- refuse cases for all eight probe payloads (noscript href and handler,
svg style, title and textarea, noembed, xmp, conditional comment)
- send fallback and editor refusal for the three that survive rendering
- escaped_send subtests check svg title/textarea sends with an x/net/html
parser oracle, scripting on and off
- accept cases for conditional comments, quoted CSS, a safe noscript
link and a literal comment opener in an attribute
- unsafeMarkupAt runs the existing HTML reading, then a tag reading in
which attribute quoting alone decides where a tag ends
(NextIsNotRawText after raw-text start tags, comment openers blanked)
- both readings share unsafeAttrs; existing refusals keep their message
and snippet
- svg/style end tag inside a quoted attribute is refused at check, save
and send; README and docs/services/mail.md describe the second reading
validateHTML ran (?i)\son[a-z]+\s*= over the rendered document, text
included, so a variable such as "buy one = get one free" stopped its own
message, and it is what made Online=Shop in the footer a send failure.
The tokenizer walk refuses every on* attribute, and since CR-02 it also
reads raw-text element and comment content as markup, so the regex is
dropped. Text shaped like a handler is now sent; such site names are shown
and saved, and the Branding refusal copy no longer mentions onclick=.
Single-URL attributes removed ASCII whitespace and control characters
before looking for a scheme, the list-valued, meta content, style and
comment checks did not, so values="java	script:..." and a tab-split
meta refresh passed. The stripping is factored out as stripURLSpace and
used by every scheme check.
The tokenizer-only walk read noscript, title, textarea, style, xmp and the
other raw-text elements as one text token and never looked inside, so an
unencoded javascript: link in <noscript> or <svg><title> passed the check
and was sent. A mail client renders with scripting off and parses svg
children as markup, so the raw content of those elements, and comment
data (Outlook conditional comments), is now checked again as markup;
nesting deeper than eight levels is refused.
- TestURLSchemeCheck: prose and scheme words in titles or queries pass; every
URL-bearing attribute, CSS, meta refresh and each entity-encoded or
whitespace-split script URL is refused with a snippet at the markup; a layout
override with javascript: falls back to the file version with inlining on
and off; the editor refuses it on content_html and CSS URLs on content_css
- TestInlineUnsafeOutput: a failing inlined result sends the style-block
version with one warning per layout
- TestFooterSiteNameLiteral: the footer prints the site name escaped and
literally on the default and system layouts; body stays Markdown
- check-phase35.sh GO_TESTS requires the three tests