bug#9318: 23.3.50; The first call of encode-coding-region() returns wrong result

[Kazuhiro Ito]

> > I think relocation of buffer may cause the problem.
> 
> Does that help?
> 
> diff --git a/src/coding.c b/src/coding.c
> index 65c8a76..f34a023 100644
> --- a/src/coding.c
> +++ b/src/coding.c
> @@ -915,8 +915,8 @@ record_conversion_result (struct coding_system *coding,
>      }
>  }
> 
> -/* This wrapper macro is used to preserve validity of pointers into
> -   buffer text across calls to decode_char, which could cause
> +/* These wrapper macros are used to preserve validity of pointers into
> +   buffer text across calls to decode_char/encode_char, which could cause
>     relocation of buffers if it loads a charset map, because loading a
>     charset map allocates large structures.  */
>  #define CODING_DECODE_CHAR(coding, src, src_base, src_end, charset, code, c) 
> \
> @@ -935,6 +935,21 @@ record_conversion_result (struct coding_system *coding,
>       src_end += offset;                                                   \
>        }                                                                      
>      \
>    } while (0)
> +#define CODING_ENCODE_CHAR(coding, dst, dst_end, charset, c, code)   \
> +  do {                                                                       
> \
> +    charset_map_loaded = 0;                                          \
> +    code = ENCODE_CHAR (charset, c);                                 \
> +    if (charset_map_loaded)                                          \
> +      {                                                                      
> \
> +     const unsigned char *orig = coding->destination;                \
> +     EMACS_INT offset;                                               \
> +                                                                     \
> +     coding_set_destination (coding);                                \
> +     offset = coding->destination - orig;                            \
> +     dst += offset;                                                  \
> +     dst_end += offset;                                              \
> +      }                                                                      
> \
> +  } while (0)
> 
> 
>  /* If there are at least BYTES length of room at dst, allocate memory
> @@ -2652,7 +2667,7 @@ encode_coding_emacs_mule (struct coding_system *coding)
>           {
>             charset = CHARSET_FROM_ID (preferred_charset_id);
>             if (CHAR_CHARSET_P (c, charset))
> -             code = ENCODE_CHAR (charset, c);
> +             CODING_ENCODE_CHAR (coding, dst, dst_end, charset, c, code);
>             else
>               charset = char_charset (c, charset_list, &code);
>           }
> @@ -4185,7 +4200,8 @@ decode_coding_iso_2022 (struct coding_system *coding)
 
>  #define ENCODE_ISO_CHARACTER(charset, c)                                \
>    do {                                                                       
>    \
> -    int code = ENCODE_CHAR ((charset), (c));                            \
> +    int code;                                                                
>    \
> +    CODING_ENCODE_CHAR (coding, dst, dst_end, charset, c, code);        \
>                                                                          \
>      if (CHARSET_DIMENSION (charset) == 1)                               \
>        ENCODE_ISO_CHARACTER_DIMENSION1 ((charset), code);                \

Andreas' patch resolved the problem partially.  It resolved the problem on
NetBSD with '-O0' CFLAGS, but failed on NetBSD with '-O2' and Windows.

I confirmed that adding the protection of coding->dst_object to
Andreas' patch resolved the problem on NetBSD with '-O2' but not on
Windows.  I don't know whether it is incorrect way or is not enough.

--- src/coding.c        2011-07-01 11:03:55 +0000
+++ src/coding.c        2011-08-24 23:39:49 +0000
@@ -7397,10 +7436,15 @@
       setup_ccl_program (&cclspec.ccl, CODING_CCL_ENCODER (coding));
     }
   do {
+    struct gcpro gcpro1;
+    GCPRO1 (coding->dst_object);
+
     coding_set_source (coding);
     consume_chars (coding, translation_table, max_lookup);
     coding_set_destination (coding);
     (*(coding->encoder)) (coding);
+
+    UNGCPRO;
   } while (coding->consumed_char < coding->src_chars);
 
   if (BUFFERP (coding->dst_object) && coding->produced_char > 0)

-- 
Kazuhiro Ito