zenai's review round made keyless explicit in the client's type
(init takes ?[]const u8; null selects the public endpoint) and removed
the always-empty description field from results:
- the .auto keyless rung and searchExplicit pass null instead of "";
the null-key call is comptime-gated to keyless-capable engines and
apiSearch's key parameter is anytype to admit both shapes
- a set-but-empty KEENABLE_API_KEY is now keyed and fails loudly
instead of silently switching to the keyless rate-limit regime
- formatKeenableMarkdown reads snippet only (description no longer
exists to fall back to)
- build.zig.zon repinned to the reviewed zenai commit
Adds keenable to the search tool's engines (KEENABLE_API_KEY, tried
after brave/tavily/exa in .auto). Unlike the other engines it also
answers without any key: its client routes an empty key to the public
endpoint (rate-limited per client IP), so .auto now tries that as the
last rung before the DuckDuckGo scrape, and an explicit
/searchEngine keenable works keyless instead of erroring. Engine table
entries gain init_options so the Keenable client can carry the
lightpanda attribution header.
Needs the zenai Keenable search client (lightpanda-io/zenai PR #9); the
zon pin points at that branch until it lands.
FormData is now encoded based on the Page's encoding on form submission, or
UTF8 for a JS-created FormData. Newlines are also normalized as needed (sole
\r and \n become \r\n in most cases).
The value for a hidden _charset input is now set to the encoding of the FormData
(who knew this was a thing?).
Added a setter to File.input
findFramebyName ignores a disconnected frame.
Most of these were driven by the WPT /FileAPI/ category, but the changes impacts
a handful of cases throughout (e.g. various cases in /html/semantics/forms/ are
also improved).
Network is less cohesive than https://github.com/lightpanda-io/browser/pull/3242
would indicate. It has two distinct and _completely_ separate responsibilities.
1 - It acts as the base for each HttpClient, providing a shared connection pool
(for http and ws) and access to the process wide Cache, RobotStore,
WebBothAuth, certificates, ...
2 - It accepts, polls and reads from CDP connection
There is zero relationship between these, and it's a particularly bad place for
this duality to exist because both parts are, in their own way, the main
multi-threaded junction in the system.
This commit is purely mechanical it:
1 - Keeps network as the base for each HttpClient.
2 - Extract the CDP interaction into the existing Server.zig
3 - Gives mcp's HttpServer its own accept loop
Network has accumulated a bit of bagged. It knows a lot about certificates, it
knows a lot of the cache. I have plans to expand Network, and wanted to clean it
up.
1 - certificate logic moved to dedicated Certificates.zig
2 - Removed CurlDebugAllocator
- this makes setup easier, to Updater can just init libcurl directly
3 - Change Updater to be a single function
4 - Cache initialization is don in the Cache
5 - ?Cache => Cache{.kind = .noop}
Move all of the blank png logic to OffscreenCanvas and apply it to the existing
convertToBlob. I agree it doesn't really make that much sense here, but Canvas
already references OffscreenCanvas, so given its current usage, it's the most
ergonomic place to put it.
Canvas's ToBlobCallback now runs microtask after executing the callback and
lazily creates the Blob for simpler blob lifecycle.
Network has accumulated a bit of bagged. It knows a lot about certificates, it
knows a lot of the cache. I have plans to expand Network, and wanted to clean it
up.
1 - certificate logic moved to dedicated Certificates.zig
2 - Removed CurlDebugAllocator
- this makes setup easier, to Updater can just init libcurl directly
3 - Change Updater to be a single function
4 - Cache initialization is don in the Cache
5 - ?Cache => Cache{.kind = .noop}
matchMedia("(min-device-width: …)") always returned false, so scripts
scanning for the first matching bound never terminated.
device-* resolves against the viewport, like window.screen.width/height.
Add missing toJSON, use Arena + Finalizers for delivery / GeolocationPosition
and coordinates. Return permission denied if permission isn't set, regardless of
override. Split structs into their own files.
Add an optional per-host rate limit. This currently only applies to the top-
level navigation. This makes it simpler to implement and simpler to reason
about. The full load of a page is only ever delayed at its head, not
sporadically through the page loading.
The use-case where a RateLimiter is most useful is when the browser is crawling
multiple pages of the same site, and in that case, the top-level rate limit
still applies some degree of limit to any linked resources (e.g. a JS on a
different host).
`--http-nav-delay` is the delay, in milliseconds, to apply to top level
navigates per host. Currently defaults to 0 (disabled).
`--http-nav-burst` is the burst allowed per host. Defaults to 1 (has no impact
when `--http-nav-delay` is disabled).
1-
Per spec, dictionary objects should be loaded in lexicographical order (a). So
the following always pushes `endings` before `type`
```js
new Blob([], {
get type() { order.push('type'); return ''; },
get endings() { order.push('endings'); return 'transparent'; },
});
```
This requirement is now enforced by a comptime check. So, a lot of files were
updated to satisfy this requirement.
2-
Blob parts now work with the @@iterator protocol. This is done more generically
(via `value.iterator())` but is currently only used by Blob (and File). Other
types that need this (e.g. URLSearchParams) can be done in a follow up.
3-
Add support for Blob.textStream
4-
Reject XHR/Fetch requests on blob URLs if the method isn't GET
5-
Strip #hash from blob URL when doing lookup
All of this fixes ~175 FileAPI WPT cases, though it's possible #1 (the
dictionary ordering) helps with a few others.
(a) https://webidl.spec.whatwg.org/#js-dictionary
Adds the Geolocation and GeolocationPositionError interfaces plus a
navigator.geolocation accessor. getCurrentPosition schedules delivery
on the frame's js.Scheduler so callbacks always fire off the calling
stack, and covers the no-override path where no
Browser.geolocation_override is set: the error callback receives a
GeolocationPositionError with code POSITION_UNAVAILABLE (2), or
PERMISSION_DENIED (1) if the geolocation permission was denied via
CDP. Reading an override into a successful GeolocationPosition and
watchPosition/clearWatch are left for later work.
1. The horizontal position of an element is now calculated based on the
horizontal position of it's siblings (up the tree). Rather than merely the
count * default width.
2. translateX is parsed on inline styles and applies to the horizontal position
1 & 2 fix the https://lseo.com/ issue, but they add overhead to APIs that can be
used a lot. So:
3. Most element dimensions api relied on getElementDimensions which calculated
both the height and width dimensions, even though they only needed once
(e.g. getClientHeight only needs the height). These methods now only get
the axis that they care about
4. Because we do more style sheet lookups, and thus more lookups in
Frame._element_styles, Element.flags now has an has_inline_style which
helps avoid the lookup.
The unconditional full load incurs a performance hit in the common case where
the about:blank is safe to use. This commit restores the fast-path only when
(a) the page is waiting for its initial navigate (same as before) and (b)
no Runtime.* methods were executed (i.e. no driver-driven JS was executed on
about:blank)
We currently implement a v8 signature check on the receiver. This throws on
failure. However, for a promise-returning function, it should reject the
promise instead. This skips the (v8) signature check for promise-returning so
that it flows through the normal code which does rejection.
Fixes a few WPT cases (in fetch and FileAPI, but probably others too).
This reject, don't throw, is an ongoing thing that we're fixing, e.g.
https://github.com/lightpanda-io/browser/pull/3095
Hooks into the xml5ever parse_error callback to capture parse errors and reject
malformed XML. Because of this stricter error handling, we need to do some
input pre-processing for edge cases (thank you WPT). We need to strip out
<!DOCTYPE svg [ <!ENTITY ...> ]> which comes form Illustrator SVG export since
that now causes errors (but should be ignored).
Finally, by default xml5ever closes any opened tags at the end of the stream.
Per WPT, these should be invalid.
The rust work was all Claude-driven.
Ilya Bogin