getgrav/grav 2.0.7

on GitHub

Bugfix

• [security] A page editor can no longer run commands on the server by hiding a callable directive in a form field's settings; dynamic field data now refuses dangerous functions and cannot be tricked into reaching one through a helper (GHSA-fj2p-qj2f-74v5).
• A page's translatedLanguages() now localizes ancestor slugs too, so a nested translation whose parent folder has a localized slug: produces the fully translated cross-language link instead of leaving parent segments in the current language. Fixes getgrav/grav#4186.
• Pointing the log stream at environment:// (for example log: environment://logs) no longer crashes the site or bin/grav clear with a "stream must either be a resource or a string" error when the per-environment folder does not exist; logging now falls back to the default logs/ folder instead. Fixes getgrav/grav#4172.
• The media:// stream now checks the per-environment user/env/<host>/media/ folder before the shared user/media/, so site media stored per environment resolves to the correct URL in the admin and in page content instead of a broken user/media/ link. Fixes getgrav/grav#4188.
• Large file downloads such as site backups are now streamed to the browser in chunks instead of being loaded into memory all at once, so a download bigger than PHP's memory limit no longer fails with a blank server error. Fixes getgrav/grav-plugin-api#12.
• Backups initialization no longer runs twice when something that bypasses the normal request middleware (such as the API plugin) also attaches the backup scheduler listener, so the listener is registered only once.
• Pages accessed with URL parameters such as pagination or taxonomy filters no longer recompile every Twig template on each request, restoring full template caching on exactly the pages that get the most traffic.
• The per-file compiled cache for YAML and markdown files now loads through its intended opcache fast path, and a source file that has been deleted no longer serves stale cached data.
• A modular page that outputs trusted theme or plugin markup, such as a form with a reCAPTCHA field, is no longer wrongly blanked by the content security scan, which now checks the editor's own content instead of the finished template output. Fixes getgrav/grav-plugin-form#636.
• Chaining media actions on page media under the content security scan, such as {{ page.media['x.jpg'].lightbox(1024,768).cropResize(176,176).html() }}, now works instead of being blocked, and the scan's list of allowed media methods stays in step with Grav's documented media actions automatically.

Improved

• Updating a plugin or theme whose required dependency is held back by a newer Grav or PHP requirement now explains the real fix. Instead of reporting that the needed version is "higher than the latest release" and suggesting a cache refresh, the updater names the newer dependency release and the Grav (or PHP) version it needs, so you know to update Grav first. Relates to getgrav/grav-plugin-admin2#93.
• Backup profiles now always appear in the scheduler where each can be switched on or off with the Enabled/Disabled toggle, instead of a profile staying hidden until its schedule was turned on; the profile's schedule setting seeds the default state and an explicit toggle takes precedence.
• Frontend requests are noticeably faster across the board: the scheduler, backups machinery, error page renderer and logger now initialize only when actually used instead of on every page view, cutting over 50 PHP files from a typical request.
• The filesystem scan that checks pages for changes now reuses its result for a couple of seconds (configurable with cache.check.interval), so busy sites no longer stat every page file on every single request; content edits still show up right away in normal editing workflows and admin saves remain instant.
• Configuration, blueprint and language file lists honor the same freshness window instead of checking every tracked file's timestamp three times per request, and theme blueprints no longer load at all on normal frontend page views.
• Class autoloading is faster: source installs get an optimized class map, APCu is suggested so the existing autoloader cache setting can engage, and plugin autoloaders no longer sit in front of the core one where every core class lookup had to pass through them first.
• Rendering a page with cached content no longer loads its whole media collection up front, the pages index no longer stores pre-computed metadata for every page, and relative markdown links resolve their target page directly instead of building the full page list per link.
• Assorted hot-path trims: string helpers use fast native functions for the common case, asset rendering skips per-asset integrity work when the feature is off, the site root URL is computed once per request instead of per asset, and debugger timers cost nothing when the debugger is disabled.
• New experimental opt-in page index (pages.lazy_index: true): pages, routes, children lists, sort orders and the taxonomy map load on demand from a per-page index instead of one large cache blob that has to be fully unserialized on every request, making per-request cost flat as sites grow: a 2,000 page test site renders as fast as a 2 page one and uses a quarter of the memory; SQLite powers the index when available with a pure PHP fallback, and the default behavior is completely unchanged until the flag is enabled.
• Page collection filters (visible, routable, published, module) now use menu flags recorded in the page index, so building a navigation menu that filters a folder to its visible pages no longer loads every hidden sibling first. On a 500 post blog under the Quark theme this cut the pages built for a page view from all 507 to 7 and roughly halved memory; it helps every site, most of all large ones with the experimental page index enabled.
• Sorting a page collection by date, title, or another common field now reads that value from the page index instead of loading every page in the collection just to read one field, and on single language sites the automatic translated filter that every collection applies no longer loads any pages at all. With the experimental page index enabled, a blog post showing a related posts grid dropped from loading every post on the site to only the handful it displays.
• The setting that scans page content for XSS moved to security.content.xss_scan_output, since it applies to all page content rather than only Twig in content; the previous security.twig_content.xss_scan_output location keeps working and is moved to the new one automatically on upgrade.