capybara-lightpanda 0.14.2 → 0.15.0

data/lib/capybara/lightpanda/options.rb

@@ -20,10 +20,11 @@ module Capybara
       # `Capybara::Lightpanda.configure { |c| c.port = 9222 }` when external
       # tooling needs a known address.
       DEFAULT_PORT = 0
-      # Mirrors Lightpanda's own `Viewport.default` (1920x1080) rather than
-      # Cuprite's 1024x768. window_size is applied for real now (see below), so
-      # a 1024x768 default would silently shrink the viewport of every existing
-      # suite and flip `@media` branches under them. Matching the browser's
+      # Mirrors Lightpanda's own `Viewport.default` (1920x1080), not the
+      # 1024x768 some drivers default to. window_size is applied for real now
+      # (see below), so a 1024x768 default would silently shrink the viewport
+      # of every existing suite and flip `@media` branches under them.
+      # Matching the browser's
       # native default keeps `window_size` truthful AND leaves default
       # behavior byte-identical to before it was wired up.
       DEFAULT_WINDOW_SIZE = [1920, 1080].freeze
@@ -36,16 +37,17 @@ module Capybara
       # no rendering engine, so getBoundingClientRect stays synthetic and
       # nothing reflows. Sizing down will not make an off-viewport element
       # report as obscured.
-      # headless is accepted for Cuprite drop-in compatibility but inert —
+      # headless is accepted so suites migrating from other drivers keep
+      # working, but is inert —
       # headless is the only mode Lightpanda runs in.
-      # save_path: directory for downloaded files (Cuprite parity). nil falls
+      # save_path: directory for downloaded files. nil falls
       # back to Capybara.save_path at create_page time; downloads stay off when
       # both are nil (Browser#create_page only opts in when a path exists).
       # raise_on_unhandled_modal: a JS dialog that opens with no
       # accept_modal/dismiss_modal pre-arm in flight is resolved by Lightpanda's
       # silent default (confirm → cancel, prompt → null, alert → dismissed), so
       # a runaway confirm cancels the action and the spec can still pass. false
-      # (default, Cuprite parity) warns on stderr; true raises
+      # (default) warns on stderr; true raises
       # UnhandledModalError from the action that opened it.
       # load_images: adds `image` to the `--load-resources` value so `<img>`
       # elements are fetched over the network (upstream #3230, build >= 8834,

data/lib/capybara/lightpanda/process.rb

@@ -264,7 +264,48 @@ module Capybara
       # end in Network.loadingFailed, which Network consumes); #3529 (9877)
       # SameSite=None without Secure refused (Cookies#set reports it);
       # #3702 (9968) click no longer focuses (CLICK_JS focuses itself).
-      MINIMUM_NIGHTLY_BUILD = Gem::Version.new("9994")
+      #
+      # 2026-10-16 bump 9994 -> 10641 (= release 1.0.1, the new lockstep cap).
+      # Retired or ungated by it:
+      #   #3958 (10613, ours) Window and Document carry the GlobalEventHandlers
+      #     `on*` attributes, so javascripts/event_handlers.js is gone — jQuery
+      #     1.x sees `'onchange' in window` natively.
+      #   #3940 (10536) a link to a non-fetchable scheme (mailto:, tel:) is a
+      #     no-op that keeps the document, so CLICK_JS no longer cancels the
+      #     click's default from a window listener.
+      #   #3755 / #3849 (10049 / 10310, ours / upstream) exceptions from timers,
+      #     rAF, inline handlers and microtasks reach `window.onerror`, so
+      #     Browser#page_errors sees them; #3757 (10052) hides a closed
+      #     [popover]; #3759 (10057) routes send_keys into an iframe; #3874
+      #     (10366) Ctrl+A selects a field — upstream_fixes_test.rb runs them
+      #     un-gated.
+      #   #3763 (10239, ours) history traverses pushState entries in place;
+      #     #3831 (10255) `window` serializes as CyclicObject, not a node;
+      #     #3843 (10261) a failed navigation commits an error document — their
+      #     tests drop the nightly_build branches.
+      #   #3952 (10575) typed text fires `change` when committed by blur/Enter.
+      # A `nightly_build` gate is nil on a release binary, so every such test
+      # skipped on 1.0.1 although it carried the fix: ungate, don't re-gate.
+      #
+      # 2026-10-23 bump 10641 -> 10965 (= release 1.0.2, the new lockstep cap;
+      # CSS nesting #3962 at 10976 and blank-page timers #4148 at 11012 stay
+      # above it). Rides along:
+      #   #4096 (10904) Runtime.exceptionThrown for unhandled rejections, and
+      #     `unhandledrejection` no longer fires for a rejection handled in
+      #     the same tick — errors.js stops reporting those false positives;
+      #     page_errors_test.rb drops its nightly_build gate. With #4115
+      #     (10910, an iframe's exception carries its context id) this
+      #     unblocks the B19 swap off errors.js.
+      #   #3997 (10685) the scheduler rework: a MessageChannel task loop no
+      #     longer delays a pending fetch tenfold (#4042), which held
+      #     wait_for_idle.
+      #   #3658 (10657) client rects follow the scroll and
+      #     DOM.scrollIntoViewIfNeeded is real — HTML5_DRAG_JS centers the
+      #     drop target itself because scrollIntoView ignores `block`.
+      MINIMUM_NIGHTLY_BUILD = Gem::Version.new("10965")
+
+      # Lines of browser output #output keeps (ferrum keeps the same 100).
+      OUTPUT_LINES = 100
 
       # Second, equivalent floor for the *release* channel.
       #
@@ -273,16 +314,16 @@ module Capybara
       # bare "0.3.6" carrying no commit counter at all, and MINIMUM_NIGHTLY_BUILD
       # has nothing to compare against. Releases are cut from the same trunk, so
       # a release is acceptable exactly when its own commit count clears the
-      # nightly floor. 1.0.0 (2026-10-02) is build 9994 — exactly the nightly
+      # nightly floor. 1.0.2 (2026-10-23) is build 10965 — exactly the nightly
       # floor, which is no accident: the floor was raised *to* the release so
-      # both channels gate the same commit. (0.4.1 = 9463 and 0.4.0 = 9058
-      # predate fixes the floor exists for.)
+      # both channels gate the same commit. (1.0.1 = 10641 predates fixes the
+      # floor exists for.)
       #
       # INVARIANT: every MINIMUM_NIGHTLY_BUILD bump must also move this to the
       # first release containing that build (`git rev-list --count <tag>` in the
       # browser repo tells you). Leaving it behind would let the release channel
       # silently accept a binary the nightly channel rejects.
-      MINIMUM_RELEASE = Gem::Version.new("1.0.0")
+      MINIMUM_RELEASE = Gem::Version.new("1.0.2")
 
       class << self
         # `lightpanda version` prints one of two shapes, and the gem supports
@@ -348,6 +389,17 @@ module Capybara
         @stderr_r = nil
         @stderr_w = nil
         @finalizer_registered = false
+        @output = []
+        @output_partial = {}
+        @output_mutex = Mutex.new
+      end
+
+      # The last OUTPUT_LINES lines Lightpanda wrote to stdout/stderr, oldest
+      # first, across restarts of this Process. Kept for diagnostics: a crash
+      # or panic only ever shows up there, and from build 10783 so do page
+      # errors (`page_error` scope, WARN). Mirrors ferrum's Process::Output.
+      def output
+        @output_mutex.synchronize { @output.dup }
       end
 
       def start
@@ -416,7 +468,7 @@ module Capybara
             next unless ready
 
             ready[0].each do |io|
-              io.read_nonblock(4096)
+              record_output(io, io.read_nonblock(4096))
             rescue IO::WaitReadable
               # No data
             rescue EOFError
@@ -430,6 +482,18 @@ module Capybara
         end
       end
 
+      # Split per pipe so a line cut across two reads is kept whole, and a
+      # stdout fragment never glues onto a stderr one.
+      def record_output(io, chunk)
+        text = chunk.dup.force_encoding(Encoding::UTF_8).scrub
+        @output_mutex.synchronize do
+          *lines, rest = "#{@output_partial[io]}#{text}".split("\n", -1)
+          @output_partial[io] = rest
+          @output.concat(lines)
+          @output.shift(@output.size - OUTPUT_LINES) if @output.size > OUTPUT_LINES
+        end
+      end
+
       def spawn_process(binary_path)
         args = build_args
 
@@ -499,6 +563,7 @@ module Capybara
             ready[0].each do |io|
               chunk = io.read_nonblock(1024)
               output << chunk
+              record_output(io, chunk)
 
               if (match = output.match(READY_PATTERN))
                 @ws_url = "ws://#{match[1]}/"

data/lib/capybara/lightpanda/version.rb

@@ -2,6 +2,6 @@
 
 module Capybara
   module Lightpanda
-    VERSION = "0.14.2"
+    VERSION = "0.15.0"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: capybara-lightpanda
 version: !ruby/object:Gem::Version
-  version: 0.14.2
+  version: 0.15.0
 platform: ruby
 authors:
 - Navid Emad
@@ -91,7 +91,6 @@ files:
 - lib/capybara/lightpanda/headers.rb
 - lib/capybara/lightpanda/javascripts/attach.js
 - lib/capybara/lightpanda/javascripts/banner.js
-- lib/capybara/lightpanda/javascripts/errors.js
 - lib/capybara/lightpanda/javascripts/predicates.js
 - lib/capybara/lightpanda/javascripts/turbo.js
 - lib/capybara/lightpanda/keyboard.rb

data/lib/capybara/lightpanda/javascripts/errors.js

@@ -1,74 +0,0 @@
-// --- Uncaught page errors ---
-// Lightpanda implements no `Runtime.exceptionThrown` (upstream gap, wishlist
-// B19), so an uncaught exception in page JS reaches no CDP client at all. It is
-// invisible to console_logs, which only ever sees explicit console.* calls, so
-// a handler that dies on a TypeError leaves no trace and the failure surfaces
-// far away as ElementNotFound on whatever it was supposed to produce. That is
-// the exact shape of the solidus taxon-tree failure in real-apps causes.yml.
-//
-// These listeners are passive — no preventDefault, so nothing the page would
-// otherwise see is swallowed, and nothing that would have propagated stops
-// propagating. They re-emit through the same console.debug sentinel channel
-// turbo.js uses, and console.rb routes them into Browser#page_errors rather
-// than #console_logs. Keeping the two apart is deliberate: console_logs stays
-// "the console.* calls the page made", which is what Chrome models (an
-// exception is exceptionThrown, not consoleAPICalled) and what Playwright and
-// Puppeteer expose (pageerror, separate from console). Folding exceptions into
-// console_logs would also break every suite already asserting that buffer is
-// free of errors.
-//
-// Delete this file once upstream emits Runtime.exceptionThrown and the floor
-// covers it: subscribe Browser::Console to that event and feed the same buffer.
-function _reportPageError(fields) {
-  try {
-    console.debug('__lightpanda_page_error_' + JSON.stringify(fields));
-  } catch (e) {
-    // A field that won't serialize (a getter that throws, a cyclic `reason`)
-    // must not turn one page error into two, so report the kind and move on.
-    try {
-      console.debug('__lightpanda_page_error_' + JSON.stringify({
-        kind: fields && fields.kind ? fields.kind : 'error',
-        message: '(page error could not be serialized)'
-      }));
-    } catch (e2) {}
-  }
-}
-
-window.addEventListener('error', function (e) {
-  // A failed <img>/<script>/<link> fetch fires `error` too, but on the element
-  // and without bubbling, so it shouldn't reach this listener. Guard anyway
-  // (Lightpanda's dispatch need not match Chrome's exactly) and identify those
-  // by their missing string `message` — a resource failure is a network fact
-  // and Network#traffic already carries it.
-  if (!e || typeof e.message !== 'string') return;
-
-  _reportPageError({
-    kind: 'error',
-    message: e.message,
-    url: typeof e.filename === 'string' ? e.filename : '',
-    line: typeof e.lineno === 'number' ? e.lineno : null,
-    column: typeof e.colno === 'number' ? e.colno : null,
-    // Cap the stack: it rides a console argument over the CDP WebSocket, and a
-    // deep recursion stack can be megabytes.
-    stack: e.error && typeof e.error.stack === 'string' ? e.error.stack.slice(0, 4000) : null
-  });
-});
-
-window.addEventListener('unhandledrejection', function (e) {
-  var reason = e ? e.reason : null;
-  var message;
-  try {
-    message = reason && reason.message ? String(reason.message) : String(reason);
-  } catch (err) {
-    message = '(rejection reason could not be stringified)';
-  }
-
-  _reportPageError({
-    kind: 'unhandledrejection',
-    message: message,
-    url: '',
-    line: null,
-    column: null,
-    stack: reason && typeof reason.stack === 'string' ? reason.stack.slice(0, 4000) : null
-  });
-});