From html5validator¶

html5validator wraps the Nu Html Checker (vnu.jar), the reference HTML5 conformance validator behind validator.nu, as a Python command and a small Validator class. It is the standard way to run conformance checks from Python – in a CI step or a pre-commit hook – and it flags the authoring mistakes that parse cleanly but violate the standard: a missing img alt, obsolete markup, a duplicate id, an undefined ARIA role. It shells out to a bundled JVM for every run.

turbohtml covers the same checks natively in turbohtml.conformance. check_html() parses a string and returns a ConformanceReport – a valid verdict plus every ConformanceMessage classified by severity – with no subprocess, no JVM, and no jar to install, so a check runs in microseconds instead of a JVM launch.

turbohtml vs html5validator¶

Dimension	turbohtml	html5validator
Engine	Native C, in-process	`vnu.jar` over a JVM subprocess
Rule coverage	Alt text, obsolete markup, duplicate ids, ARIA roles, heading and section structure, document title and language	The full Nu Html Checker rule set
Result	Severity-classified messages with a stable `code` and source position	vnu text, JSON, or gnu output parsed from the subprocess
Runtime dependency	None	A Java runtime plus the bundled jar
Performance	Microseconds per document, no process spawn	Seconds per run, dominated by JVM startup
Typing	Fully annotated, `py.typed`	Untyped

The rule sets differ in breadth: vnu enforces the whole standard, while turbohtml checks the high-value authoring rules listed above. For those rules the two agree on the verdict, which the differential suite pins by cross-checking turbohtml against vnu.jar itself.

Feature overlap¶

html5validator.Validator().validate([path]), which returns a count of errors, maps to not check_html(markup).valid – a boolean verdict from the report’s error findings.
vnu’s per-message type/subType (error, info/warning) maps to ConformanceMessage.severity ("error", "warning", "info"), and its lastLine/firstColumn to the message line and column.
Reading vnu’s JSON output to gate a build maps to the report’s errors, warnings, and infos views.
Validating an already-parsed tree maps to turbohtml.conformance.check(), which skips the reparse.

from turbohtml.conformance import check_html

report = check_html(
    "<!DOCTYPE html><html lang=en><head><title>Page</title></head>"
    "<body><img src=logo.png><font>old</font></body></html>"
)
print(report.valid)
for message in report.errors:
    print(message.code, message.line, message.message)

False
img-missing-alt 1 img element has no alt attribute
obsolete-element 1 the font element is obsolete

input	turbohtml	html5validator
check page (3 KiB)	154 µs	960 ms (6245x)
check page (27 KiB)	1.34 ms	1.03 s (769x)