Fix #6806: linkcheck: Failure on parsing content

CHANGES

@@ -48,6 +48,7 @@ Bugs fixed
   supported LaTeX engines: ¶, §, €, ∞, ±, →, ‣, –, superscript and subscript
   digits go through "as is" (as default OpenType font supports them)
 * #6704: linkcheck: Be defensive and handle newly defined HTTP error code
+* #6806: linkcheck: Failure on parsing content
 * #6655: image URLs containing ``data:`` causes gettext builder crashed
 * #6584: i18n: Error when compiling message catalogs on Hindi
 * #6718: i18n: KeyError is raised if section title and table title are same

sphinx/builders/linkcheck.py

@@ -59,6 +59,9 @@ def check_anchor(response: requests.requests.Response, anchor: str) -> bool:
     # Read file in chunks. If we find a matching anchor, we break
     # the loop early in hopes not to have to download the whole thing.
     for chunk in response.iter_content(chunk_size=4096, decode_unicode=True):
+        if isinstance(chunk, bytes):    # requests failed to decode
+            chunk = chunk.decode()      # manually try to decode it
+
         parser.feed(chunk)
         if parser.found:
             break