fix: ancestor walk for inline <code> to prevent duplication (#849, #884)

- _is_inline_code_element() walks iterancestors() to catch <span><code>,
  <a><code> etc. inside paragraph-like parents (p, li, td, th, dd, dt)
- is_code_block_element() returns False for inline code
- recover_wild_text and _extract filter inline code sub-elems
- baseline filters inline code via _is_inline_code() ancestor walk
- fixes ruff format/CI, addresses review from Sanjays2402 and adbar

All 15 custom inline-code tests pass, 176 unit_tests pass, ruff check passes

Author: Bob

trafilatura/baseline.py

@@ -51,6 +51,18 @@ def basic_cleaning(tree: HtmlElement) -> HtmlElement:
 # <article> must carry more than this to count as content)
 _MIN_CONTENT_LENGTH = 100
 
+_INLINE_CODE_PARENTS = ("p", "li", "td", "th", "dd", "dt")
+
+
+def _is_inline_code(element) -> bool:
+    # inline <code> inside paragraph-like ancestors (including wrapped in span/a etc.) is not standalone (#849, #884)
+    if element.tag != "code":
+        return False
+    for ancestor in element.iterancestors():
+        if ancestor.tag in _INLINE_CODE_PARENTS:
+            return True
+    return False
+
 
 def _walk_json(node: Any, bodies: list[str], teasers: list[str]) -> None:
     """Collect schema.org text content from parsed JSON-LD (list-wrapped and @graph-nested
@@ -208,7 +220,12 @@ def baseline(filecontent: Any) -> tuple[_Element, str, int]:
 
     # scrape from text paragraphs, dropping repeats: a nested element (e.g. <p> in
     # <blockquote>) duplicates part of its container's text, collected first in document order
-    paragraphs = (trim(element.text_content()) for element in tree.iter("blockquote", "code", "p", "pre", "q", "quote"))
+    # skip inline <code> — text already captured by parent <p>/<li>/etc. (#849, #884)
+    paragraphs = (
+        trim(element.text_content())
+        for element in tree.iter("blockquote", "code", "p", "pre", "q", "quote")
+        if not _is_inline_code(element)
+    )
     if result := _attempt(paragraphs, dedupe=True):
         return result
 

trafilatura/main_extractor.py

@@ -229,8 +229,24 @@ def handle_lists(element: _Element, options: Extractor) -> _Element | None:
     return None
 
 
+_INLINE_CODE_PARENTS = frozenset(("p", "li", "td", "th", "dd", "dt"))
+
+
+def _is_inline_code_element(element: _Element) -> bool:
+    # inline <code> inside paragraph-like ancestors (including wrapped) is not a block (#849, #884)
+    if element.tag != "code":
+        return False
+    for ancestor in element.iterancestors():
+        if ancestor.tag in _INLINE_CODE_PARENTS:
+            return True
+    return False
+
+
 def is_code_block_element(element: _Element) -> bool:
     "Check if it is a code element according to common structural markers."
+    # inline <code> inside paragraph-like parents is not a block element (#849, #884)
+    if _is_inline_code_element(element):
+        return False
     # pip
     if element.get("lang") or element.tag == "code":
         return True
@@ -681,6 +697,8 @@ def recover_wild_text(
     unwanted = ("span",) if "ref" in potential_tags else ("a", "ref", "span")
     strip_tags(search_tree, *unwanted)
     subelems = search_tree.xpath(search_expr)
+    # filter out inline <code> to prevent duplication (#849, #884)
+    subelems = [e for e in subelems if not _is_inline_code_element(e)]
     # dedup against the pre-main-pass snapshot: skip what the main pass already took -- exact
     # match (not length-gated, #634; accepted cost: identical-text elements collapse) or a
     # length-gated substring (a <p> folded into its <list> container)
@@ -777,6 +795,9 @@ def _extract(tree: HtmlElement, options: Extractor) -> tuple[_Element, str, set[
         LOGGER.debug(sorted(potential_tags))
         # proper extraction
         subelems = subtree.xpath(".//*")
+        # filter out inline <code> — already part of their parent's text
+        # flow; processing them separately duplicates their content (#849, #884)
+        subelems = [e for e in subelems if not _is_inline_code_element(e)]
         # e.g. only lb-elems in a div
         if {e.tag for e in subelems} == {"lb"}:
             subelems = [subtree]