Make attribute maps live, preserve parser script execution state, and expose
element scrolling used by React commit callbacks. Connect fallback web streams
so React Router and Remix hydration payloads flow through codec transforms.
Skip <link rel=stylesheet> sheets that are disabled or gated to
print/dark, and resolve @import chains inside fetched sheets so the
imported CSS reaches the cascade.
- disabled links (JS-toggled alternate themes) no longer paint pages
dark under a light Chromium
- media="(prefers-color-scheme: dark)" link sheets are skipped
- @import url(...) is fetched recursively; Sphinx docs pull their layout
from basic.css via classic.css and it was being dropped
Pixel fidelity vs headless Chromium: docs.python 95.5 to 10.2 percent,
web.dev 91.1 to 10.5, developer.chrome 88.0 to 18.9. No regressions on
the good-site set; obstacle course still 33/33.
A <link rel=stylesheet media="print"> must not apply on screen; fetching
and injecting it let print rules override the screen layout (e.g. the
Guardian's print sheet). Skip links whose media is print-only.
The render layer had no CSS Grid support, so Vector 2022's page layout
(three nested grids) collapsed to a vertical stack with the article
landing tens of thousands of pixels down the page. Wikipedia rendered
as a near-blank page and Hacker News only worked through per-site
hardcoded CSS overrides.
Removed the HN-specific hardcoding and fixed the general engine instead:
- CSS Grid layout (grid-template-areas/columns/rows via taffy's grid mode)
- Indexed, specificity-correct cascade in a new css.rs, replacing the
old test-every-selector-against-every-node scan. Cascade time on
Wikipedia went from 9.5s to under 200ms.
- Ancestor bloom-filter fast-rejection for selector matching in
obscura-dom, needed to make the indexed cascade actually fast on
descendant combinators.
- :link/:visited pseudo-classes were previously unparseable, so any
rule using them (a:link{color:...}, extremely common) silently
dropped the whole rule from the cascade.
- overflow:hidden clipping, fixing the pervasive visually-hidden-but-
accessible pattern used for skip links and screen-reader-only labels.
- @media parser robustness: it did a literal substring search for
max-width: and silently dropped the constraint on max-width : 750px
(space before colon) or max-width:calc(750px - 1px), treating
mobile-only or narrow-viewport rules as always active. The calc()
case was hiding Wikipedia's TOC sidebar via a rule meant only for
narrow screens.
- Table auto-layout approximation: the trailing auto-width td in a row
absorbs leftover space, so a width:100% table's unsized last column
reaches the right edge instead of shrinking to content.
- Inline formatting context approximation: a block holding inline-level
children (a paragraph or list item with links) is promoted to a
wrapping flex row so text and links read as continuous prose instead
of one fragment per line.
- Whitespace-only text nodes no longer get a full line-height box. They
used to stack up as blank rows, turning Wikipedia's collapsed TOC
subsections and the page header into hundreds of pixels of dead space.
- SVG rendering via resvg/usvg, plus resolving relative image URLs
against the page's base URL.
- background-image support including background-size/background-position
for correctly sized and positioned icons.
- Per-screenshot image fetch cache.
- Fixed the layout-only (no paint feature) build, which didn't compile
due to an unconditional call into paint-gated code.
End-to-end rasterization behind the render feature (which now enables
obscura-render/paint):
- obscura-render: screenshot_png(dom, viewport) paints and encodes to PNG
via tiny-skia's encode_png (no extra dep).
- obscura-browser: Page::screenshot renders the live DOM at a viewport.
- CLI: obscura fetch --screenshot out.png writes the settled page to a PNG.
- CDP: Page.captureScreenshot returns the PNG base64-encoded, so
Puppeteer/Playwright page.screenshot() works. captureSnapshot still
errors (it is a DOM/layer snapshot, not an image).
Verified: both paths produce correct 1280x720 RGBA PNGs with elements
laid out and colored as the CSS specifies (block stacking, backgrounds,
white page). Default build unchanged; obstacle course 33/33.
Text, borders, and images are not yet painted; the box layout and
background colors are rendered.
getBoundingClientRect, offsetWidth/Height, and offsetTop/Left now read
real box geometry from the obscura-render layout cache when the render
feature is compiled in. elementFromPoint inherits correct hit-testing
because it already delegates to getBoundingClientRect per element.
Adds a render feature on obscura-js (propagated through obscura-browser
and obscura-cli), a lazily-computed layout cache and viewport in
ObscuraState (feature-gated, cleared on navigation), and an
op_layout_geometry op. bootstrap.js probes the op with typeof, so the
default scraping build is unchanged and falls through to the synthetic
rect.
With --features render, block siblings stack and flex children place
horizontally at correct positions, and elementFromPoint resolves the
right element. Default build returns the synthetic rect unchanged.
Obstacle course 33/33; obscura-js unit failures identical to main.
Keep pumping the event loop past the general 500ms settle limit only while the dynamic-script queue remains pending, with a 3s hard cap.
Add a deterministic CDP regression for direct and nested external scripts whose fetch crosses the original settle deadline.
The premise was wrong. deno_core 0.350's run_event_loop is:
pub async fn run_event_loop(&mut self, opts) -> Result<(), CoreError> {
poll_fn(|cx| self.poll_event_loop(cx, opts)).await
}
so replacing `timeout(budget, run_event_loop())` with a hand-rolled
`poll_fn(poll_event_loop)` loop is the same code with the same cancellation
semantics, not safer ones. The mechanism I described cannot happen either:
`poll` is a synchronous `&mut self` call, a future can only be dropped between
polls, and the HandleScope/ContextScope live inside poll_event_loop as stack
locals that are destroyed before it returns. The isolate is exited between
polls both before and after that change.
What actually fixes#430 is confining each connection's isolates to their own
OS thread, which the next commits do. The regression test still reproduces the
abort on the pre-thread-per-connection tree and still passes here without this
commit, so nothing depended on it.
It was not merely a no-op. Two regressions it introduced:
- resolve_promises_until inverted the meaning of tick_ms. It was "pump the
event loop for up to tick_ms, returning early on idle"; it became one single
poll followed by an unconditional sleep(tick_ms). With the 1->50ms backoff
that settles at 20 event-loop turns per second, so an awaited promise chain
needing many turns crawls, and past the budget the caller silently reads a
stale __obscura_await_meta.
- drive_module_eval replaced `e?` with `let _ = poll_event_loop(...)`,
discarding event-loop errors. A module stalled on a top-level await used to
exit immediately with a logged error; it instead spun for the whole budget
logging nothing -- the exact #374 starvation the surrounding comment claims
to prevent.
Keep the drive_module_eval extraction (it genuinely dedupes load_module and
load_inline_module) but restore the original select!-based body.
Refs #430.
- Delete the now-unused CdpContext::new_full (its only caller switched to
new_with_shared_context); have _new_inner build the shared BrowserContext and
delegate, so the struct's field list lives in one place.
- Extract drive_module_eval, shared verbatim by load_module and
load_inline_module.
- Update comments that still described the deleted process-wide V8 lock and the
single-LocalSet model to the per-connection reality, drop the dangling
reference to the removed obscura-js/src/v8_lock.rs, and soften the
ISOLATE_CREATE_LOCK note now that the main-thread warmup is the real guard
against the InitializeBuiltinJSDispatchTable segfault.
Refs #430.
Several event-loop operations bounded their runtime with
tokio::time::timeout(budget, run_event_loop()) or a select! against a
timer. When the timer wins, tokio drops the run_event_loop future
mid-poll; deno_core enters the isolate while polling, so the drop leaves
the isolate entered on the OS thread. Under concurrent CDP work a second
page's execute_script then trips heap->isolate() == TryGetCurrent() and
aborts the process (#430).
Drive the loop with poll_event_loop inside a poll_fn instead. It creates
and drops its own isolate scope per call, so stopping between ticks never
leaves the isolate entered. Convert resolve_promises_until,
run_event_loop_bounded, resolve_promises, the mod_evaluate settle in
load_module and load_inline_module, and the two navigation settle loops
in page.rs. Add poll_event_loop_once as the cancellation-safe single-tick
primitive the settle loops use.
Keeps the obstacle course at 33/33. This is hardening toward #430; the
residual concurrency abort still needs the thread-per-runtime refactor
that v8_lock.rs calls the "Option 2" fix.
Refs #430.
The on_request/on_response Vec lived on the ObscuraHttpClient shared by
every page in the context, so a callback registered on one page fired for
all pages (#408 cross-page bleed) and outlived its page.
Callbacks now live in a per-page CallbackRegistry (Arc shared with the
page's JS runtime state). Page-driven fetch paths (navigation, POST,
scripts, stylesheets, robots.txt, module loads, JS fetch/XHR in both
plain and stealth modes) pass the registry through; the HTTP client
itself is callback-free.
Adds callbacks_do_not_bleed_across_pages: page A's callback stays silent
for page B's requests and stops firing after page A is dropped.
on_request/on_response stored callbacks in append-only Vecs with no way
to remove one, so a crawler that installed a response-body capture hook
for one phase could never detach it: it fired for the shared client's
whole lifetime, re-registering across navigations piled up duplicates,
and (because the client is shared per context) it saw sibling pages'
requests too.
Tag each callback with a stable id. on_request/on_response now return
the id and off_request/off_response (remove_on_request/remove_on_response
on the client) detach a specific one. Storage is Vec<(u64, Callback)>;
all fire sites destructure the tuple. Closes#408.
Requests made by page JS (fetch/XHR/dynamic script and stylesheet loads)
run in the V8 op layer and never produced a NetworkEvent, so CDP emitted
Network.requestWillBeSent / responseReceived only for the static
navigation subresources. Puppeteer/Playwright page.on('request') and
on('response') therefore missed every XHR/JSON response, which breaks
response capture (and is the root cause of the Aviasales half of #394).
op_fetch_url now records a JsNetworkEvent keyed by the same fetch-{N} id
under which it already stores the response body. The Page drains these
into its network_events before the CDP layer emits, so they flow through
emit_navigation_events as requestWillBeSent / responseReceived /
loadingFinished alongside the document subresources, and
Network.getResponseBody resolves for them via the shared id. Applied to
both navigation paths (server process_navigation and in-process
do_navigate). The queue is capped at 4096 events per page.
Closes#406.
server.rs logged previews of attacker-controlled strings with raw byte
slicing: `&text[..text.len().min(200)]` for a malformed CDP frame (inside a
`warn!`, which is active at the worker's default log level) and the same for
an intercepted event JSON. `&s[..n]` panics when byte `n` falls inside a
multi-byte UTF-8 character, so a single WebSocket frame that is invalid JSON,
>=201 bytes, with a multi-byte char straddling byte 200 panics the CDP
processor task — an unauthenticated one-frame crash. page.rs had the same
pattern (behind --verbose debug logging) for evaluated JS expressions.
Add truncate_on_char_boundary (str::floor_char_boundary is still unstable) and
route all four sites through it. Regression test constructs the exact 'a'*199 +
'€' input, asserts the old raw slice panics (via catch_unwind) and the helper
truncates on the boundary instead.
Network.getResponseBody returns the whole body in one base64 blob and the
cache drops anything over 2 MiB, so large downloads (images, PDFs,
archives) either failed or forced the long-running server to hold the
full body in memory (issue #360).
Add Fetch.takeResponseBodyAsStream + IO.read + IO.close. The body is moved
out of the page cache into a stream handle and delivered in bounded chunks
via IO.read, then freed on IO.close, so it is held once and released after
the download. Binary bodies are byte-exact (stored base64). To stream a
large body, raise OBSCURA_NETWORK_BODY_BUFFER_BYTES so it is cached.
This is entirely opt-in: nothing runs unless a client calls the new
methods, so the default fetch/render/scrape path is unchanged (verified:
obstacle course 33/33, obscura-vs-Chrome head-to-head within noise of the
pre-change baseline).
Fixes#360.
The script-execution phase had a 10s soft deadline. On a heavy SPA the
remaining scripts were skipped before the app finished booting, so it
never fired its XHR/fetch calls and page.on('response') captured zero API
responses (issue #361). The knob to extend it, OBSCURA_SCRIPT_DEADLINE_MS,
existed but was undocumented, so it was undiscoverable.
Raise the default to 30s and document the env var. Only pages that keep
running past the deadline are affected; fast pages finish and return well
before it, so the larger budget costs them nothing, and the per-phase
watchdog (armed at deadline + 1s) still bounds a real synchronous hang.
Very heavy SPAs on slow networks can raise it further with
OBSCURA_SCRIPT_DEADLINE_MS, paired with a matching client nav timeout.
Fixes#361.
Navigating straight to an image, PDF, or other resource could not read the
main response body over CDP, and Page.frameNavigated always reported
text/html.
Two problems: the main resource's body was stored under its internal
request id, but the client sees it as the navigation loaderId (Chrome's
requestId === loaderId convention), so Network.getResponseBody(loaderId)
missed. And binary bodies were stored as UTF-8-lossy text, corrupting the
bytes. Alias the stored body to the loaderId, store non-text-like content
types (images, PDF, octet-stream) base64, and report the response's real
Content-Type in frameNavigated. HTML pages are unchanged.
Fixes#340.
op_fetch_url ignored the url/method/headers/body fields on
InterceptResolution::Continue, so the interception channel could pass, mock, or
block a request but not rewrite it. Apply them: shadow url/method/body and
derive custom_headers from the override after the channel resolves Continue.
Also fix enable_interception() called before the first navigation: init_js
propagated intercept_tx to the new runtime but not intercept_enabled, so the
runtime started with interception disabled and op_fetch_url never intercepted.
Propagate intercept_enabled in init_js as well.
Add a test that rewrites a fetch's url via Continue and asserts the rewritten
response is captured.
Expose the interception infrastructure that already existed internally but was
private, so embedders can intercept, observe, mock, or block JS fetch()/XHR
without forking or unsafe casts.
On obscura::Page:
- add_preload_script: inject JS before any page <script> (CDP
Page.addScriptToEvaluateOnNewDocument). The field already ran the scripts;
this only adds the method.
- enable_interception: returns a receiver for every JS fetch()/XHR request;
resolve each through its resolver with InterceptResolution::{Continue,
Fulfill, Fail} to pass, mock, or block (CDP Fetch.enable equivalent).
- on_request / on_response: passive callbacks. on_response is the main path for
capturing API response payloads from SPAs.
on_request/on_response previously fired only for navigation; op_fetch_url now
invokes them for JS fetch()/XHR too (both reqwest and stealth paths), gated on
a non-empty callback vector so the common no-callback case pays nothing.
Re-export InterceptedRequest, InterceptResolution, RequestInfo, ResourceType,
RequestCallback, ResponseCallback, Response from the obscura crate.
Integration test tests/interception.rs exercises all three against a local echo
server.
Resolved conflicts:
- page.rs: keep both set_blocked_urls (#315) and stealth client wiring
- ops.rs: keep SHA-512/224 + SHA-512/256 from stealth branch; keep
the unreachable catch-all (vec![]) from #314 instead of silent
SHA-256 fallback; updated comment accordingly
- bootstrap.js: extend the digest algorithm allowlist to include
SHA-512/224 and SHA-512/256 to match the Rust additions
op_fetch_url backed JS fetch() and XHR with a plain reqwest/rustls client, so
in stealth mode scripted requests carried a non-browser TLS fingerprint and
none of the Chrome client hints the main navigation sends. Bot managers that
score subresource requests (the AWS WAF challenge verify call, Akamai sensors)
read that as a script and reject it. Add StealthHttpClient::send_single and a
stealth_fetch_all path that op_fetch_url uses when a stealth client is
installed, preserving the redirect, SSRF, and CORS handling so JS-level
requests share the navigation's Chrome TLS fingerprint and sec-ch-ua client
hints. The reqwest path is unchanged for non-stealth. Adds a stealth feature to
obscura-js gated on obscura-net/stealth.
Stealth mode forced a Linux navigator UA and platform but reported Windows
in userAgentData.getHighEntropyValues and served Windows Direct3D11 WebGL
renderers, a combination no real Linux Chrome ever produces and that
cross-layer fingerprint checks flag. Present one consistent Windows Chrome
145 identity instead: the wreq emulation uses Platform::Windows, and the
navigator surfaces read the UA and platform from shared constants so the
HTTP User-Agent header, sec-ch-ua-platform, navigator.userAgent, and
userAgentData all agree. userAgentData brands are now derived from the
Chrome major version via Chromium's GREASE algorithm rather than a single
hardcoded token, and a Linux WebGL renderer pool is added so a Linux UA can
never report a Direct3D11 renderer.
Added Network.setBlockedURLs dispatch handling.
Stores blocked URL patterns on each page.
Applies blocking to page subresources and JS fetch / XHR requests.
Supports wildcard CDP-style patterns like *://*.gstatic.com/*.woff2.
--dump assets only walked static DOM attributes (script[src], img[src], ...),
so resources a page pulled in with fetch() or XHR never showed up (issue #301).
Record each absolute URL requested through op_fetch_url in ObscuraState,
expose it via the runtime and Page, and append those URLs to --dump assets as
type "fetch", deduped against URLs already listed from the DOM.
wreq upgraded to 6.0.0-rc.29 and wreq-util to 3.0.0-rc.12 to apply the
latest Chrome 145 TLS fingerprint (16 cipher suites, proper extension
order, ALPS, GREASE).
__obscura_stealth global variable introduced and passed from Rust to JS,
allowing stealth-specific JS behaviors to run only when the stealth flag
is enabled.
Tightened hardwareConcurrency and deviceMemory pools in stealth mode
to desktop-realistic values (e.g. min 4GB RAM, 4 cores), eliminating
mobile-tier values that leak the spoof.
navigator.userAgentData and getHighEntropyValues() are now overridden
in stealth mode to perfectly match the sec-ch-ua header sent by
wreq_util::Profile::Chrome145.
Fixed usedJSHeapSize sometimes exceeding totalJSHeapSize.
navigator.platform returned "Win32" while navigator.userAgent said Linux when
using the stealth client. page.rs now calls set_platform("Linux x86_64", "Linux", "")
alongside set_user_agent when the stealth client is active.
bootstrap.js: pre-declare 71 internal _* globals as enumerable:false before
var runs, so they don't show up in Object.keys(window) on real pages.
The geolocation shim returned a fixed Frankfurt position. With the timezone now
configurable, a deployment in another region would report a location that does
not match its exit IP or timezone, the mismatch issue #228 describes.
Set OBSCURA_GEOLOCATION="lat,lon" to override the reported coordinates; the
value is range-checked and falls back to the Frankfurt default when unset or
malformed. The existing positional jitter is preserved.
SGavrl