A merged index that failed to load made mergeIndex throw, and the
failed instance stayed in the list, so search stopped working for
every index. Indexes passed with `optional: true` are now dropped
with a warning when they fail to load, and search runs over the
indexes that did load.
Fixes#887
Relative images were resolved against the page URL, but only the
pathname was kept, so with a baseUrl on another domain the image
still loaded from the search site. Keep the origin when the page URL
has one, in the Component UI's resolveUrl filter and in both Default
UI result templates. result_with_subs.svelte didn't resolve images at
all before.
Fixes#573
exclude_selectors is the only way the config can reach into a page, and it
carries the semantics of data-pagefind-ignore="index". Every other control
Pagefind has lives on an attribute, so reaching it from the config meant a
new flag per attribute.
Add a rules list that pairs a selector with the attributes to set on the
elements it matches. The attributes are applied before every other handler,
so the rest of the parser reads them exactly as if the page had carried
them, and every documented attribute is reachable without new flags.
An attribute already present on the element wins, so a rule never overrides
markup. A selector that cannot be parsed is reported and skipped.
Rules are config-file only: a list of selectors with attribute maps does not
fit a CLI flag or an environment variable.
--exclude-selectors marked a match as "ignored", the same as a bare
data-pagefind-ignore, so metadata and filters inside the match were still
collected. Excluding img.emoji, for example, still let the emoji become the
page's auto-detected image.
The status was also applied after the element had been processed. For a void
element such as <img> that is too late twice over: its metadata has already
been read, and the parser has already popped back to the parent, so the
status landed on the parent instead and wiped out the rest of the page.
Match the selectors in a handler registered ahead of the main element
handler, which now reads that flag and gives the node Excluded status up
front. The built-in REMOVE_SELECTORS (head, nav, footer, ...) keep the old
Ignored status, so metadata tagged inside them still counts.
Fixes#692
A file named Müsli.html was indexed with the URL /Müsli.html, which only
resolves if the host happens to serve the same byte sequence the index was
built from. Encode the URL that build_url derives from the file path using
the URL spec's path percent-encode set, leaving / and % alone so path
separators and already-encoded filenames are unaffected.
URLs supplied by the caller (data-pagefind-meta="url:...", addCustomRecord)
do not go through build_url and are untouched.
Fixes#619
build_url() escaped %, # and ? one by one, so a file name carrying any
other character that a URL path cannot hold - a space, a quote, an angle
bracket, a control character - was emitted raw and only resolved if the
host happened to serve the same byte sequence the index was built from.
Encode the URL that build_url derives from the file path with the URL
spec's path percent-encode set, leaving / alone so the separators between
path segments survive. % stays in the set, so a file literally named in
encoded form keeps pointing at the file that is on disk.
URLs supplied by the caller (data-pagefind-meta="url:...", addCustomRecord)
do not go through build_url and are untouched.
Fixes#619
Every --exclude-selectors value was joined into a single selector string
handed to lol_html, so one selector the parser does not support took the
whole run down with an unwrap panic. Selectors are now validated one by
one and unsupported ones are reported through the logger and dropped.
Fixes#931
dualfroz