin *todo.Checklist path string selector string matches []string }{ { title: "all todo gadgets are proven", mannequin: todo.NewList(). Add("Foo"). Add("Bar"), selector: "ul.todo-list li", matches: []string{"Foo", "Bar"}, }, // ... the opposite instances { title: "highlighted navigation hyperlink: All", path: "/", selector: "ul.filters a.chosen", matches: []string{"All"}, }, { title: "highlighted navigation hyperlink: Energetic", path: "/lively", selector: "ul.filters a.chosen", matches: []string{"Energetic"}, }, { title: "highlighted navigation hyperlink: Accomplished", path: "/accomplished", selector: "ul.filters a.chosen", matches: []string{"Accomplished"}, }, }
Java
report TestCase(String title, TodoList mannequin, String path, String selector, Checklist<String> matches) { @Override public String toString() { return title; } } public static TestCase[] indexTestCases() { return new TestCase[]{ new TestCase( "all todo gadgets are proven", new TodoList() .add("Foo") .add("Bar"), "/", "ul.todo-list li", Checklist.of("Foo", "Bar")), // ... the earlier instances new TestCase( "highlighted navigation hyperlink: All", new TodoList(), "/", "ul.filters a.chosen", Checklist.of("All")), new TestCase( "highlighted navigation hyperlink: Energetic", new TodoList(), "/lively", "ul.filters a.chosen", Checklist.of("Energetic")), new TestCase( "highlighted navigation hyperlink: Accomplished", new TodoList(), "/accomplished", "ul.filters a.chosen", Checklist.of("Accomplished")), }; }
We discover that for the three new instances, the mannequin is irrelevant;
whereas for the earlier instances, the trail is irrelevant. The Go syntax permits us
to initialize a struct with simply the fields we’re interested by, however Java doesn’t have
an identical function, so we’re pushed to move further data, and this makes the check instances
desk tougher to know.
A developer would possibly take a look at the primary check case and marvel if the anticipated conduct relies upon
on the trail being set to "/"
, and may be tempted so as to add extra instances with
a special path. In the identical method, when studying the
highlighted navigation hyperlink check instances, the developer would possibly marvel if the
anticipated conduct relies on the mannequin being set to an empty todo listing. In that case, one would possibly
be led so as to add irrelevant check instances for the highlighted hyperlink with non-empty todo-lists.
We need to optimize for the time of the builders, so it is worthwhile to keep away from including irrelevant
knowledge to our check case. In Java we would move null
for the
irrelevant fields, however there’s a greater method: we are able to use
the builder sample,
popularized by Joshua Bloch.
We will rapidly write one for the Java TestCase
report this fashion:
Java
report TestCase(String title,
TodoList mannequin,
String path,
String selector,
Checklist<String> matches) {
@Override
public String toString() {
return title;
}
public static last class Builder {
String title;
TodoList mannequin;
String path;
String selector;
Checklist<String> matches;
public Builder title(String title) {
this.title = title;
return this;
}
public Builder mannequin(TodoList mannequin) {
this.mannequin = mannequin;
return this;
}
public Builder path(String path) {
this.path = path;
return this;
}
public Builder selector(String selector) {
this.selector = selector;
return this;
}
public Builder matches(String ... matches) {
this.matches = Arrays.asList(matches);
return this;
}
public TestCase construct() {
return new TestCase(title, mannequin, path, selector, matches);
}
}
}
Hand-coding builders is just a little tedious, however doable, although there are
automated methods to put in writing them.
Now we are able to rewrite our Java check instances with the Builder
, to
obtain higher readability:
Java
public static TestCase[] indexTestCases() { return new TestCase[]{ new TestCase.Builder() .title("all todo gadgets are proven") .mannequin(new TodoList() .add("Foo") .add("Bar")) .selector("ul.todo-list li") .matches("Foo", "Bar") .construct(), // ... different instances new TestCase.Builder() .title("highlighted navigation hyperlink: Accomplished") .path("/accomplished") .selector("ul.filters a.chosen") .matches("Accomplished") .construct(), }; }
So, the place are we with our assessments? At current, they fail for the flawed cause: null-pointer exceptions
as a result of lacking mannequin
and path
values.
With a purpose to get our new check instances to fail for the best cause, specifically that the template does
not but have logic to focus on the proper hyperlink, we should
present default values for mannequin
and path
. In Go, we are able to do that
within the check technique:
Go
func Test_indexTemplate(t *testing.T) {
for _, check := vary testCases {
t.Run(check.title, func(t *testing.T) {
if check.mannequin == nil {
check.mannequin = todo.NewList()
}
buf := renderTemplate(check.mannequin, check.path)
// ... identical as earlier than
})
}
}
In Java, we are able to present default values within the builder:
Java
public static last class Builder { String title; TodoList mannequin = new TodoList(); String path = "/"; String selector; Checklist<String> matches; // ... }
With these adjustments, we see that the final two check instances, those for the highlighted hyperlink Energetic
and Accomplished fail, for the anticipated cause that the highlighted hyperlink doesn’t change:
Go
=== RUN Test_indexTemplate/highlighted_navigation_link:_Active index_template_test.go:82: Error Hint: .../tdd-templates/go/index_template_test.go:82 Error: Not equal: anticipated: "Energetic" precise : "All" === RUN Test_indexTemplate/highlighted_navigation_link:_Completed index_template_test.go:82: Error Hint: .../tdd-templates/go/index_template_test.go:82 Error: Not equal: anticipated: "Accomplished" precise : "All"
Java
IndexTemplateTest > testIndexTemplate(TestCase) > [5] highlighted navigation hyperlink: Energetic FAILED org.opentest4j.AssertionFailedError: Anticipating: <"All"> to be equal to: <"Energetic"> however was not. IndexTemplateTest > testIndexTemplate(TestCase) > [6] highlighted navigation hyperlink: Accomplished FAILED org.opentest4j.AssertionFailedError: Anticipating: <"All"> to be equal to: <"Accomplished"> however was not.
To make the assessments move, we make these adjustments to the template:
Go
<ul class="filters"> <li> <a class="{{ if eq .path "/" }}chosen{{ finish }}" href="#/">All</a> </li> <li> <a class="{{ if eq .path "/lively" }}chosen{{ finish }}" href="#/lively">Energetic</a> </li> <li> <a class="{{ if eq .path "/accomplished" }}chosen{{ finish }}" href="#/accomplished">Accomplished</a> </li> </ul>
Java – jmustache
<ul class="filters"> <li> <a class="{{ #pathRoot }}chosen{{ /pathRoot }}" href="#/">All</a> </li> <li> <a class="{{ #pathActive }}chosen{{ /pathActive }}" href="#/lively">Energetic</a> </li> <li> <a class="{{ #pathCompleted }}chosen{{ /pathCompleted }}" href="#/accomplished">Accomplished</a> </li> </ul>
For the reason that Mustache template language doesn’t enable for equality testing, we should change the
knowledge handed to the template in order that we execute the equality assessments earlier than rendering the template:
Java
personal String renderTemplate(String templateName, TodoList mannequin, String path) { var template = Mustache.compiler().compile( new InputStreamReader( getClass().getResourceAsStream(templateName))); var knowledge = Map.of( "mannequin", mannequin, "pathRoot", path.equals("/"), "pathActive", path.equals("/lively"), "pathCompleted", path.equals("/accomplished") ); return template.execute(knowledge); }
And with these adjustments, all of our assessments now move.
To recap this part, we made the check code just a little bit extra sophisticated, in order that the check
instances are clearer: this can be a superb tradeoff!
Degree 3: testing HTML behaviour
Within the story up to now, we examined the behaviour of the HTML
templates, by checking the construction of the generated HTML.
That is good, however what if we wished to check the behaviour of the HTML
itself, plus any CSS and JavaScript it could use?
The behaviour of HTML by itself is often fairly apparent, as a result of
there may be not a lot of it. The one components that may work together with the
consumer are the anchor (<a>
), <kind>
and
<enter>
components, however the image adjustments utterly when
we add CSS, that may conceal, present, transfer round issues and much extra, and
with JavaScript, that may add any behaviour to a web page.
In an software that’s primarily rendered server-side, we count on
that almost all behaviour is carried out by returning new HTML with a
round-trip to the consumer, and this may be examined adequately with the
strategies we have seen up to now, however what if we wished to hurry up the
software behaviour with a library equivalent to HTMX? This library works by means of particular
attributes which can be added to components so as to add Ajax behaviour. These
attributes are in impact a DSL that we would need to
check.
How can we check the mixture of HTML, CSS and JavaScript in
a unit check?
Testing HTML, CSS and JavaScript requires one thing that is ready to
interpret and execute their behaviours; in different phrases, we’d like a
browser! It’s customary to make use of headless browsers in end-to-end assessments;
can we use them for unitary assessments as an alternative? I feel that is potential,
utilizing the next strategies, though I need to admit I’ve but to attempt
this on an actual venture.
We’ll use the Playwright
library, that’s obtainable for each Go and
Java. The assessments we
are going to put in writing might be slower, as a result of we must wait a number of
seconds for the headless browser to start out, however will retain a few of the
necessary traits of unit assessments, primarily that we’re testing
simply the HTML (and any related CSS and JavaScript), in isolation from
every other server-side logic.
Persevering with with the TodoMVC
instance, the subsequent factor we would need to check is what occurs when the
consumer clicks on the checkbox of a todo merchandise. What we might wish to occur is
that:
- A POST name to the server is made, in order that the appliance is aware of
that the state of a todo merchandise has modified - The server returns new HTML for the dynamic a part of the web page,
specifically all the part with class “todoapp”, in order that we are able to present the
new state of the appliance together with the depend of remaining “lively”
gadgets (see the template above) - The web page replaces the outdated contents of the “todoapp” part with
the brand new ones.
Loading the web page within the Playwright browser
We begin with a check that may simply load the preliminary HTML. The check
is just a little concerned, so I present the whole code right here, after which I’ll
remark it little by little.
Go
func Test_toggleTodoItem(t *testing.T) { // render the preliminary HTML mannequin := todo.NewList(). Add("One"). Add("Two") initialHtml := renderTemplate("index.tmpl", mannequin, "/") // open the browser web page with Playwright web page := openPage() defer web page.Shut() logActivity(web page) // stub community calls err := web page.Route("**", func(route playwright.Route) { if route.Request().URL() == "http://localhost:4567/index.html" { // serve the preliminary HTML stubResponse(route, initialHtml.String(), "textual content/html") } else { // keep away from surprising requests panic("surprising request: " + route.Request().URL()) } }) if err != nil { t.Deadly(err) } // load preliminary HTML within the web page response, err := web page.Goto("http://localhost:4567/index.html") if err != nil { t.Deadly(err) } if response.Standing() != 200 { t.Fatalf("surprising standing: %d", response.Standing()) } }
Java
public class IndexBehaviourTest { static Playwright playwright; static Browser browser; @BeforeAll static void launchBrowser() { playwright = Playwright.create(); browser = playwright.chromium().launch(); } @AfterAll static void closeBrowser() { playwright.shut(); } @Take a look at void toggleTodoItem() { // Render the preliminary html TodoList mannequin = new TodoList() .add("One") .add("Two"); String initialHtml = renderTemplate("/index.tmpl", mannequin, "/"); attempt (Web page web page = browser.newPage()) { logActivity(web page); // stub community calls web page.route("**", route -> { if (route.request().url().equals("http://localhost:4567/index.html")) { // serve the preliminary HTML route.fulfill(new Route.FulfillOptions() .setContentType("textual content/html") .setBody(initialHtml)); } else { // we do not need surprising calls fail(String.format("Sudden request: %s %s", route.request().technique(), route.request().url())); } }); // load preliminary html web page.navigate("http://localhost:4567/index.html"); } } }
At the beginning of the check, we initialize the mannequin with two todo
gadgets “One” and “Two”, then we render the template as earlier than:
Go
mannequin := todo.NewList(). Add("One"). Add("Two") initialHtml := renderTemplate("index.tmpl", mannequin, "/")
Java
TodoList mannequin = new TodoList() .add("One") .add("Two"); String initialHtml = renderTemplate("/index.tmpl", mannequin, "/");
Then we open the Playwright “web page”, which can begin a headless
browser
Go
web page := openPage() defer web page.Shut() logActivity(web page)
Java
attempt (Web page web page = browser.newPage()) { logActivity(web page);
The openPage
operate in Go returns a Playwright
Web page
object,
Go
func openPage() playwright.Web page { pw, err := playwright.Run() if err != nil { log.Fatalf("couldn't begin playwright: %v", err) } browser, err := pw.Chromium.Launch() if err != nil { log.Fatalf("couldn't launch browser: %v", err) } web page, err := browser.NewPage() if err != nil { log.Fatalf("couldn't create web page: %v", err) } return web page }
and the logActivity
operate offers suggestions on what
the web page is doing
Go
func logActivity(web page playwright.Web page) { web page.OnRequest(func(request playwright.Request) { log.Printf(">> %s %sn", request.Technique(), request.URL()) }) web page.OnResponse(func(response playwright.Response) { log.Printf("<< %d %sn", response.Standing(), response.URL()) }) web page.OnLoad(func(web page playwright.Web page) { log.Println("Loaded: " + web page.URL()) }) web page.OnConsole(func(message playwright.ConsoleMessage) { log.Println("! " + message.Textual content()) }) }
Java
personal void logActivity(Web page web page) { web page.onRequest(request -> System.out.printf(">> %s %spercentn", request.technique(), request.url())); web page.onResponse(response -> System.out.printf("<< %s %spercentn", response.standing(), response.url())); web page.onLoad(page1 -> System.out.println("Loaded: " + page1.url())); web page.onConsoleMessage(consoleMessage -> System.out.println("! " + consoleMessage.textual content())); }
Then we stub all community exercise that the web page would possibly attempt to do
Go
err := web page.Route("**", func(route playwright.Route) {
if route.Request().URL() == "http://localhost:4567/index.html" {
// serve the preliminary HTML
stubResponse(route, initialHtml.String(), "textual content/html")
} else {
// keep away from surprising requests
panic("surprising request: " + route.Request().URL())
}
})
Java
// stub community calls
web page.route("**", route -> {
if (route.request().url().equals("http://localhost:4567/index.html")) {
// serve the preliminary HTML
route.fulfill(new Route.FulfillOptions()
.setContentType("textual content/html")
.setBody(initialHtml));
} else {
// we do not need surprising calls
fail(String.format("Sudden request: %s %s", route.request().technique(), route.request().url()));
}
});
and we ask the web page to load the preliminary HTML
Go
response, err := web page.Goto("http://localhost:4567/index.html")
Java
web page.navigate("http://localhost:4567/index.html");
With all this equipment in place, we run the check; it succeeds and
it logs the stubbed community exercise on normal output:
Go
=== RUN Test_toggleTodoItem >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html Loaded: http://localhost:4567/index.html --- PASS: Test_toggleTodoItem (0.89s)
Java
IndexBehaviourTest > toggleTodoItem() STANDARD_OUT >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html Loaded: http://localhost:4567/index.html IndexBehaviourTest > toggleTodoItem() PASSED
So with this check we at the moment are in a position to load arbitrary HTML in a
headless browser. Within the subsequent sections we’ll see find out how to simulate consumer
interplay with components of the web page, and observe the web page’s
behaviour. However first we have to clear up an issue with the dearth of
identifiers in our area mannequin.
Figuring out todo gadgets
Now we need to click on on the “One” checkbox. The issue we now have is
that at current, we now have no method to determine particular person todo gadgets, so
we introduce an Id
subject within the todo merchandise:
Go – up to date mannequin with Id
kind Merchandise struct { Id int Title string IsCompleted bool } func (l *Checklist) AddWithId(id int, title string) *Checklist { merchandise := Merchandise{ Id: id, Title: title, } l.Gadgets = append(l.Gadgets, &merchandise) return l } // Add creates a brand new todo.Merchandise with a random Id func (l *Checklist) Add(title string) *Checklist { merchandise := Merchandise{ Id: generateRandomId(), Title: title, } l.Gadgets = append(l.Gadgets, &merchandise) return l } func generateRandomId() int { return abs(rand.Int()) }
Java – up to date mannequin with Id
public class TodoList { personal last Checklist<TodoItem> gadgets = new ArrayList<>(); public TodoList add(String title) { gadgets.add(new TodoItem(generateRandomId(), title, false)); return this; } public TodoList addCompleted(String title) { gadgets.add(new TodoItem(generateRandomId(), title, true)); return this; } public TodoList add(int id, String title) { gadgets.add(new TodoItem(id, title, false)); return this; } personal static int generateRandomId() { return new Random().nextInt(0, Integer.MAX_VALUE); } } public report TodoItem(int id, String title, boolean isCompleted) { public boolean isActive() { return !isCompleted; } }
And we replace the mannequin in our check so as to add specific Ids
Go – including Id within the check knowledge
func Test_toggleTodoItem(t *testing.T) { // render the preliminary HTML mannequin := todo.NewList(). AddWithId(101, "One"). AddWithId(102, "Two") initialHtml := renderTemplate("index.tmpl", mannequin, "/") // ... }
Java – including Id within the check knowledge
@Take a look at void toggleTodoItem() { // Render the preliminary html TodoList mannequin = new TodoList() .add(101, "One") .add(102, "Two"); String initialHtml = renderTemplate("/index.tmpl", mannequin, "/"); }
We at the moment are prepared to check consumer interplay with the web page.
Clicking on a todo merchandise
We need to simulate consumer interplay with the HTML web page. It may be
tempting to proceed to make use of CSS selectors to determine the particular
checkbox that we need to click on, however there’s a greater method: there’s a
consensus amongst front-end builders that one of the simplest ways to check
interplay with a web page is to make use of it
the identical method that customers do. As an example, you do not search for a
button by means of a CSS locator equivalent to button.purchase
; as an alternative,
you search for one thing clickable with the label “Purchase”. In apply,
this implies figuring out elements of the web page by means of their
ARIA roles.
To this finish, we add code to our check to search for a checkbox labelled
“One”:
Go
func Test_toggleTodoItem(t *testing.T) { // ... // click on on the "One" checkbox checkbox := web page.GetByRole(*playwright.AriaRoleCheckbox, playwright.PageGetByRoleOptions{Title: "One"}) if err := checkbox.Click on(); err != nil { t.Deadly(err) } }
Java
@Take a look at void toggleTodoItem() { // ... // click on on the "One" checkbox var checkbox = web page.getByRole(AriaRole.CHECKBOX, new Web page.GetByRoleOptions().setName("One")); checkbox.click on(); } }
We run the check, and it fails:
Go
>> GET http://localhost:4567/index.html
<< 200 http://localhost:4567/index.html
Loaded: http://localhost:4567/index.html
--- FAIL: Test_toggleTodoItem (32.74s)
index_behaviour_test.go:50: playwright: timeout: Timeout 30000ms exceeded.
Java
IndexBehaviourTest > toggleTodoItem() STANDARD_OUT
>> GET http://localhost:4567/index.html
<< 200 http://localhost:4567/index.html
Loaded: http://localhost:4567/index.html
IndexBehaviourTest > toggleTodoItem() FAILED
com.microsoft.playwright.TimeoutError: Error {
message="hyperlink the label to the checkbox correctly:
generated HTML with dangerous accessibility
<li>
<div class="view">
<enter class="toggle" kind="checkbox">
<label>One</label>
<button class="destroy"></button>
</div>
</li>
We repair it by utilizing the for
attribute within the
template,
index.tmpl – Go
<li>
<div class="view">
<enter id="checkbox-{{.Id}}" class="toggle" kind="checkbox">
<label for="checkbox-{{.Id}}">{{.Title}}</label>
<button class="destroy"></button>
</div>
</li>
index.tmpl – Java
<li>
<div class="view">
<enter id="checkbox-{{ id }}" class="toggle" kind="checkbox">
<label for="checkbox-{{ id }}">{{ title }}</label>
<button class="destroy"></button>
</div>
</li>
In order that it generates correct, accessible HTML:
generated HTML with higher accessibility
<li>
<div class="view">
<enter id="checkbox-101" class="toggle" kind="checkbox">
<label for="checkbox-101">One</label>
<button class="destroy"></button>
</div>
</li>
We run once more the check, and it passes.
On this part we noticed how testing the HTML in the identical was as customers
work together with it led us to make use of ARIA roles, which led to bettering
accessibility of our generated HTML. Within the subsequent part, we are going to see
find out how to check that the clicking on a todo merchandise triggers a distant name to the
server, that ought to end in swapping part of the present HTML with
the HTML returned by the XHR name.
Spherical-trip to the server
Now we are going to prolong our check. We inform the check that if name to
POST /toggle/101
is acquired, it ought to return some
stubbed HTML.
Go
} else if route.Request().URL() == "http://localhost:4567/toggle/101" && route.Request().Technique() == "POST" { // we count on {that a} POST /toggle/101 request is made after we click on on the "One" checkbox const stubbedHtml = ` <part class="todoapp"> <p>Stubbed html</p> </part>` stubResponse(route, stubbedHtml, "textual content/html")
Java
} else if (route.request().url().equals("http://localhost:4567/toggle/101") && route.request().technique().equals("POST")) { // we count on {that a} POST /toggle/101 request is made after we click on on the "One" checkbox String stubbedHtml = """ <part class="todoapp"> <p>Stubbed html</p> </part> """; route.fulfill(new Route.FulfillOptions() .setContentType("textual content/html") .setBody(stubbedHtml));
And we stub the loading of the HTMX library, which we load from a
native file:
Go
} else if route.Request().URL() == "https://unpkg.com/htmx.org@1.9.12" {
// serve the htmx library
stubResponse(route, readFile("testdata/htmx.min.js"), "software/javascript")
Go
} else if (route.request().url().equals("https://unpkg.com/htmx.org@1.9.12")) {
// serve the htmx library
route.fulfill(new Route.FulfillOptions()
.setContentType("textual content/html")
.setBody(readFile("/htmx.min.js")));
Lastly, we add the expectation that, after we click on the checkbox,
the part of the HTML that incorporates a lot of the software is
reloaded.
Go
// click on on the "One" checkbox checkbox := web page.GetByRole(*playwright.AriaRoleCheckbox, playwright.PageGetByRoleOptions{Title: "One"}) if err := checkbox.Click on(); err != nil { t.Deadly(err) } // examine that the web page has been up to date doc := parseHtml(t, content material(t, web page)) components := doc.Discover("physique > part.todoapp > p") assert.Equal(t, "Stubbed html", components.Textual content(), should(web page.Content material()))
java
// click on on the "One" checkbox var checkbox = web page.getByRole(AriaRole.CHECKBOX, new Web page.GetByRoleOptions().setName("One")); checkbox.click on(); // examine that the web page has been up to date var doc = parseHtml(web page.content material()); var components = doc.choose("physique > part.todoapp > p"); assertThat(components.textual content()) .describedAs(web page.content material()) .isEqualTo("Stubbed html");
We run the check, and it fails, as anticipated. With a purpose to perceive
why precisely it fails, we add to the error message the entire HTML
doc.
Go
assert.Equal(t, "Stubbed html", components.Textual content(), should(web page.Content material()))
Java
assertThat(components.textual content())
.describedAs(web page.content material())
.isEqualTo("Stubbed html");
The error message may be very verbose, however we see that the explanation it
fails is that we do not see the stubbed HTML within the output. This implies
that the web page didn’t make the anticipated XHR name.
Go – Java is analogous
--- FAIL: Test_toggleTodoItem (2.75s) === RUN Test_toggleTodoItem >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html Loaded: http://localhost:4567/index.html index_behaviour_test.go:67: Error Hint: .../index_behaviour_test.go:67 Error: Not equal: anticipated: "Stubbed html" precise : "" ... Take a look at: Test_toggleTodoItem Messages: <!DOCTYPE html><html lang="en"><head> <meta charset="utf-8"> <meta title="viewport" content material="width=device-width, initial-scale=1"> <title>Template • TodoMVC</title> <script src="https://unpkg.com/htmx.org@1.9.12"></script> <physique> <part class="todoapp"> ... <li class=""> <div class="view"> <enter id="checkbox-101" class="toggle" kind="checkbox"> <label for="checkbox-101">One</label> <button class="destroy"></button> </div> </li> ...
We will make this check move by altering the HTML template to make use of HTMX
to make an XHR name again to the server. First we load the HTMX
library:
index.tmpl
<title>Template • TodoMVC</title>
<script src="https://unpkg.com/htmx.org@1.9.12"></script>
Then we add the HTMX attributes to the checkboxes:
index.tmpl
<enter data-hx-post="/toggle/{{.Id}}" data-hx-target="part.todoapp" id="checkbox-{{.Id}}" class="toggle" kind="checkbox">
The data-hx-post
annotation will make HTMX do a POST
name to the desired url. The data-hx-target
tells HTMX
to repeat the HTML returned by the decision, to the aspect specified by the
part.todoapp
CSS locator.
We run once more the check, and it nonetheless fails!
Go – Java is analogous
--- FAIL: Test_toggleTodoItem (2.40s) === RUN Test_toggleTodoItem >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html >> GET https://unpkg.com/htmx.org@1.9.12 << 200 https://unpkg.com/htmx.org@1.9.12 Loaded: http://localhost:4567/index.html >> POST http://localhost:4567/toggle/101 << 200 http://localhost:4567/toggle/101 index_behaviour_test.go:67: Error Hint: .../index_behaviour_test.go:67 Error: Not equal: anticipated: "Stubbed html" precise : "" ... Take a look at: Test_toggleTodoItem Messages: <!DOCTYPE html><html lang="en"><head> <meta charset="utf-8"> <meta title="viewport" content material="width=device-width, initial-scale=1"> <title>Template • TodoMVC</title> <script src="https://unpkg.com/htmx.org@1.9.12"></script> ... <physique> <part class="todoapp"><part class="todoapp"> <p>Stubbed html</p> </part></part> ... </physique></html>
The log traces present that the POST name occurred as anticipated, however
examination of the error message exhibits that the HTML construction we
anticipated just isn’t there: we now have a part.todoapp
nested
inside one other. Which means we’re not utilizing the HTMX annotations
appropriately, and exhibits why this type of check will be worthwhile. We add the
lacking annotation
index.tmpl
<enter
data-hx-post="/toggle/{{.Id}}"
data-hx-target="part.todoapp"
data-hx-swap="outerHTML"
id="checkbox-{{.Id}}"
class="toggle"
kind="checkbox">
The default behaviour of HTMX is to interchange the inside HTML of the
goal aspect. The data-hx-swap="outerHTML"
annotation
tells HTMX to interchange the outer HTML as an alternative.
and we check once more, and this time it passes!
Go
=== RUN Test_toggleTodoItem >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html >> GET https://unpkg.com/htmx.org@1.9.12 << 200 https://unpkg.com/htmx.org@1.9.12 Loaded: http://localhost:4567/index.html >> POST http://localhost:4567/toggle/101 << 200 http://localhost:4567/toggle/101 --- PASS: Test_toggleTodoItem (1.39s)
Java
IndexBehaviourTest > toggleTodoItem() STANDARD_OUT >> GET http://localhost:4567/index.html << 200 http://localhost:4567/index.html >> GET https://unpkg.com/htmx.org@1.9.12 << 200 https://unpkg.com/htmx.org@1.9.12 Loaded: http://localhost:4567/index.html >> POST http://localhost:4567/toggle/101 << 200 http://localhost:4567/toggle/101 IndexBehaviourTest > toggleTodoItem() PASSED
On this part we noticed find out how to write a check for the behaviour of our
HTML that, whereas utilizing the sophisticated equipment of a headless browser,
nonetheless feels extra like a unit check than an integration check. It’s in
reality testing simply an HTML web page with any related CSS and JavaScript,
in isolation from different elements of the appliance equivalent to controllers,
companies or repositories.
The check prices 2-3 seconds of ready time for the headless browser to return up, which is often an excessive amount of for a unit check; nevertheless, like a unit check, it is vitally steady, as it’s not flaky, and its failures are documented with a comparatively clear error message.
Bonus stage: Stringly asserted
Esko Luontola, TDD skilled and creator of the net course tdd.mooc.fi, urged another to testing HTML with CSS selectors: the thought is to remodel HTML right into a human-readable canonical kind.
Let’s take for instance this snippet of generated HTML:
<ul class="todo-list"> <li class=""> <div class="view"> <enter id="checkbox-100" class="toggle" kind="checkbox"> <label for="checkbox-100">One</label> <button class="destroy"></button> </div> </li> <li class=""> <div class="view"> <enter id="checkbox-200" class="toggle" kind="checkbox"> <label for="checkbox-200">Two</label> <button class="destroy"></button> </div> </li> <li class="accomplished"> <div class="view"> <enter id="checkbox-300" class="toggle" kind="checkbox"> <label for="checkbox-300">Three</label> <button class="destroy"></button> </div> </li> </ul>
We may visualize the above HTML by:
- deleting all HTML tags
- decreasing each sequence of whitespace characters to a single clean
to reach at:
One Two Three
This, nevertheless, removes an excessive amount of of the HTML construction to be helpful. As an example, it doesn’t allow us to distinguish between lively and accomplished gadgets. Some HTML aspect signify seen content material: as an illustration
<enter worth="foo" />
exhibits a textual content field with the phrase “foo” that is a crucial a part of the method we understand HTML. To visualise these components, Esko suggests so as to add a data-test-icon
attribute that provides some textual content for use instead of the aspect when visualizing it for testing. With this,
<enter worth="foo" data-test-icon="[foo]" />
the enter aspect is visualized as [foo]
, with the sq. brackets hinting that the phrase “foo” sits inside an editable textual content field. Now if we add test-icons to our HTML template,
Go — Java is analogous
<ul class="todo-list"> {{ vary .mannequin.AllItems }} <li class="{{ if .IsCompleted }}accomplished{{ finish }}"> <div class="view"> <enter data-hx-post="/toggle/{{ .Id }}" data-hx-target="part.todoapp" data-hx-swap="outerHTML" id="checkbox-{{ .Id }}" class="toggle" kind="checkbox" data-test-icon="{{ if .IsCompleted }}✅{{ else }}⬜{{ finish }}"> <label for="checkbox-{{ .Id }}">{{ .Title }}</label> <button class="destroy" data-test-icon="❌️"></button> </div> </li> {{ finish }} </ul>
we are able to assert in opposition to its canonical visible illustration like this:
Go
func Test_visualize_html_example(t *testing.T) { mannequin := todo.NewList(). Add("One"). Add("Two"). AddCompleted("Three") buf := renderTemplate("todo-list.tmpl", mannequin, "/") anticipated := ` ⬜ One ❌️ ⬜ Two ❌️ ✅ Three ❌️ ` assert.Equal(t, normalizeWhitespace(anticipated), visualizeHtml(buf.String())) }
Java
@Take a look at void visualize_html_example() { var mannequin = new TodoList() .add("One") .add("Two") .addCompleted("Three"); var html = renderTemplate("/todo-list.tmpl", mannequin, "/"); assertThat(visualizeHtml(html)) .isEqualTo(normalizeWhitespace(""" ⬜ One ❌️ ⬜ Two ❌️ ✅ Three ❌️ """)); }
Right here is Esko Luontola’s Java implementation of the 2 capabilities that make this potential, and my translation to Go of his code.
Go
func visualizeHtml(html string) string span func normalizeWhitespace(s string) string { return strings.TrimSpace(replaceAll(s, "s+", " ")) } func replaceAll(src, regex, repl string) string { re := regexp.MustCompile(regex) return re.ReplaceAllString(src, repl) }
Java
public static String visualizeHtml(String html) code public static String normalizeWhitespace(String s) { return s.replaceAll("s+", " ").trim(); }
On this part, we now have seen a way for asserting HTML content material that’s a substitute for the CSS selector-based approach utilized in the remainder of the article. Esko Luontola has reported nice success with it, and I hope readers have success with it too!
This system of asserting in opposition to massive, sophisticated knowledge constructions equivalent to HTML pages by decreasing them to a canonical string model has no title that I do know of. Martin Fowler urged “stringly asserted”, and from his suggestion comes the title of this part.