A command-line tool to find broken links in your static site.
Fast. docs.sentry.io produces
1.1 GB of HTML files. All alternatives we tried were slower
than hyperlink on this site.
hyperlink handles this amount of data in 4 seconds on a MacBook Pro 2018.
Pay for what you need. By default, hyperlink checks for hard 404s in
internal links only. Anything beyond that is opt-in. See Options
for a list of features to enable.
Maps back errors to source files. If your static site was created from
Markdown files, hyperlink can try to find the original broken link by
fuzzy-matching the content around it. See the --sources option.
Supports traversing file-system paths only, no arbitrary URLs.
No support for the <base> tag.
No support for external links. It does not know how to speak HTTP.
Even if you don't have a static site, you can put hyperlink to work by first downloading the entire website using e.g. suckit. In certain cases this is faster than other tools too.
Does not honor robots.txt. A broken link is still broken for users even if
not indexed by Google.
Does not parse CSS files, as broken links in CSS have not been a practical concern for us. We are concerned about broken link in the page content, not the chrome around it.
Only supports UTF-8 encoded HTML files.
Download the latest binary and:
```bash
./hyperlink public/
./hyperlink public/ --check-anchors
./hyperlink public/ --sources src/ ```
Or as GitHub action:
yaml
- uses: untitaker/hyperlink@0.1.24
with:
args: public/ --sources src/
Or build from source using Cargo (needs Rust installed):
bash
cargo install hyperlink # latest stable release
cargo install --git https://github.com/untitaker/hyperlink # latest git SHA
Or install via npm:
bash
npm install -g @untitaker/hyperlink
When invoked without options, hyperlink only checks for 404s of internal
links. However, it can do more.
-j/--jobs: How many threads to spawn for parsing HTML. By default
hyperlink will attempt to saturate your CPU.
--check-anchors: Opt-in, check for validity of anchors on pages. Broken
anchors are considered warnings, meaning that hyperlink will exit 2 if
there are only broken anchors but no hard 404s.
--sources: A folder of markdown files that were the input for the HTML
hyperlink has to check. This is used to provide better error messages that
point at the actual file to edit. hyperlink does very simple content-based
matching to figure out which markdown files may have been involved in the
creation of a HTML file.
Why not just crawl and validate links in Markdown at this point? Answer:
There are countless of proprietary extensions to markdown out there for creating intra-page links that are generally not supported by link checking tools.
The structure of your markdown content does not necessarily match the
structure of your HTML (i.e. what the user actually sees). With this setup,
hyperlink does not have to assume anything about your build pipeline.
--github-actions: Emit GitHub actions
errors,
i.e. add error messages in-line to PR diffs. This is only useful with
--sources set.
If you are using hyperlink through the GitHub action this option is already
set. It is only useful if you are downloading/building and running hyperlink
yourself in CI.
exit 1: There have been errors (hard 404s)exit 2: There have been only warnings (broken anchors)(roughly ranked by performance, determined by some unserious benchmark. this section contains partially dated measurements and is not continuously updated with regards to either performance or featureset)
None of the listed alternatives have an equivalent to hyperlink's --sources
and --github-actions feature.
lychee, like hyperlink, is a great
choice for obscenely large static sites. Additionally it can check
external/outbound links. An invocation of lychee --offline -b . '**/*.htm*'
-n is more or less equivalent to hyperlink ..
liche seems to be fairly fast, but is unmaintained.
htmltest seems to be fairly fast as well, and is more of a general-purpose HTML linting tool.
muffet seems to have similar performance
as htmltest. We tested muffet with
http-server and webfsd without
noticing a change in timings.
linkcheck is faster than linkchecker
but still quite slow on large sites.
We tried linkcheck together with
http-server on localhost,
although that does not seem to be the bottleneck at all.
wummel/linkchecker seems to be the fairly feature-rich, but was a non-starter due to performance. This applies to other countless link checkers we tried that are not mentioned here.
We use Hyperlink to check for dead links on Graphviz's static-site user documentation, because:
- Hyperlink is blazingly fast, checking 700 HTML pages in 220ms (default) and 850ms (with
--check-anchors).- Hyperlink's single-binary release, with no library dependencies, was trivial to integrate into our continuous integration tests.
- High coverage: Hyperlink immediately spotted over a thousand broken page links within both
<a>tags and HTML redirects, and a further 62 broken anchor-links with--check-anchors.- Hyperlink's design decision to crawl only static files (avoiding HTTP), avoids test flakiness from network requests, allowing me to confidently block merging if Hyperlink reports an error.
In conclusion, Hyperlink fills the "static site continuous testing" niche really nicely.
-- Mark Hansen, Graphviz documentation maintainer
Licensed under the MIT, see ./LICENSE.