diff mbox series

[v3,1/5] Copy utf-7 module to modified-utf-7

Message ID 20210125090226.39967-2-mg@max.gautier.name
State Changes Requested
Headers show
Series iconv: module for IMAP-UTF-7 | expand

Commit Message

Max Gautier Jan. 25, 2021, 9:02 a.m. UTC
---
 iconvdata/Makefile         |   2 +-
 iconvdata/modified-utf-7.c | 531 +++++++++++++++++++++++++++++++++++++
 2 files changed, 532 insertions(+), 1 deletion(-)
 create mode 100644 iconvdata/modified-utf-7.c

Comments

Andreas Schwab Jan. 25, 2021, 9:31 a.m. UTC | #1
Why are you still using MODIFIED-UTF-7?

Andreas.
Max Gautier Jan. 25, 2021, 1:51 p.m. UTC | #2
I merged the name change on the other patches. I can extract them back
and merged the relevant bits on this one if needed.
Florian Weimer Feb. 7, 2021, 9:42 a.m. UTC | #3
* Max Gautier via Libc-alpha:

> I merged the name change on the other patches. I can extract them back
> and merged the relevant bits on this one if needed. 

Given that UTF-7 conversion (either variant) is not
performance-critical, I suggest to have just one implementation file.

You can use step->__data to keep track of which variant is active.  See
iconvdata/iso646.c for an example.  There is no need to allocate a
separate object; you can store the flag directly in the __data member.

You could remove the UTF7_ENCODE_OPTIONAL_CHARS from the existing UTF-7
codec in a first, separate patch.

Thanks,
Florian
Max Gautier Feb. 7, 2021, 12:29 p.m. UTC | #4
* Florian Weimer via Libc-alpha:
> Given that UTF-7 conversion (either variant) is not
> performance-critical, I suggest to have just one implementation file.
> 
> You can use step->__data to keep track of which variant is active.  See
> iconvdata/iso646.c for an example.  There is no need to allocate a
> separate object; you can store the flag directly in the __data member.

I'll work on that.
I might use some advice on specific parts. About the classification of
characters (direct or not etc) :
- utf-7.c
+ utf-7-imap.c
> -static const unsigned char direct_tab[128 / 8] =
> -  {
> -    0x00, 0x26, 0x00, 0x00, 0x81, 0xf3, 0xff, 0x87,
> -    0xfe, 0xff, 0xff, 0x07, 0xfe, 0xff, 0xff, 0x07
> -  };
> -
>  static int
>  isdirect (uint32_t ch)
>  {
> -  return (ch < 128 && ((direct_tab[ch >> 3] >> (ch & 7)) & 1));
> -}
> -
> -
> -/* The set of "direct and optional direct characters":
> -   A-Z a-z 0-9 ' ( ) , - . / : ? space tab lf cr
> -   ! " # $ % & * ; < = > @ [ ] ^ _ ` { | }
> -*/
> -
> -static const unsigned char xdirect_tab[128 / 8] =
> -  {
> -    0x00, 0x26, 0x00, 0x00, 0xff, 0xf7, 0xff, 0xff,
> -    0xff, 0xff, 0xff, 0xef, 0xff, 0xff, 0xff, 0x3f
> -  };
> -
> -static int
> -isxdirect (uint32_t ch)
> -{
> -  return (ch < 128 && ((xdirect_tab[ch >> 3] >> (ch & 7)) & 1));
> +  return ((ch == '\n' || ch == '\t' || ch == '\r')
> +		  || (ch >= 0x20 && ch <= 0x7e && ch != '&'));
>  }
>  
> -
> -/* The set of "extended base64 characters":
> -   A-Z a-z 0-9 + / -
> +/* The set of "modified base64 characters":
> +   A-Z a-z 0-9 + , -
>  */
>  
> -static const unsigned char xbase64_tab[128 / 8] =
> -  {
> -    0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xff, 0x03,
> -    0xfe, 0xff, 0xff, 0x07, 0xfe, 0xff, 0xff, 0x07
> -  };
> -
>  static int
> -isxbase64 (uint32_t ch)
> +ismbase64 (uint32_t ch)
>  {
> -  return (ch < 128 && ((xbase64_tab[ch >> 3] >> (ch & 7)) & 1));
> +  return ((ch >= 'a' && ch <= 'z')
> +			  || (ch >= 'A' && ch <= 'Z')
> +			  || (ch >= '0' && ch <= '9')
> +			  || (ch == '+' || ch == ','));
>  }

When I initially looked at utf-7.c, the use of the _tab arrays with
magic values and the subsequent shifting didn't make a lot of sense to
me, which is why I modified them like this for utf-7-imap. If they are
in the same file, it's probably better to use the same method.
So do you see any benefits to keeping the old method ?
Testing directly for the actual characters seems a lot more readable to
me, is shorter, and can be mapped to the RFC definition.

I didn't find the reason for using the current method by looking at the
git history. The only one I could think of is performance, but I don't
see how or what it would improve. If someone has some hints, let me
know.

 
> You could remove the UTF7_ENCODE_OPTIONAL_CHARS from the existing UTF-7
> codec in a first, separate patch.

Do you mean I should modify the utf-7 conversion to not encode the
optional chars ?  That would change the result of utf-7 conversions,
wouldn't it ? I'm not opposed to it, but isn't that going to break
things ?

Thanks
Florian Weimer Feb. 7, 2021, 12:34 p.m. UTC | #5
* Max Gautier via Libc-alpha:

> When I initially looked at utf-7.c, the use of the _tab arrays with
> magic values and the subsequent shifting didn't make a lot of sense to
> me, which is why I modified them like this for utf-7-imap. If they are
> in the same file, it's probably better to use the same method.
> So do you see any benefits to keeping the old method ?

I think the old method was an optimization in the long-gone times when
memory access was relatively fast compared to computation.

GCC has also improved since: it can use shifts and bit lookups to
combine comparisons against multiple constants into one check.

> Testing directly for the actual characters seems a lot more readable to
> me, is shorter, and can be mapped to the RFC definition.

Agreed.

>> You could remove the UTF7_ENCODE_OPTIONAL_CHARS from the existing UTF-7
>> codec in a first, separate patch.
>
> Do you mean I should modify the utf-7 conversion to not encode the
> optional chars ?  That would change the result of utf-7 conversions,
> wouldn't it ? I'm not opposed to it, but isn't that going to break
> things ?

Sorry, I meant to remove the constructs that enable this type of
compile-time configurability.  That is, remove the macro, but do not
change the generated code.

Thanks,
Florian
diff mbox series

Patch

diff --git a/iconvdata/Makefile b/iconvdata/Makefile
index c8c532a3e4..7f932e10ed 100644
--- a/iconvdata/Makefile
+++ b/iconvdata/Makefile
@@ -61,7 +61,7 @@  modules	:= ISO8859-1 ISO8859-2 ISO8859-3 ISO8859-4 ISO8859-5		 \
 	   IBM5347 IBM9030 IBM9066 IBM9448 IBM12712 IBM16804             \
 	   IBM1364 IBM1371 IBM1388 IBM1390 IBM1399 ISO_11548-1 MIK BRF	 \
 	   MAC-CENTRALEUROPE KOI8-RU ISO8859-9E				 \
-	   CP770 CP771 CP772 CP773 CP774
+	   CP770 CP771 CP772 CP773 CP774 MODIFIED-UTF-7
 
 # If lazy binding is disabled, use BIND_NOW for the gconv modules.
 ifeq ($(bind-now),yes)
diff --git a/iconvdata/modified-utf-7.c b/iconvdata/modified-utf-7.c
new file mode 100644
index 0000000000..fc6a8dfcfd
--- /dev/null
+++ b/iconvdata/modified-utf-7.c
@@ -0,0 +1,531 @@ 
+/* Conversion module for UTF-7.
+   Copyright (C) 2000-2020 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* UTF-7 is a legacy encoding used for transmitting Unicode within the
+   ASCII character set, used primarily by mail agents.  New programs
+   are encouraged to use UTF-8 instead.
+
+   UTF-7 is specified in RFC 2152 (and old RFC 1641, RFC 1642).  The
+   original Base64 encoding is defined in RFC 2045.  */
+
+#include <dlfcn.h>
+#include <gconv.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+
+/* Define this to 1 if you want the so-called "optional direct" characters
+      ! " # $ % & * ; < = > @ [ ] ^ _ ` { | }
+   to be encoded. Define to 0 if you want them to be passed straight
+   through, like the so-called "direct" characters.
+   We set this to 1 because it's safer.
+ */
+#define UTF7_ENCODE_OPTIONAL_CHARS 1
+
+
+/* The set of "direct characters":
+   A-Z a-z 0-9 ' ( ) , - . / : ? space tab lf cr
+*/
+
+static const unsigned char direct_tab[128 / 8] =
+  {
+    0x00, 0x26, 0x00, 0x00, 0x81, 0xf3, 0xff, 0x87,
+    0xfe, 0xff, 0xff, 0x07, 0xfe, 0xff, 0xff, 0x07
+  };
+
+static int
+isdirect (uint32_t ch)
+{
+  return (ch < 128 && ((direct_tab[ch >> 3] >> (ch & 7)) & 1));
+}
+
+
+/* The set of "direct and optional direct characters":
+   A-Z a-z 0-9 ' ( ) , - . / : ? space tab lf cr
+   ! " # $ % & * ; < = > @ [ ] ^ _ ` { | }
+*/
+
+static const unsigned char xdirect_tab[128 / 8] =
+  {
+    0x00, 0x26, 0x00, 0x00, 0xff, 0xf7, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xef, 0xff, 0xff, 0xff, 0x3f
+  };
+
+static int
+isxdirect (uint32_t ch)
+{
+  return (ch < 128 && ((xdirect_tab[ch >> 3] >> (ch & 7)) & 1));
+}
+
+
+/* The set of "extended base64 characters":
+   A-Z a-z 0-9 + / -
+*/
+
+static const unsigned char xbase64_tab[128 / 8] =
+  {
+    0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xff, 0x03,
+    0xfe, 0xff, 0xff, 0x07, 0xfe, 0xff, 0xff, 0x07
+  };
+
+static int
+isxbase64 (uint32_t ch)
+{
+  return (ch < 128 && ((xbase64_tab[ch >> 3] >> (ch & 7)) & 1));
+}
+
+
+/* Converts a value in the range 0..63 to a base64 encoded char.  */
+static unsigned char
+base64 (unsigned int i)
+{
+  if (i < 26)
+    return i + 'A';
+  else if (i < 52)
+    return i - 26 + 'a';
+  else if (i < 62)
+    return i - 52 + '0';
+  else if (i == 62)
+    return '+';
+  else if (i == 63)
+    return '/';
+  else
+    abort ();
+}
+
+
+/* Definitions used in the body of the `gconv' function.  */
+#define CHARSET_NAME		"UTF-7//"
+#define DEFINE_INIT		1
+#define DEFINE_FINI		1
+#define FROM_LOOP		from_utf7_loop
+#define TO_LOOP			to_utf7_loop
+#define MIN_NEEDED_FROM		1
+#define MAX_NEEDED_FROM		6
+#define MIN_NEEDED_TO		4
+#define MAX_NEEDED_TO		4
+#define ONE_DIRECTION		0
+#define PREPARE_LOOP \
+  mbstate_t saved_state;						      \
+  mbstate_t *statep = data->__statep;
+#define EXTRA_LOOP_ARGS		, statep
+
+
+/* Since we might have to reset input pointer we must be able to save
+   and restore the state.  */
+#define SAVE_RESET_STATE(Save) \
+  if (Save)								      \
+    saved_state = *statep;						      \
+  else									      \
+    *statep = saved_state
+
+
+/* First define the conversion function from UTF-7 to UCS4.
+   The state is structured as follows:
+     __count bit 2..0: zero
+     __count bit 8..3: shift
+     __wch: data
+   Precise meaning:
+     shift      data
+       0         --          not inside base64 encoding
+     1..32  XX..XX00..00     inside base64, (32 - shift) bits pending
+   This state layout is simpler than relying on STORE_REST/UNPACK_BYTES.
+
+   When shift = 0, __wch needs to store at most one lookahead byte (see
+   __GCONV_INCOMPLETE_INPUT below).
+*/
+#define MIN_NEEDED_INPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_INPUT	MAX_NEEDED_FROM
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_TO
+#define MAX_NEEDED_OUTPUT	MAX_NEEDED_TO
+#define LOOPFCT			FROM_LOOP
+#define BODY \
+  {									      \
+    uint_fast8_t ch = *inptr;						      \
+									      \
+    if ((statep->__count >> 3) == 0)					      \
+      {									      \
+	/* base64 encoding inactive.  */				      \
+	if (isxdirect (ch))						      \
+	  {								      \
+	    inptr++;							      \
+	    put32 (outptr, ch);						      \
+	    outptr += 4;						      \
+	  }								      \
+	else if (__glibc_likely (ch == '+'))				      \
+	  {								      \
+	    if (__glibc_unlikely (inptr + 2 > inend))			      \
+	      {								      \
+		/* Not enough input available.  */			      \
+		result = __GCONV_INCOMPLETE_INPUT;			      \
+		break;							      \
+	      }								      \
+	    if (inptr[1] == '-')					      \
+	      {								      \
+		inptr += 2;						      \
+		put32 (outptr, ch);					      \
+		outptr += 4;						      \
+	      }								      \
+	    else							      \
+	      {								      \
+		/* Switch into base64 mode.  */				      \
+		inptr++;						      \
+		statep->__count = (32 << 3);				      \
+		statep->__value.__wch = 0;				      \
+	      }								      \
+	  }								      \
+	else								      \
+	  {								      \
+	    /* The input is invalid.  */				      \
+	    STANDARD_FROM_LOOP_ERR_HANDLER (1);				      \
+	  }								      \
+      }									      \
+    else								      \
+      {									      \
+	/* base64 encoding active.  */					      \
+	uint32_t i;							      \
+	int shift;							      \
+									      \
+	if (ch >= 'A' && ch <= 'Z')					      \
+	  i = ch - 'A';							      \
+	else if (ch >= 'a' && ch <= 'z')				      \
+	  i = ch - 'a' + 26;						      \
+	else if (ch >= '0' && ch <= '9')				      \
+	  i = ch - '0' + 52;						      \
+	else if (ch == '+')						      \
+	  i = 62;							      \
+	else if (ch == '/')						      \
+	  i = 63;							      \
+	else								      \
+	  {								      \
+	    /* Terminate base64 encoding.  */				      \
+									      \
+	    /* If accumulated data is nonzero, the input is invalid.  */      \
+	    /* Also, partial UTF-16 characters are invalid.  */		      \
+	    if (__builtin_expect (statep->__value.__wch != 0, 0)	      \
+		|| __builtin_expect ((statep->__count >> 3) <= 26, 0))	      \
+	      {								      \
+		STANDARD_FROM_LOOP_ERR_HANDLER ((statep->__count = 0, 1));    \
+	      }								      \
+									      \
+	    if (ch == '-')						      \
+	      inptr++;							      \
+									      \
+	    statep->__count = 0;					      \
+	    continue;							      \
+	  }								      \
+									      \
+	/* Concatenate the base64 integer i to the accumulator.  */	      \
+	shift = (statep->__count >> 3);					      \
+	if (shift > 6)							      \
+	  {								      \
+	    uint32_t wch;						      \
+									      \
+	    shift -= 6;							      \
+	    wch = statep->__value.__wch | (i << shift);			      \
+									      \
+	    if (shift <= 16 && shift > 10)				      \
+	      {								      \
+		/* An UTF-16 character has just been completed.  */	      \
+		uint32_t wc1 = wch >> 16;				      \
+									      \
+		/* UTF-16: When we see a High Surrogate, we must also decode  \
+		   the following Low Surrogate. */			      \
+		if (!(wc1 >= 0xd800 && wc1 < 0xdc00))			      \
+		  {							      \
+		    wch = wch << 16;					      \
+		    shift += 16;					      \
+		    put32 (outptr, wc1);				      \
+		    outptr += 4;					      \
+		  }							      \
+	      }								      \
+	    else if (shift <= 10 && shift > 4)				      \
+	      {								      \
+		/* After a High Surrogate, verify that the next 16 bit	      \
+		   indeed form a Low Surrogate.  */			      \
+		uint32_t wc2 = wch & 0xffff;				      \
+									      \
+		if (! __builtin_expect (wc2 >= 0xdc00 && wc2 < 0xe000, 1))    \
+		  {							      \
+		    STANDARD_FROM_LOOP_ERR_HANDLER ((statep->__count = 0, 1));\
+		  }							      \
+	      }								      \
+									      \
+	    statep->__value.__wch = wch;				      \
+	  }								      \
+	else								      \
+	  {								      \
+	    /* An UTF-16 surrogate pair has just been completed.  */	      \
+	    uint32_t wc1 = (uint32_t) statep->__value.__wch >> 16;	      \
+	    uint32_t wc2 = ((uint32_t) statep->__value.__wch & 0xffff)	      \
+			   | (i >> (6 - shift));			      \
+									      \
+	    statep->__value.__wch = (i << shift) << 26;			      \
+	    shift += 26;						      \
+									      \
+	    assert (wc1 >= 0xd800 && wc1 < 0xdc00);			      \
+	    assert (wc2 >= 0xdc00 && wc2 < 0xe000);			      \
+	    put32 (outptr,						      \
+		   0x10000 + ((wc1 - 0xd800) << 10) + (wc2 - 0xdc00));	      \
+	    outptr += 4;						      \
+	  }								      \
+									      \
+	statep->__count = shift << 3;					      \
+									      \
+	/* Now that we digested the input increment the input pointer.  */    \
+	inptr++;							      \
+      }									      \
+  }
+#define LOOP_NEED_FLAGS
+#define EXTRA_LOOP_DECLS	, mbstate_t *statep
+#include <iconv/loop.c>
+
+
+/* Next, define the conversion from UCS4 to UTF-7.
+   The state is structured as follows:
+     __count bit 2..0: zero
+     __count bit 4..3: shift
+     __count bit 8..5: data
+   Precise meaning:
+     shift      data
+       0         0           not inside base64 encoding
+       1         0           inside base64, no pending bits
+       2       XX00          inside base64, 2 bits known for next byte
+       3       XXXX          inside base64, 4 bits known for next byte
+
+   __count bit 2..0 and __wch are always zero, because this direction
+   never returns __GCONV_INCOMPLETE_INPUT.
+*/
+#define MIN_NEEDED_INPUT	MIN_NEEDED_TO
+#define MAX_NEEDED_INPUT	MAX_NEEDED_TO
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_OUTPUT	MAX_NEEDED_FROM
+#define LOOPFCT			TO_LOOP
+#define BODY \
+  {									      \
+    uint32_t ch = get32 (inptr);					      \
+									      \
+    if ((statep->__count & 0x18) == 0)					      \
+      {									      \
+	/* base64 encoding inactive */					      \
+	if (UTF7_ENCODE_OPTIONAL_CHARS ? isdirect (ch) : isxdirect (ch))      \
+	  {								      \
+	    *outptr++ = (unsigned char) ch;				      \
+	  }								      \
+	else								      \
+	  {								      \
+	    size_t count;						      \
+									      \
+	    if (ch == '+')						      \
+	      count = 2;						      \
+	    else if (ch < 0x10000)					      \
+	      count = 3;						      \
+	    else if (ch < 0x110000)					      \
+	      count = 6;						      \
+	    else							      \
+	      STANDARD_TO_LOOP_ERR_HANDLER (4);				      \
+									      \
+	    if (__glibc_unlikely (outptr + count > outend))		      \
+	      {								      \
+		result = __GCONV_FULL_OUTPUT;				      \
+		break;							      \
+	      }								      \
+									      \
+	    *outptr++ = '+';						      \
+	    if (ch == '+')						      \
+	      *outptr++ = '-';						      \
+	    else if (ch < 0x10000)					      \
+	      {								      \
+		*outptr++ = base64 (ch >> 10);				      \
+		*outptr++ = base64 ((ch >> 4) & 0x3f);			      \
+		statep->__count = ((ch & 15) << 5) | (3 << 3);		      \
+	      }								      \
+	    else if (ch < 0x110000)					      \
+	      {								      \
+		uint32_t ch1 = 0xd800 + ((ch - 0x10000) >> 10);		      \
+		uint32_t ch2 = 0xdc00 + ((ch - 0x10000) & 0x3ff);	      \
+									      \
+		ch = (ch1 << 16) | ch2;					      \
+		*outptr++ = base64 (ch >> 26);				      \
+		*outptr++ = base64 ((ch >> 20) & 0x3f);			      \
+		*outptr++ = base64 ((ch >> 14) & 0x3f);			      \
+		*outptr++ = base64 ((ch >> 8) & 0x3f);			      \
+		*outptr++ = base64 ((ch >> 2) & 0x3f);			      \
+		statep->__count = ((ch & 3) << 7) | (2 << 3);		      \
+	      }								      \
+	    else							      \
+	      abort ();							      \
+	  }								      \
+      }									      \
+    else								      \
+      {									      \
+	/* base64 encoding active */					      \
+	if (UTF7_ENCODE_OPTIONAL_CHARS ? isdirect (ch) : isxdirect (ch))      \
+	  {								      \
+	    /* deactivate base64 encoding */				      \
+	    size_t count;						      \
+									      \
+	    count = ((statep->__count & 0x18) >= 0x10) + isxbase64 (ch) + 1;  \
+	    if (__glibc_unlikely (outptr + count > outend))		      \
+	      {								      \
+		result = __GCONV_FULL_OUTPUT;				      \
+		break;							      \
+	      }								      \
+									      \
+	    if ((statep->__count & 0x18) >= 0x10)			      \
+	      *outptr++ = base64 ((statep->__count >> 3) & ~3);		      \
+	    if (isxbase64 (ch))						      \
+	      *outptr++ = '-';						      \
+	    *outptr++ = (unsigned char) ch;				      \
+	    statep->__count = 0;					      \
+	  }								      \
+	else								      \
+	  {								      \
+	    size_t count;						      \
+									      \
+	    if (ch < 0x10000)						      \
+	      count = ((statep->__count & 0x18) >= 0x10 ? 3 : 2);	      \
+	    else if (ch < 0x110000)					      \
+	      count = ((statep->__count & 0x18) >= 0x18 ? 6 : 5);	      \
+	    else							      \
+	      STANDARD_TO_LOOP_ERR_HANDLER (4);				      \
+									      \
+	    if (__glibc_unlikely (outptr + count > outend))		      \
+	      {								      \
+		result = __GCONV_FULL_OUTPUT;				      \
+		break;							      \
+	      }								      \
+									      \
+	    if (ch < 0x10000)						      \
+	      {								      \
+		switch ((statep->__count >> 3) & 3)			      \
+		  {							      \
+		  case 1:						      \
+		    *outptr++ = base64 (ch >> 10);			      \
+		    *outptr++ = base64 ((ch >> 4) & 0x3f);		      \
+		    statep->__count = ((ch & 15) << 5) | (3 << 3);	      \
+		    break;						      \
+		  case 2:						      \
+		    *outptr++ =						      \
+		      base64 (((statep->__count >> 3) & ~3) | (ch >> 12));    \
+		    *outptr++ = base64 ((ch >> 6) & 0x3f);		      \
+		    *outptr++ = base64 (ch & 0x3f);			      \
+		    statep->__count = (1 << 3);				      \
+		    break;						      \
+		  case 3:						      \
+		    *outptr++ =						      \
+		      base64 (((statep->__count >> 3) & ~3) | (ch >> 14));    \
+		    *outptr++ = base64 ((ch >> 8) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 2) & 0x3f);		      \
+		    statep->__count = ((ch & 3) << 7) | (2 << 3);	      \
+		    break;						      \
+		  default:						      \
+		    abort ();						      \
+		  }							      \
+	      }								      \
+	    else if (ch < 0x110000)					      \
+	      {								      \
+		uint32_t ch1 = 0xd800 + ((ch - 0x10000) >> 10);		      \
+		uint32_t ch2 = 0xdc00 + ((ch - 0x10000) & 0x3ff);	      \
+									      \
+		ch = (ch1 << 16) | ch2;					      \
+		switch ((statep->__count >> 3) & 3)			      \
+		  {							      \
+		  case 1:						      \
+		    *outptr++ = base64 (ch >> 26);			      \
+		    *outptr++ = base64 ((ch >> 20) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 14) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 8) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 2) & 0x3f);		      \
+		    statep->__count = ((ch & 3) << 7) | (2 << 3);	      \
+		    break;						      \
+		  case 2:						      \
+		    *outptr++ =						      \
+		      base64 (((statep->__count >> 3) & ~3) | (ch >> 28));    \
+		    *outptr++ = base64 ((ch >> 22) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 16) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 10) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 4) & 0x3f);		      \
+		    statep->__count = ((ch & 15) << 5) | (3 << 3);	      \
+		    break;						      \
+		  case 3:						      \
+		    *outptr++ =						      \
+		      base64 (((statep->__count >> 3) & ~3) | (ch >> 30));    \
+		    *outptr++ = base64 ((ch >> 24) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 18) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 12) & 0x3f);		      \
+		    *outptr++ = base64 ((ch >> 6) & 0x3f);		      \
+		    *outptr++ = base64 (ch & 0x3f);			      \
+		    statep->__count = (1 << 3);				      \
+		    break;						      \
+		  default:						      \
+		    abort ();						      \
+		  }							      \
+	      }								      \
+	    else							      \
+	      abort ();							      \
+	  }								      \
+      }									      \
+									      \
+    /* Now that we wrote the output increment the input pointer.  */	      \
+    inptr += 4;								      \
+  }
+#define LOOP_NEED_FLAGS
+#define EXTRA_LOOP_DECLS	, mbstate_t *statep
+#include <iconv/loop.c>
+
+
+/* Since this is a stateful encoding we have to provide code which resets
+   the output state to the initial state.  This has to be done during the
+   flushing.  */
+#define EMIT_SHIFT_TO_INIT \
+  if (FROM_DIRECTION)							      \
+    /* Nothing to emit.  */						      \
+    memset (data->__statep, '\0', sizeof (mbstate_t));			      \
+  else									      \
+    {									      \
+      /* The "to UTF-7" direction.  Flush the remaining bits and terminate    \
+	 with a '-' byte.  This will guarantee correct decoding if more	      \
+	 UTF-7 encoded text is added afterwards.  */			      \
+      int state = data->__statep->__count;				      \
+									      \
+      if (state & 0x18)							      \
+	{								      \
+	  /* Deactivate base64 encoding.  */				      \
+	  size_t count = ((state & 0x18) >= 0x10) + 1;			      \
+									      \
+	  if (__glibc_unlikely (outbuf + count > outend))		      \
+	    /* We don't have enough room in the output buffer.  */	      \
+	    status = __GCONV_FULL_OUTPUT;				      \
+	  else								      \
+	    {								      \
+	      /* Write out the shift sequence.  */			      \
+	      if ((state & 0x18) >= 0x10)				      \
+		*outbuf++ = base64 ((state >> 3) & ~3);			      \
+	      *outbuf++ = '-';						      \
+									      \
+	      data->__statep->__count = 0;				      \
+	    }								      \
+	}								      \
+      else								      \
+	data->__statep->__count = 0;					      \
+    }
+
+
+/* Now define the toplevel functions.  */
+#include <iconv/skeleton.c>