summaryrefslogtreecommitdiff
path: root/src/utf8.cc
diff options
context:
space:
mode:
Diffstat (limited to 'src/utf8.cc')
-rw-r--r--src/utf8.cc12
1 files changed, 6 insertions, 6 deletions
diff --git a/src/utf8.cc b/src/utf8.cc
index 3d25f4f5..71b4159d 100644
--- a/src/utf8.cc
+++ b/src/utf8.cc
@@ -16,7 +16,7 @@
// C++ functions with C linkage ----------------------------------------------
-/*
+/**
* Return index of the last byte of the UTF-8-encoded character that str + i
* points to or into.
*/
@@ -36,7 +36,7 @@ uint_t a_Utf8_end_of_char(const char *str, uint_t i)
return i;
}
-/*
+/**
* Decode a single UTF-8-encoded character starting at p.
* The resulting Unicode value (in the range 0-0x10ffff) is returned,
* and len is set to the number of bytes in the UTF-8 encoding.
@@ -48,7 +48,7 @@ uint_t a_Utf8_decode(const char* str, const char* end, int* len)
return fl_utf8decode(str, end, len);
}
-/*
+/**
* Write UTF-8 encoding of ucs into buf and return number of bytes written.
*/
int a_Utf8_encode(unsigned int ucs, char *buf)
@@ -56,7 +56,7 @@ int a_Utf8_encode(unsigned int ucs, char *buf)
return fl_utf8encode(ucs, buf);
}
-/*
+/**
* Examine first srclen bytes of src.
* Return 0 if not legal UTF-8, 1 if all ASCII, 2 if all below 0x800,
* 3 if all below 0x10000, and 4 otherwise.
@@ -66,8 +66,8 @@ int a_Utf8_test(const char* src, unsigned int srclen)
return fl_utf8test(src, srclen);
}
-/*
- * Does s point to a UTF-8-encoded ideographic character?
+/**
+ * Does s point to a UTF-8-encoded ideographic character?.
*
* This is based on http://unicode.org/reports/tr14/#ID plus some guesses
* for what might make the most sense for Dillo. Surprisingly, they include