`bootstrap.js` publishes its DOM interfaces with `globalThis.Node = Node`,
which defaults to `enumerable: true`. A real browser exposes every WebIDL
interface on the global as `enumerable: false`, so the difference is a
one-liner for any script that cares:
Object.getOwnPropertyDescriptor(window, 'Node').enumerable
// real Chrome: false obscura: true
It also means `Object.keys(window)` and `for...in` over the global enumerate
22 names a browser would never list.
The file already has the mechanism for this -- `_preHideInternals()`
pre-declares internal globals with `enumerable: false`, and its own comment
notes that later `globalThis.x = value` assignments only update the value and
leave the descriptor intact. This just adds the WebIDL names to that list, so
no assignment site changes.
Before: 9 of 10 sampled interfaces enumerable. After: 0 of 10, matching Chrome.
Regression: interfaces still callable (`typeof Node === 'function'`),
`instanceof Node` / `instanceof Element` hold, `querySelector` works, Hacker
News still yields 30 results under `--stealth`, snapshot builds clean.
Composes with #537 -- verified with both applied: `var Node` pages run, the
JSF/RichFaces case loads (jQuery/A4J/RichFaces all live), and enumerability
stays at 0/10.
The #525 fix landed on the reqwest path (`client.rs`: `configured_root_paths()`
-> `add_root_certificate`) but did not cover `wreq_client.rs`. The stealth
client builds its `wreq::Client` without a certificate store, so those
variables are silently ignored whenever `--stealth` is on.
The consequence is inverted: the transport with the best TLS fingerprint is the
only one that cannot reach a host behind a private or incomplete CA chain --
precisely the environments #525 described.
Same binary on current main, same private-CA server, only the flag differs:
SSL_CERT_FILE=ca.pem obscura fetch https://127.0.0.1:8443/ -> ok
SSL_CERT_FILE=ca.pem obscura --stealth fetch https://127.0.0.1:8443/ -> fails
`wreq` already exposes the hook -- `ClientBuilder::tls_cert_store(CertStore)`,
and `CertStore::builder().set_default_paths()` reads both variables. It is
simply never called.
Two constraints, both of which fail silently if got wrong:
* `tls_cert_store`, not `tls_options`. `emulation()` is documented as
"This will overwrite the existing configuration" and internally calls
`self.tls_options(emulation.tls_options)`; setting TLS options here would
discard the Chrome fingerprint, which is the whole point of this client.
`tls_cert_store` is a separate config field and composes with emulation.
* Opt-in only. Supplying a store REPLACES the webpki roots (see
`set_cert_store` in `tls/conn/ext.rs`), it does not add to them. Applying it
unconditionally would break ordinary sites whenever the bundle is
incomplete. With neither variable set, behaviour is byte-for-byte unchanged.
Verified: with the variable, a private-CA host resolves under --stealth;
without it, the same host still fails; with a deliberately EMPTY bundle,
example.com fails -- which is what proves the store is actually applied rather
than the bundle merely happening to be right.
Refs #538, #525
reqwest was pinned to rustls-tls, so HTTPS only trusted the bundled
webpki roots: SSL_CERT_FILE, SSL_CERT_DIR, and the system CA store were
ignored, and obscura-net exposes no add-CA hook. Deployments behind a
TLS-inspecting egress proxy or an internal CA cannot fetch at all, and a
caching proxy in front of --proxy cannot terminate TLS.
Enable rustls-tls-native-roots alongside rustls-tls. reqwest merges the
two root stores into one, so the default trust surface only grows and
certificate verification is otherwise unchanged: unknown CAs are still
rejected. The stealth (wreq) client has its own TLS stack and is out of
scope here.
Tests: an in-process HTTPS fixture with an rcgen-minted private CA,
served via tokio-rustls (pinned to the ring provider so the test binary
has a single rustls CryptoProvider). With SSL_CERT_FILE pointing at the
CA the fetch succeeds; without it the same fixture is rejected.
Fixes#525.
Obscura lacked these two standard event constructors on globalThis.
The missing PromiseRejectionEvent caused core-js to misdetect the
environment and override native Promise with a broken polyfill,
breaking Vue application rendering. StorageEvent completes the
localStorage/sessionStorage API surface.
Both are also registered in document.createEvent() for legacy callers.
Fixes#521, fixes#522.
resolveOrFallback(undefined) returned undefined, so applyVirtual set
__virtualUrl to null and location reverted to the document URL. Per the HTML
spec a missing url keeps the current document URL; return __currentUrl() so the
new history entry captures the current location instead of clearing it.
Fixes#496
before(), after() and replaceWith() only converted string arguments to Text
nodes; other primitives (numbers, booleans, null) were passed straight to
insert_before with no _nid, corrupting the call. Route them through the shared
_convertNodes helper (the same path append/prepend use), so every non-node
argument becomes document.createTextNode(String(n)).
Fixes#498
The old sub-millisecond bump (_sub += 0.1) was uncapped, so after >10 calls in
one millisecond it exceeded 1.0; when the wall clock then ticked over and the
bump reset to 0, now() returned a value below the previous call. Track the last
returned value and never go below it (advance by a tiny epsilon when the clock
has not moved past it).
Fixes#497
set_quirks_mode was a no-op and the selector MatchingContext hardcoded
QuirksMode::NoQuirks, so class/id selectors were always case-sensitive even on
quirks-mode (no-doctype) pages where browsers match .foo / #bar
ASCII-case-insensitively. Store the parser-reported quirks mode on DomTree, pass
it into the selector context, and skip the exact-case #id index fast path in
quirks mode so it falls through to case-insensitive matching.
Fixes#499
remove() freed a node and its descendants by directly indexing inner.nodes and
unconditionally pushing each slot onto the free list. An out-of-range id panicked
out-of-bounds, and removing an already-freed node pushed its slot a second time,
so two later allocations could be handed the same NodeId (aliasing/corruption).
Free a slot only when it is currently live, via checked access.
Fixes#500
cloneNode(true) round-tripped through a throwaway <div>'s innerHTML, so the
HTML fragment parser discarded elements that are not valid children of <div>
(<tr>, <td>, <option>, <tbody>, …) — createElement('tr').cloneNode(true)
returned null — and JS-set inline styles were lost. The shallow path also
re-derived attributes by regex-scraping serialized HTML.
Clone structurally with real DOM APIs: shallow-clone each node via
createElement/createElementNS plus a getAttributeNames() attribute copy (no
parsing context, every attribute preserved), then descend into children with an
explicit stack so a deep subtree cannot overflow the JS stack. Text, Comment,
DocumentFragment and <template> contents are handled.
Fixes#490
The MCP HTTP server serves connections sequentially on the single
current-thread runtime (the browser session is !Send). The SSE branch
(GET /mcp, Accept: text/event-stream) ran an infinite keep-alive ping loop
inline, so handle_connection never returned and the accept loop never ran
again — one open SSE stream hung every later POST /mcp tool call
indefinitely (functional break + trivial unauthenticated DoS).
The keep-alive carries no browser state and its write half is Send + 'static,
so spawn the ping loop on its own task and return from the handler. POST
handling stays serialized (correct for a single browser session), but a live
SSE stream no longer blocks the accept loop.
Fixes#489
CSSStyleDeclaration was backed by an in-memory _props map that was never
bridged to the element's style content attribute. Parsed inline styles were
invisible to el.style.*, setAttribute('style', …) did not update el.style, and
el.style.x = … never reached the attribute or serialization; getComputedStyle
read the same disconnected declaration.
Give the declaration its owner element: lazily pull the style attribute into
_props on read (skipping the reparse when unchanged) and push _props back to the
attribute on every mutation. Owner-less declarations (getComputedStyle fallback,
stylesheet rules) stay purely in-memory, so their behavior is unchanged.
Fixes#488
get namespaceURI() derived the namespace from the tag name and only ever
recognised the <svg> root, so every element parsed inside an <svg> subtree
reported the XHTML namespace. get href() gates the SVGAnimatedString
reflection on the SVG namespace, so for parsed content the gate never
opened: href returned a plain string and .animVal/.baseVal were undefined,
making new URL(el.href.animVal) throw. Only createElementNS-built elements
worked, because createElementNS records _ns explicitly.
The tree builder already applies the HTML5 foreign-content rules, so
QualName.ns is correct for parsed SVG; it just was not exposed to JS. Add a
namespace_uri DOM op and read it in get namespaceURI(), keeping _ns as the
override and the old tag-name heuristic as the fallback when no element name
is recorded. Namespaces are immutable, so the result is cached like _lname.
Fixes#484
The template.content bridge (#463) routed every element through
content_source(), which called get_node().clone() to read the Copy
template_contents field. That deep-cloned each element's Node (its attrs
Vec plus a String per attribute) on every innerHTML/outerHTML/dump-html
serialization, once per non-void element.
template_contents is already destructured from the node in the serialize
loop, so read it there directly. content_source() (inner_html path) now
reads it through with_node() by reference. Serialization output is
byte-identical; ~10% faster on serialization-heavy pages.
SGavrl