MarkdownRenderer.codespan always wrapped the content in a single backtick
pair, so a code span whose content contains backticks re-parsed wrong: the
inner backtick closed the span early, and a lone backtick rendered as an
empty fenced code block.

Size the delimiter to one more than the longest backtick run in the content,
and pad with a space on each side when the content begins or ends with a
backtick (the parser strips that padding back off), matching parse_codespan
and CommonMark.

CommonMark rules 9 and 10 require the multiple-of-3 delimiter test to use
the lengths of the original delimiter runs, not the lengths remaining after
a run has been partially consumed. mistune mutated _Delimiter.length as it
consumed markers and then reused that mutated value in the mod-3 test, so the
leftover of a partially consumed run (e.g. the ** left after *a* closes in
'*a***a*') could no longer match a later delimiter.

Track the original run length on _Delimiter and use it in
_can_match_emphasis_delimiters. Fixes cases such as '*a***a*' and
'*foo***bar*', which two CommonMark reference implementations
(markdown-it-py, commonmark.py) render as <em>a</em>*<em>a</em> and
<em>foo</em>*<em>bar</em>.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

…ength

Fix emphasis multiple-of-3 rule mis-rendering (e.g. *foo***bar*)

Fix MarkdownRenderer re-rendering of code spans containing backticks

…blocks (#456)

* fix: set prev token in render_list_item and add block_text to ignore_blocks

The RSTRenderer.block_quote method accesses token.prev to decide
whether to emit a .. separator before the blockquote.  When a
blockquote appears inside a list item, render_list_item called
renderer.render_token directly without tracking prev, so
blockquote tokens never received a prev reference.  This caused a
KeyError in older versions and silently degraded output in the
current version (missing or incorrect .. separators).

- Track prev in render_list_item, mirroring how RSTRenderer.iter_tokens
  already does for top-level tokens.
- Add block_text (the list-internal paragraph token type) to the
  ignore_blocks tuple so a blockquote following list-internal text
  does not get a spurious .. separator.

* Test RST list item blockquote separation

A title containing a double quote was emitted unescaped inside double
quotes, e.g. [t](/u 'a"b') re-rendered as [t](/u "a"b"). On a
re-parse the embedded quote closes the title early, so the title is
silently dropped (or the link fails to parse). Escape backslashes and
the closing quote when writing inline-link, image and reference-link
titles so they survive a round trip.

…nk-title-quote

Escape quotes in MarkdownRenderer link and image titles

MarkdownRenderer.block_quote indented every line with '> ' (including the
blank line a paragraph leaves behind) and then ran rstrip('> \n') to drop
that dangling marker. Because rstrip strips a character set, it also ate a
'>' that ended the quote's own content, so an autolink, an inline HTML tag,
or a literal '>' lost its final character (e.g. '> <https://x.com>' became
'> <https://x.com', which reparses as escaped text).

Strip the children's trailing blank lines before indenting instead, so the
marker is never added to an empty trailing line and there is nothing to peel
back off.

…-trailing-gt

MarkdownRenderer.text() only escaped backticks, so a text token that is a
literal "*"/"_" emphasis delimiter was emitted unescaped and re-parsed as
emphasis on the round-trip. This silently changed the document's meaning,
violating the invariant asserted by TestMarkdownRendererRoundTrip
("reformatting valid Markdown must not change its meaning").

An escaped marker in the source (``\*not emphasis\*``) is stored by the
inline parser as bare "*" text tokens (the backslash is consumed and the
parser's internal no-emphasis flag is dropped before the AST). Re-emitting
those tokens without a backslash let them pair up again:

    reformat(r"\*not emphasis\*")  ->  "*not emphasis*"
    html(reformat(...))            ->  "<p><em>not emphasis</em></p>"   # was <p>*not emphasis*</p>

The same happened for "_", "**"/"__", and unmatched leftover markers.

Fix: in text(), if a text token consists entirely of "*"/"_" it is a
literal delimiter (an escaped marker or an unmatched leftover), so
backslash-escape each character. Prose punctuation such as "2 * 3" or
"snake_case" always arrives mixed with other characters, so it is left
untouched and is not over-escaped (guarded by the existing
test_prose_not_over_escaped and a new test_real_emphasis_not_over_escaped).

Added test_escaped_emphasis_markers and test_real_emphasis_not_over_escaped
to TestMarkdownRendererRoundTrip. The regression test fails without the
source change (asserts <em>...</em> reappears) and passes with it; full
suite 1122 passed, ruff check/format and mypy clean.

…-emphasis