gh-83273: Rewrite csv.Sniffer dialect detection using trial parsing

Doc/library/csv.rst

@@ -312,6 +312,22 @@ The :mod:`!csv` module defines the following classes:
       is given, it is interpreted as a string containing possible valid
       delimiter characters.
 
+      The dialect is deduced by parsing the sample with every plausible
+      combination of parameters
+      and choosing the combination which splits the sample into rows
+      with the most consistent number of fields,
+      so the returned dialect is consistent with how :func:`reader`
+      will parse the sample.
+      Raise :exc:`Error` if no combination fits the sample,
+      in particular if it is a single column,
+      so there is no delimiter to find.
+
+      .. versionchanged:: next
+         The dialect is now deduced by trial parsing
+         and the results may differ from those of earlier Python versions.
+         The *escapechar* parameter can now be detected,
+         and the requested *delimiters* are not restricted to ASCII.
+
 
    .. method:: has_header(sample)
 

Doc/whatsnew/3.16.rst

@@ -96,6 +96,21 @@ codecs
   even when a built-in codec of the same name exists.
   (Contributed by Serhiy Storchaka in :gh:`152997`.)
 
+csv
+---
+
+* :meth:`csv.Sniffer.sniff` now deduces the dialect by trial parsing
+  instead of heuristics based on matching isolated fragments
+  and on character frequencies,
+  so the result is consistent with how :func:`csv.reader` will parse the sample.
+  It can now detect the *escapechar* parameter,
+  accepts non-ASCII delimiters in the *delimiters* argument,
+  no longer misdetects the delimiter
+  when the sample contains delimiter characters inside quoted fields,
+  and no longer takes quadratic time on quoted samples.
+  The results may differ from those of earlier Python versions.
+  (Contributed by Serhiy Storchaka in :gh:`83273`.)
+
 curses
 ------