Issue #19405: Fixed outdated comments in the _sre module.
diff --git a/Lib/sre_compile.py b/Lib/sre_compile.py
index bd40705..471753e 100644
--- a/Lib/sre_compile.py
+++ b/Lib/sre_compile.py
@@ -276,10 +276,10 @@
# set is constructed. Then, this bitmap is sliced into chunks of 256
# characters, duplicate chunks are eliminated, and each chunk is
# given a number. In the compiled expression, the charset is
-# represented by a 16-bit word sequence, consisting of one word for
-# the number of different chunks, a sequence of 256 bytes (128 words)
+# represented by a 32-bit word sequence, consisting of one word for
+# the number of different chunks, a sequence of 256 bytes (64 words)
# of chunk numbers indexed by their original chunk position, and a
-# sequence of chunks (16 words each).
+# sequence of 256-bit chunks (8 words each).
# Compression is normally good: in a typical charset, large ranges of
# Unicode will be either completely excluded (e.g. if only cyrillic
@@ -294,7 +294,7 @@
# In UCS-4 mode, the BIGCHARSET opcode still supports only subsets
# of the basic multilingual plane; an efficient representation
-# for all of UTF-16 has not yet been developed. This means,
+# for all of Unicode has not yet been developed. This means,
# in particular, that negated charsets cannot be represented as
# bigcharsets.