XPath-syntaks-jukseark

XPath navigates an XML or HTML document as a tree. Every step has the form axis::node-test[predicate], and most of it has a short form: //div[@class="card"]/a/@href reads “the href of every a inside any div with class card”.

⚠️ Browsers, Selenium and most scrapers implement XPath 1.0. ends-with(), matches(), lower-case() and replace() are 2.0+ and only work in tools like xmllint, Saxon or lxml with an XSLT 2.0 processor. Workarounds for 1.0 are given below.

🧭 Path Syntax

XPath	Selects
`/`	From the document root — an absolute path
`//`	Anywhere in the document, at any depth
`.`	The current (context) node
`..`	The parent of the context node
`@attr`	An attribute
`*`	Any element
`@*`	Any attribute
`text()`	The text nodes of an element
`node()`	Any node at all, text and comments included
`[n]`	The nth match — 1-based, not 0-based
`[last()]`	The last match
`a \| b`	The union of two expressions

/html/body            (: absolute: the body of the page :)
//text()              (: every text node :)
./div/b               (: relative to the context node :)
(//p)[2]              (: the second p in the whole document :)
//p[2]                (: every p that is the second p of its parent :)

🏷 Elements & Attributes

XPath	Description	CSS
`//E`	Every `<E>` element	`E`
`(//E)[2]`	The second `<E>` in the document	—
`//E[@A]`	`<E>` carrying attribute `A`	`E[A]`
`//E[@A="t"]`	`<E>` where `A` is exactly `t`	`E[A='t']`
`//E[contains(@A,"t")]`	`<E>` where `A` contains `t`	`E[A*='t']`
`//E[starts-with(@A,"t")]`	`<E>` where `A` starts with `t`	`E[A^='t']`
`//E[substring(@A, string-length(@A)-2)="t"]`	`<E>` where `A` ends with `t` (1.0 trick)	`E[A$='t']`
`//E[ends-with(@A,"t")]`	The same, XPath 2.0+ only	`E[A$='t']`
`//E[matches(@A,"r")]`	`A` matches regex `r` (2.0+)	—
`//E/@A`	The value of attribute `A`	—
`//*/@A`	Attribute `A` of any element	—
`//E[@A2="t"]/@A1`	`A1` where `A2` is `t`	—
`//E1[@id="I1"] \| //E2[@id="I2"]`	Either element	`E1#I1, E2#I2`
`//E[@id="I1" or @id="I2"]`	`<E>` with either id	`E#I1, E#I2`

🆔 Id, Name, Class & Language

XPath	Description	CSS
`//*[@id="I"]`	The element with id `I`	`#I`
`//E[@id="I"]`	`<E>` with id `I`	`E#I`
`//*[@name="N"]`	Any element named `N`	`[name='N']`
`//*[@name="N"][@value="v"]`	Named `N` with value `v`	`[name='N'][value='v']`
`//*[contains(concat(" ", normalize-space(@class), " "), " C ")]`	Element with class `C` — the padding avoids matching `C2`	`.C`
`//E[contains(concat(" ", normalize-space(@class), " "), " C ")]`	`<E>` with class `C`	`E.C`
`//E[@lang="L" or starts-with(@lang, concat("L","-"))]`	`<E>` in language `L` or a subcode	`E[lang\|='L']`

📝 Text & Links

XPath	Description	CSS
`//*[.="t"]`	Element whose whole text is exactly `t`	—
`//E[contains(text(),"t")]`	`<E>` with a text node containing `t`	—
`//E[contains(.,"t")]`	`<E>` whose text, descendants included, contains `t`	—
`//E[normalize-space()="t"]`	Exact text, ignoring surrounding whitespace	—
`//a`	Every link	`a`
`//a[.="t"]`	The link labelled exactly `t`	—
`//a[@href="url"]`	The link pointing at `url`	`a[href='url']`
`//a[.="t"]/@href`	The URL behind the label `t`	—
`//a[starts-with(@href,"https")]`	Secure links only	`a[href^='https']`

👪 Parents & Children

XPath	Description	CSS
`//E/*[1]`	First child of `<E>`	`E > *:first-child`
`//E/*[last()]`	Last child of `<E>`	`E > *:last-child`
`//E[1]`	Each `<E>` that is the first of its type in its parent	`E:first-of-type`
`//E[last()]`	Last `<E>` of its type	`E:last-of-type`
`//E[2]`	Second `<E>` of its type	`E:nth-of-type(2)`
`//*[2][name()="E"]`	Second child, if it is an `<E>`	`E:nth-child(2)`
`//E[last()-1]`	Second-to-last `<E>` of its type	`E:nth-last-of-type(2)`
`//E/..`	The parent of `<E>`	—
`//div/form/parent::*`	Every element that has a form child	`div:has(> form)`
`//*[@id="I"]//E`	`<E>` anywhere under id `I`	`#I E`
`//E[count(*)=0]`	`<E>` with no element children	`E:empty`
`//E[count(*)=1]`	`<E>` with exactly one child	—
`//E[count(preceding-sibling::)+count(following-sibling::)=0]`	`<E>` that is an only child	`E:only-child`
`//E[count(../E)=1]`	`<E>` with no `<E>` siblings	`E:only-of-type`
`//E1[E2 and not(*[not(self::E2)])]`	`<E1>` whose only children are `<E2>`	—
`//E[position() mod N = M + 1]`	Every Nth element, offset by M	`E:nth-child(Nn+M)`

👥 Siblings

XPath	Description	CSS
`//E2/following-sibling::E1`	Every `<E1>` after a sibling `<E2>`	`E2 ~ E1`
`//E2/following-sibling::*[1][name()="E1"]`	`<E1>` immediately after `<E2>`	`E2 + E1`
`//E2/following-sibling::*[2][name()="E1"]`	`<E1>` two siblings after `<E2>`	`E2 + * + E1`
`//E/following-sibling::*[1]`	Whatever follows `<E>` directly	`E + *`
`//E2/preceding-sibling::E1`	Every `<E1>` before a sibling `<E2>`	—
`//E2/preceding-sibling::*[1][name()="E1"]`	`<E1>` immediately before `<E2>`	—
`//E/preceding-sibling::*[1]`	Whatever precedes `<E>` directly	—

📊 Tables & Form State

XPath	Description	CSS
`//*[@id="T"]//tr[3]//td[2]`	Third row, second cell	`#T tr:nth-child(3) td:nth-child(2)`
`//td[preceding-sibling::td="t"]`	The cell right after the one reading `t`	—
`//td[preceding-sibling::td[contains(.,"t")]]`	Cells following one containing `t`	`td:has(~ td)`
`//table[count(tr)=1 and count(tr/td)=2]`	Tables with one row and two columns	—
`//table/tr[last()]`	The last row of a table	`table tr:last-child`
`//E[@disabled]`	A disabled control	`E:disabled`
`//E[not(@disabled)]`	A control that is not disabled	`E:enabled`
`//*[@checked]`	A checked checkbox or radio	`:checked`

🧮 Functions

Strings

Function	Returns
`contains(haystack, needle)`	True when the first string contains the second
`starts-with(haystack, needle)`	True when it starts with it
`concat(s1, s2, …)`	The strings joined together
`normalize-space(s)`	The string with runs of whitespace collapsed and trimmed
`string-length(s)`	The number of characters
`substring(s, start [, length])`	A slice — 1-based
`substring-before(haystack, needle)`	Everything before the first occurrence
`substring-after(haystack, needle)`	Everything after it
`translate(s, from, to)`	Characters mapped one by one — the 1.0 way to fold case

Numbers, logic & nodes

Function	Returns
`count(node-set)`	How many nodes matched
`sum(node-set)`	The sum of their numeric values
`position()`	The index of the context node within its step
`last()`	The size of the context — so `[last()]` is the final node
`name([node-set])`	The qualified name of the first node
`namespace-uri([node-set])`	Its namespace URI
`lang(s)`	True when the node is in that language
`not(expr)`	The negation
`true()` / `false()`	Boolean constants
`boolean(expr)` / `string(o)` / `number(o)`	Type conversions
`ceiling(n)` / `floor(n)` / `round(n)`	Rounding

💡 Case-insensitive matching in XPath 1.0: //E[contains(translate(@A,"ABCDEFGHIJKLMNOPQRSTUVWXYZ","abcdefghijklmnopqrstuvwxyz"), "t")]

🧭 Axes

Axis	Short form	Selects
`child`	(default)	The direct children of the context node
`descendant`	`//`	Every node below the context node
`descendant-or-self`	`//`	The context node and everything below it
`parent`	`..`	The single parent node
`ancestor`	—	Parent, grandparent, up to the root
`ancestor-or-self`	—	The same, including the context node
`following`	—	Everything after the node in document order, except descendants
`following-sibling`	—	Later siblings under the same parent
`preceding`	—	Everything before it, except its ancestors
`preceding-sibling`	—	Earlier siblings under the same parent
`attribute`	`@`	The attributes of the node
`self`	`.`	The context node itself

//td[.="Total"]/ancestor::table          (: the table a cell belongs to :)
//h2[.="Specs"]/following-sibling::ul[1] (: the list after a heading :)
//input[@name="q"]/ancestor-or-self::form (: the form wrapping a field :)

🧠 Worked Examples

//hr[@class="edge" and position()=1]
(: the first hr of class 'edge' under each parent :)

//table[parent::div[@class="pad"] and not(@id)]//a
(: links in unnamed tables inside a padded div :)

/html/body/div/*[preceding-sibling::h4]
(: whatever comes after an h4 :)

//tr/td[font[@class="head" and text()="TRACK"]]
(: cells whose font child says TRACK :)

//*[count(*)=3]
(: elements with exactly three children :)

//rdf:Seq/rdf:li/em:id
(: namespaced elements — the prefixes must be bound by the host tool :)

//var | //acronym
(: a union of two node sets :)

🛠 Running XPath

// browser devtools console
$x('//a[contains(@href, "docs")]');

// plain DOM API
document.evaluate('//h1', document, null, XPathResult.ANY_TYPE, null).iterateNext();

# from the shell, over XML or HTML
xmllint --xpath '//book/title/text()' catalog.xml
xmllint --html --xpath '//a/@href' page.html 2>/dev/null

from lxml import html
tree = html.parse('page.html')
tree.xpath('//div[@class="card"]/a/@href')

# Selenium and Playwright both accept XPath locators
driver.find_element(By.XPATH, '//button[normalize-space()="Save"]')
page.locator('xpath=//button[normalize-space()="Save"]').click()

⚠️ Notes

Indexes start at 1. [0] never matches anything.
//E[1] is per-parent, (//E)[1] is document-wide — the single most common XPath bug.
Namespaced documents need prefixes registered with the host library; in XPath 1.0 a common workaround is //*[local-name()="title"].
// is expensive on large documents; anchor the path when you can.
Prefer text matching with normalize-space() — HTML is full of stray whitespace and line breaks.

📚 Resources

Credit to the original author: https://gist.github.com/jmaccabee/550a0b9fcfdc7e6b170cd34c6ec7bc56