v2.0.0-next.43

Patch Changes

• 7872c96: A base with a trailing slash, such as Vite's import.meta.env.BASE_URL ("/app/"), no longer produces double slashes. Router.paths hrefs and navigate("/about") now give /app/about instead of /app//about, which matched no route.
• 1865ccc: Stamp a pushed history entry's _depth as one above the current entry instead of history.length - 1. Browsers cap session history (50 entries in Chromium and Firefox), so once the cap was reached every new entry got the same depth. That broke scroll restoration and useBeforeLeave back-navigation deltas on long sessions. Entries the router did not push, such as a #anchor jump, are stamped one above the last known depth as well.
• d4c5e6e: Peer dependencies on solid-js and @solidjs/web move to ^2.0.0-rc.19.
• ea3d8ba: Server link claims no longer parse a root-relative href with URL when its path and query only use characters URL keeps unchanged and it has no ./.. segments, and comparablePath skips its parse for such paths. Other hrefs still go through URL. On a 1000-link SSR page this cuts the claims' cost per request from about 1.2 ms to about 0.4 ms.
• 8d2a9e4: query.get no longer returns a value from a fetch that was superseded. When an older fetch for a key settles after revalidate() started a newer one, or after query.set wrote a value, its result is no longer written over the newer value.
• 5f02d81: A server-side navigate() now writes its 302 status and Location header into the existing event.response instead of replacing the object. Headers already on the response, such as Set-Cookie and the flash-cookie clear, are kept, and a redirect issued after createSSRResponse captured the head still reaches it. Query redirects during SSR resolve Location against the router base as a result.
• 7488be9: Link preloading and click handling resolve an anchor's href against document.baseURI. An anchor cloned from a <template> keeps its raw href (e.g. /about) until it is adopted into the page, so eagerPreload's idle flush threw Invalid URL when it reached one that a transition still held off the page (#704).