2

在 PostgreSQL 的 libpq 库(C API)中,我试图从以文本表示形式返回的 bytea 字段转换为原始二进制字符串。

例如,对于单个换行符,文本表示是"\\x0a"(即字符'\\'、和'x',以空字节终止)。'0''a'

根据文档,这个文本表示可以通过使用 PQunescapeBytea() 转换回二进制表示。但是,当我使用它时,我只是得到一个char *完全相同的,但删除了前导斜杠"x0a"。我究竟做错了什么?

char * value  = PGgetvalue(res, row, col);
size_t length = 0;
char * bytes  = PQunescapeBytea(value, &length);
// inspection of 'bytes' for length shows it's the same, with the '\' removed
4

1 回答 1

1

嗯,我正在开发的机器有 PostgreSQL 8.4,而服务器有 9.1,它正在做 bytea 的十六进制表示,客户端不理解。我需要去做:

SET bytea_output = escape

但是,由于我不想篡改用户设置(甚至可能从我下面改变),我选择提供一个后备实现并在字符串看起来像新格式时绕过 libpq。为简洁起见,删除了命名空间前缀。并不是说我要返回一个 Ruby 字符串,因此我的代码中有rb_*andVALUE东西。

/** Predicate to test of string is of the form \x0afe... */
#define NEW_HEX_P(s, len) (len > 2 && s[0] == '\\' && s[1] == 'x')

/** Lookup table for fast conversion of bytea hex strings to binary data */
static char * HexLookup;

/** Cast from a bytea to a String according to the new (PG 9.0) hex format */
static VALUE cast_bytea_hex(char * hex, size_t len) {
  if ((len % 2) != 0) {
    rb_raise(rb_eRuntimeError,
        "Bad hex value provided for bytea (length not divisible by 2)");
  }

  size_t   buflen = (len - 2) / 2;
  char   * buffer = malloc(sizeof(char) * buflen);
  char   * s      = hex + 2;
  char   * b      = buffer;

  if (buffer == NULL) {
    rb_raise(rb_eRuntimeError,
        "Failed to allocate %ld bytes for bytea conversion", buflen);
  }

  for (; *s; s += 2, ++b)
    *b = (HexLookup[*s] << 4) + (HexLookup[*(s + 1)]);

  VALUE str = rb_str_new(buffer, buflen);
  free(buffer);

  return str;
}

/** Cast from a bytea to a String according to a regular escape format */
static VALUE cast_bytea_escape(char * escaped, size_t len) {
  unsigned char * buffer  = PQunescapeBytea(escaped, &len);

  if (buffer == NULL) {
    rb_raise(rb_eRuntimeError,
        "Failed to allocate memory for PQunescapeBytea() conversion");
  }

  VALUE str = rb_str_new(buffer, len);
  PQfreemem(buffer);

  return str;
}

/** Get the value as a ruby type */
VALUE cast_value(PGresult * res, int row, int col) {
  if (PQgetisnull(res, row, col)) {
    return Qnil;
  }

  char * value  = PQgetvalue(res, row, col);
  int    length = PQgetlength(res, row, col);

  switch (PQftype(res, col)) {
    case INT2OID:
    case INT4OID:
    case INT8OID:
      return rb_cstr2inum(value, 10);

    case BOOLOID:
      return (value[0] == 't') ? Qtrue : Qfalse;

    case BYTEAOID:
      if (NEW_HEX_P(value, length)) {
        return cast_bytea_hex(value, length);
      } else {
        return cast_bytea_escape(value, length);
      }

    default:
      return rb_str_new(value, length);
  }
}

/* Initialize hex decoding lookup table. Must be invoked once, before use. */
void Init_casts(void) {
  HexLookup = malloc(sizeof(char) * 128);

  if (HexLookup == NULL) {
    rb_raise(rb_eRuntimeError,
        "Failed to allocate 128 bytes for internal lookup table");
  }

  char c;

  for (c = '\0'; c < '\x7f'; ++c)
    HexLookup[c] = 0; // Default to NULLs so we don't crash. May be a bad idea.

  for (c = '0'; c <= '9'; ++c)
    HexLookup[c] = c - '0';

  for (c = 'a'; c <= 'f'; ++c)
    HexLookup[c] = 10 + c - 'a';

  for (c = 'A'; c <= 'F'; ++c)
    HexLookup[c] = 10 + c - 'A';
}

所以基本上,当我尝试转换一个 OID 为 的值时BYTEAOID,我首先检查它是否看起来像新的十六进制格式,如果是,我对查找表执行解码,否则我只是PQunescapeBytea()照常进行。有点糟糕,但它是两害相权取其轻。

Init_casts();

VALUE rubyval = cast_value(res, 1, 6); // Get the ruby type in row 1, column 6
于 2012-09-21T16:24:11.090 回答