diff --git a/src/mjson.c b/src/mjson.c
new file mode 100644
index 0000000..549b0d5
--- /dev/null
+++ b/src/mjson.c
@@ -0,0 +1,1048 @@
+// Copyright (c) 2018-2020 Cesanta Software Limited
+// All rights reserved
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+#include <float.h>
+#include <math.h>
+
+#include <import/mjson.h>
+
+#if defined(_MSC_VER)
+#define alloca(x) _alloca(x)
+#endif
+
+#if defined(_MSC_VER) && _MSC_VER < 1700
+#define va_copy(x, y) (x) = (y)
+#define isinf(x) !_finite(x)
+#define isnan(x) _isnan(x)
+#endif
+
+static double mystrtod(const char *str, char **end);
+
+static int mjson_esc(int c, int esc) {
+  const char *p, *esc1 = "\b\f\n\r\t\\\"", *esc2 = "bfnrt\\\"";
+  for (p = esc ? esc1 : esc2; *p != '\0'; p++) {
+    if (*p == c) return esc ? esc2[p - esc1] : esc1[p - esc2];
+  }
+  return 0;
+}
+
+static int mjson_escape(int c) {
+  return mjson_esc(c, 1);
+}
+
+static int mjson_pass_string(const char *s, int len) {
+  int i;
+  for (i = 0; i < len; i++) {
+    if (s[i] == '\\' && i + 1 < len && mjson_escape(s[i + 1])) {
+      i++;
+    } else if (s[i] == '\0') {
+      return MJSON_ERROR_INVALID_INPUT;
+    } else if (s[i] == '"') {
+      return i;
+    }
+  }
+  return MJSON_ERROR_INVALID_INPUT;
+}
+
+int mjson(const char *s, int len, mjson_cb_t cb, void *ud) {
+  enum { S_VALUE, S_KEY, S_COLON, S_COMMA_OR_EOO } expecting = S_VALUE;
+  unsigned char nesting[MJSON_MAX_DEPTH];
+  int i, depth = 0;
+#define MJSONCALL(ev) \
+  if (cb != NULL && cb(ev, s, start, i - start + 1, ud)) return i + 1;
+
+// In the ascii table, the distance between `[` and `]` is 2.
+// Ditto for `{` and `}`. Hence +2 in the code below.
+#define MJSONEOO()                                                     \
+  do {                                                                 \
+    if (c != nesting[depth - 1] + 2) return MJSON_ERROR_INVALID_INPUT; \
+    depth--;                                                           \
+    if (depth == 0) {                                                  \
+      MJSONCALL(tok);                                                  \
+      return i + 1;                                                    \
+    }                                                                  \
+  } while (0)
+
+  for (i = 0; i < len; i++) {
+    int start = i;
+    unsigned char c = ((unsigned char *) s)[i];
+    int tok = c;
+    if (c == ' ' || c == '\t' || c == '\n' || c == '\r') continue;
+    // printf("- %c [%.*s] %d %d\n", c, i, s, depth, expecting);
+    switch (expecting) {
+      case S_VALUE:
+        if (c == '{') {
+          if (depth >= (int) sizeof(nesting)) return MJSON_ERROR_TOO_DEEP;
+          nesting[depth++] = c;
+          expecting = S_KEY;
+          break;
+        } else if (c == '[') {
+          if (depth >= (int) sizeof(nesting)) return MJSON_ERROR_TOO_DEEP;
+          nesting[depth++] = c;
+          break;
+        } else if (c == ']' && depth > 0) {  // Empty array
+          MJSONEOO();
+        } else if (c == 't' && i + 3 < len && memcmp(&s[i], "true", 4) == 0) {
+          i += 3;
+          tok = MJSON_TOK_TRUE;
+        } else if (c == 'n' && i + 3 < len && memcmp(&s[i], "null", 4) == 0) {
+          i += 3;
+          tok = MJSON_TOK_NULL;
+        } else if (c == 'f' && i + 4 < len && memcmp(&s[i], "false", 5) == 0) {
+          i += 4;
+          tok = MJSON_TOK_FALSE;
+        } else if (c == '-' || ((c >= '0' && c <= '9'))) {
+          char *end = NULL;
+          mystrtod(&s[i], &end);
+          if (end != NULL) i += (int) (end - &s[i] - 1);
+          tok = MJSON_TOK_NUMBER;
+        } else if (c == '"') {
+          int n = mjson_pass_string(&s[i + 1], len - i - 1);
+          if (n < 0) return n;
+          i += n + 1;
+          tok = MJSON_TOK_STRING;
+        } else {
+          return MJSON_ERROR_INVALID_INPUT;
+        }
+        if (depth == 0) {
+          MJSONCALL(tok);
+          return i + 1;
+        }
+        expecting = S_COMMA_OR_EOO;
+        break;
+
+      case S_KEY:
+        if (c == '"') {
+          int n = mjson_pass_string(&s[i + 1], len - i - 1);
+          if (n < 0) return n;
+          i += n + 1;
+          tok = MJSON_TOK_KEY;
+          expecting = S_COLON;
+        } else if (c == '}') {  // Empty object
+          MJSONEOO();
+          expecting = S_COMMA_OR_EOO;
+        } else {
+          return MJSON_ERROR_INVALID_INPUT;
+        }
+        break;
+
+      case S_COLON:
+        if (c == ':') {
+          expecting = S_VALUE;
+        } else {
+          return MJSON_ERROR_INVALID_INPUT;
+        }
+        break;
+
+      case S_COMMA_OR_EOO:
+        if (depth <= 0) return MJSON_ERROR_INVALID_INPUT;
+        if (c == ',') {
+          expecting = (nesting[depth - 1] == '{') ? S_KEY : S_VALUE;
+        } else if (c == ']' || c == '}') {
+          MJSONEOO();
+        } else {
+          return MJSON_ERROR_INVALID_INPUT;
+        }
+        break;
+    }
+    MJSONCALL(tok);
+  }
+  return MJSON_ERROR_INVALID_INPUT;
+}
+
+struct msjon_get_data {
+  const char *path;     // Lookup json path
+  int pos;              // Current path index
+  int d1;               // Current depth of traversal
+  int d2;               // Expected depth of traversal
+  int i1;               // Index in an array
+  int i2;               // Expected index in an array
+  int obj;              // If the value is array/object, offset where it starts
+  const char **tokptr;  // Destination
+  int *toklen;          // Destination length
+  int tok;              // Returned token
+};
+
+#include <stdio.h>
+
+static int plen1(const char *s) {
+  int i = 0, n = 0;
+  while (s[i] != '\0' && s[i] != '.' && s[i] != '[')
+    n++, i += s[i] == '\\' ? 2 : 1;
+  // printf("PLEN: s: [%s], [%.*s] => %d\n", s, i, s, n);
+  return n;
+}
+
+static int plen2(const char *s) {
+  int i = 0, n = 0;
+  while (s[i] != '\0' && s[i] != '.' && s[i] != '[')
+    n++, i += s[i] == '\\' ? 2 : 1;
+  // printf("PLEN: s: [%s], [%.*s] => %d\n", s, i, s, n);
+  return i;
+}
+
+static int kcmp(const char *a, const char *b, int n) {
+  int i = 0, j = 0, r = 0;
+  for (i = 0, j = 0; j < n; i++, j++) {
+    if (b[i] == '\\') i++;
+    if ((r = a[j] - b[i]) != 0) return r;
+  }
+  // printf("KCMP: a: [%.*s], b:[%.*s] ==> %d\n", n, a, i, b, r);
+  return r;
+}
+
+static int mjson_get_cb(int tok, const char *s, int off, int len, void *ud) {
+  struct msjon_get_data *data = (struct msjon_get_data *) ud;
+  // printf("--> %2x %2d %2d %2d %2d\t'%s'\t'%.*s'\t\t'%.*s'\n", tok, data->d1,
+  // data->d2, data->i1, data->i2, data->path + data->pos, off, s, len,
+  // s + off);
+  if (data->tok != MJSON_TOK_INVALID) return 1;  // Found
+
+  if (tok == '{') {
+    if (!data->path[data->pos] && data->d1 == data->d2) data->obj = off;
+    data->d1++;
+  } else if (tok == '[') {
+    if (data->d1 == data->d2 && data->path[data->pos] == '[') {
+      data->i1 = 0;
+      data->i2 = (int) mystrtod(&data->path[data->pos + 1], NULL);
+      if (data->i1 == data->i2) {
+        data->d2++;
+        data->pos += 3;
+      }
+    }
+    if (!data->path[data->pos] && data->d1 == data->d2) data->obj = off;
+    data->d1++;
+  } else if (tok == ',') {
+    if (data->d1 == data->d2 + 1) {
+      data->i1++;
+      if (data->i1 == data->i2) {
+        while (data->path[data->pos] != ']') data->pos++;
+        data->pos++;
+        data->d2++;
+      }
+    }
+  } else if (tok == MJSON_TOK_KEY && data->d1 == data->d2 + 1 &&
+             data->path[data->pos] == '.' && s[off] == '"' &&
+             s[off + len - 1] == '"' &&
+             plen1(&data->path[data->pos + 1]) == len - 2 &&
+             kcmp(s + off + 1, &data->path[data->pos + 1], len - 2) == 0) {
+    data->d2++;
+    data->pos += plen2(&data->path[data->pos + 1]) + 1;
+  } else if (tok == MJSON_TOK_KEY && data->d1 == data->d2) {
+    return 1;  // Exhausted path, not found
+  } else if (tok == '}' || tok == ']') {
+    data->d1--;
+    // data->d2--;
+    if (!data->path[data->pos] && data->d1 == data->d2 && data->obj != -1) {
+      data->tok = tok - 2;
+      if (data->tokptr) *data->tokptr = s + data->obj;
+      if (data->toklen) *data->toklen = off - data->obj + 1;
+      return 1;
+    }
+  } else if (MJSON_TOK_IS_VALUE(tok)) {
+    // printf("TOK --> %d\n", tok);
+    if (data->d1 == data->d2 && !data->path[data->pos]) {
+      data->tok = tok;
+      if (data->tokptr) *data->tokptr = s + off;
+      if (data->toklen) *data->toklen = len;
+      return 1;
+    }
+  }
+  return 0;
+}
+
+enum mjson_tok mjson_find(const char *s, int len, const char *jp,
+                          const char **tokptr, int *toklen) {
+  struct msjon_get_data data = {jp, 1,  0,      0,      0,
+                                0,  -1, tokptr, toklen, MJSON_TOK_INVALID};
+  if (jp[0] != '$') return MJSON_TOK_INVALID;
+  if (mjson(s, len, mjson_get_cb, &data) < 0) return MJSON_TOK_INVALID;
+  return (enum mjson_tok) data.tok;
+}
+
+int mjson_get_number(const char *s, int len, const char *path, double *v) {
+  const char *p;
+  int tok, n;
+  if ((tok = mjson_find(s, len, path, &p, &n)) == MJSON_TOK_NUMBER) {
+    if (v != NULL) *v = mystrtod(p, NULL);
+  }
+  return tok == MJSON_TOK_NUMBER ? 1 : 0;
+}
+
+int mjson_get_bool(const char *s, int len, const char *path, int *v) {
+  int tok = mjson_find(s, len, path, NULL, NULL);
+  if (tok == MJSON_TOK_TRUE && v != NULL) *v = 1;
+  if (tok == MJSON_TOK_FALSE && v != NULL) *v = 0;
+  return tok == MJSON_TOK_TRUE || tok == MJSON_TOK_FALSE ? 1 : 0;
+}
+
+static unsigned char mjson_unhex_nimble(const char *s) {
+  unsigned char i, v = 0;
+  for (i = 0; i < 2; i++) {
+    int c = s[i];
+    if (i > 0) v <<= 4;
+    v |= (c >= '0' && c <= '9') ? c - '0'
+                                : (c >= 'A' && c <= 'F') ? c - '7' : c - 'W';
+  }
+  return v;
+}
+
+static int mjson_unescape(const char *s, int len, char *to, int n) {
+  int i, j;
+  for (i = 0, j = 0; i < len && j < n; i++, j++) {
+    if (s[i] == '\\' && i + 5 < len && s[i + 1] == 'u') {
+      //  \uXXXX escape. We could process a simple one-byte chars
+      // \u00xx from the ASCII range. More complex chars would require
+      // dragging in a UTF8 library, which is too much for us
+      if (s[i + 2] != '0' || s[i + 3] != '0') return -1;  // Too much, give up
+      to[j] = mjson_unhex_nimble(s + i + 4);
+      i += 5;
+    } else if (s[i] == '\\' && i + 1 < len) {
+      int c = mjson_esc(s[i + 1], 0);
+      if (c == 0) return -1;
+      to[j] = c;
+      i++;
+    } else {
+      to[j] = s[i];
+    }
+  }
+  if (j >= n) return -1;
+  if (n > 0) to[j] = '\0';
+  return j;
+}
+
+int mjson_get_string(const char *s, int len, const char *path, char *to,
+                     int n) {
+  const char *p;
+  int sz;
+  if (mjson_find(s, len, path, &p, &sz) != MJSON_TOK_STRING) return -1;
+  return mjson_unescape(p + 1, sz - 2, to, n);
+}
+
+int mjson_get_hex(const char *s, int len, const char *x, char *to, int n) {
+  const char *p;
+  int i, j, sz;
+  if (mjson_find(s, len, x, &p, &sz) != MJSON_TOK_STRING) return -1;
+  for (i = j = 0; i < sz - 3 && j < n; i += 2, j++) {
+    ((unsigned char *) to)[j] = mjson_unhex_nimble(p + i + 1);
+  }
+  if (j < n) to[j] = '\0';
+  return j;
+}
+
+#if MJSON_ENABLE_BASE64
+static int mjson_base64rev(int c) {
+  if (c >= 'A' && c <= 'Z') {
+    return c - 'A';
+  } else if (c >= 'a' && c <= 'z') {
+    return c + 26 - 'a';
+  } else if (c >= '0' && c <= '9') {
+    return c + 52 - '0';
+  } else if (c == '+') {
+    return 62;
+  } else if (c == '/') {
+    return 63;
+  } else {
+    return 64;
+  }
+}
+
+int mjson_base64_dec(const char *src, int n, char *dst, int dlen) {
+  const char *end = src + n;
+  int len = 0;
+  while (src + 3 < end && len < dlen) {
+    int a = mjson_base64rev(src[0]), b = mjson_base64rev(src[1]),
+        c = mjson_base64rev(src[2]), d = mjson_base64rev(src[3]);
+    dst[len++] = (a << 2) | (b >> 4);
+    if (src[2] != '=' && len < dlen) {
+      dst[len++] = (b << 4) | (c >> 2);
+      if (src[3] != '=' && len < dlen) {
+        dst[len++] = (c << 6) | d;
+      }
+    }
+    src += 4;
+  }
+  if (len < dlen) dst[len] = '\0';
+  return len;
+}
+
+int mjson_get_base64(const char *s, int len, const char *path, char *to,
+                     int n) {
+  const char *p;
+  int sz;
+  if (mjson_find(s, len, path, &p, &sz) != MJSON_TOK_STRING) return 0;
+  return mjson_base64_dec(p + 1, sz - 2, to, n);
+}
+#endif  // MJSON_ENABLE_BASE64
+
+#if MJSON_ENABLE_NEXT
+struct nextdata {
+  int off, len, depth, t, vo, arrayindex;
+  int *koff, *klen, *voff, *vlen, *vtype;
+};
+
+static int next_cb(int tok, const char *s, int off, int len, void *ud) {
+  struct nextdata *d = (struct nextdata *) ud;
+  // int i;
+  switch (tok) {
+    case '{':
+    case '[':
+      if (d->depth == 0 && tok == '[') d->arrayindex = 0;
+      if (d->depth == 1 && off > d->off) {
+        d->vo = off;
+        d->t = tok == '{' ? MJSON_TOK_OBJECT : MJSON_TOK_ARRAY;
+        if (d->voff) *d->voff = off;
+        if (d->vtype) *d->vtype = d->t;
+      }
+      d->depth++;
+      break;
+    case '}':
+    case ']':
+      d->depth--;
+      if (d->depth == 1 && d->vo) {
+        d->len = off + len;
+        if (d->vlen) *d->vlen = d->len - d->vo;
+        if (d->arrayindex >= 0) {
+          if (d->koff) *d->koff = d->arrayindex;  // koff holds array index
+          if (d->klen) *d->klen = 0;              // klen holds 0
+        }
+        return 1;
+      }
+      if (d->depth == 1 && d->arrayindex >= 0) d->arrayindex++;
+      break;
+    case ',':
+    case ':':
+      break;
+    case MJSON_TOK_KEY:
+      if (d->depth == 1 && d->off < off) {
+        if (d->koff) *d->koff = off;  // And report back to the user
+        if (d->klen) *d->klen = len;  // If we have to
+      }
+      break;
+    default:
+      if (d->depth != 1) break;
+      // If we're iterating over the array
+      if (off > d->off) {
+        d->len = off + len;
+        if (d->vlen) *d->vlen = len;    // value length
+        if (d->voff) *d->voff = off;    // value offset
+        if (d->vtype) *d->vtype = tok;  // value type
+        if (d->arrayindex >= 0) {
+          if (d->koff) *d->koff = d->arrayindex;  // koff holds array index
+          if (d->klen) *d->klen = 0;              // klen holds 0
+        }
+        return 1;
+      }
+      if (d->arrayindex >= 0) d->arrayindex++;
+      break;
+  }
+  (void) s;
+  return 0;
+}
+
+int mjson_next(const char *s, int n, int off, int *koff, int *klen, int *voff,
+               int *vlen, int *vtype) {
+  struct nextdata d = {off, 0, 0, 0, 0, -1, koff, klen, voff, vlen, vtype};
+  mjson(s, n, next_cb, &d);
+  return d.len;
+}
+#endif
+
+#if MJSON_ENABLE_PRINT
+int mjson_print_fixed_buf(const char *ptr, int len, void *fndata) {
+  struct mjson_fixedbuf *fb = (struct mjson_fixedbuf *) fndata;
+  int i, left = fb->size - 1 - fb->len;
+  if (left < len) len = left;
+  for (i = 0; i < len; i++) fb->ptr[fb->len + i] = ptr[i];
+  fb->len += len;
+  fb->ptr[fb->len] = '\0';
+  return len;
+}
+
+// This function allocates memory in chunks of size MJSON_DYNBUF_CHUNK
+// to decrease memory fragmentation, when many calls are executed to
+// print e.g. a base64 string or a hex string.
+int mjson_print_dynamic_buf(const char *ptr, int len, void *fndata) {
+  char *s, *buf = *(char **) fndata;
+  size_t curlen = buf == NULL ? 0 : strlen(buf);
+  size_t new_size = curlen + len + 1 + MJSON_DYNBUF_CHUNK;
+  new_size -= new_size % MJSON_DYNBUF_CHUNK;
+
+  if ((s = (char *) realloc(buf, new_size)) == NULL) {
+    return 0;
+  } else {
+    memcpy(s + curlen, ptr, len);
+    s[curlen + len] = '\0';
+    *(char **) fndata = s;
+    return len;
+  }
+}
+
+int mjson_print_null(const char *ptr, int len, void *userdata) {
+  (void) ptr;
+  (void) userdata;
+  return len;
+}
+
+int mjson_print_buf(mjson_print_fn_t fn, void *fnd, const char *buf, int len) {
+  return fn(buf, len, fnd);
+}
+
+int mjson_print_long(mjson_print_fn_t fn, void *fnd, long val, int is_signed) {
+  unsigned long v = val, s = 0, n, i;
+  char buf[20], t;
+  if (is_signed && val < 0) {
+    buf[s++] = '-', v = -val;
+  }
+  // This loop prints a number in reverse order. I guess this is because we
+  // write numbers from right to left: least significant digit comes last.
+  // Maybe because we use Arabic numbers, and Arabs write RTL?
+  for (n = 0; v > 0; v /= 10) buf[s + n++] = "0123456789"[v % 10];
+  // Reverse a string
+  for (i = 0; i < n / 2; i++)
+    t = buf[s + i], buf[s + i] = buf[s + n - i - 1], buf[s + n - i - 1] = t;
+  if (val == 0) buf[n++] = '0';  // Handle special case
+  return fn(buf, s + n, fnd);
+}
+
+int mjson_print_int(mjson_print_fn_t fn, void *fnd, int v, int s) {
+  return mjson_print_long(fn, fnd, s ? (long) v : (unsigned) v, s);
+}
+
+static int addexp(char *buf, int e, int sign) {
+  int n = 0;
+  buf[n++] = 'e';
+  buf[n++] = sign;
+  if (e > 400) return 0;
+  if (e < 10) buf[n++] = '0';
+  if (e >= 100) buf[n++] = (e / 100) + '0', e -= 100 * (e / 100);
+  if (e >= 10) buf[n++] = (e / 10) + '0', e -= 10 * (e / 10);
+  buf[n++] = e + '0';
+  return n;
+}
+
+int mjson_print_dbl(mjson_print_fn_t fn, void *fnd, double d, int width) {
+  char buf[40];
+  int i, s = 0, n = 0, e = 0;
+  double t, mul, saved;
+  if (d == 0.0) return fn("0", 1, fnd);
+  if (isinf(d)) return fn(d > 0 ? "inf" : "-inf", d > 0 ? 3 : 4, fnd);
+  if (isnan(d)) return fn("nan", 3, fnd);
+  if (d < 0.0) d = -d, buf[s++] = '-';
+
+  // Round
+  saved = d;
+  mul = 1.0;
+  while (d >= 10.0 && d / mul >= 10.0) mul *= 10.0;
+  while (d <= 1.0 && d / mul <= 1.0) mul /= 10.0;
+  for (i = 0, t = mul * 5; i < width; i++) t /= 10.0;
+  d += t;
+  // Calculate exponent, and 'mul' for scientific representation
+  mul = 1.0;
+  while (d >= 10.0 && d / mul >= 10.0) mul *= 10.0, e++;
+  while (d < 1.0 && d / mul < 1.0) mul /= 10.0, e--;
+  // printf(" --> %g %d %g %g\n", saved, e, t, mul);
+
+  if (e >= width) {
+    struct mjson_fixedbuf fb = {buf + s, (int) sizeof(buf) - s, 0};
+    n = mjson_print_dbl(mjson_print_fixed_buf, &fb, saved / mul, width);
+    // printf(" --> %.*g %d [%.*s]\n", 10, d / t, e, fb.len, fb.ptr);
+    n += addexp(buf + s + n, e, '+');
+    return fn(buf, s + n, fnd);
+  } else if (e <= -width) {
+    struct mjson_fixedbuf fb = {buf + s, (int) sizeof(buf) - s, 0};
+    n = mjson_print_dbl(mjson_print_fixed_buf, &fb, saved / mul, width);
+    // printf(" --> %.*g %d [%.*s]\n", 10, d / mul, e, fb.len, fb.ptr);
+    n += addexp(buf + s + n, -e, '-');
+    return fn(buf, s + n, fnd);
+  } else {
+    for (i = 0, t = mul; d >= 1.0 && s + n < (int) sizeof(buf); i++) {
+      int ch = (int) (d / t);
+      if (n > 0 || ch > 0) buf[s + n++] = ch + '0';
+      d -= ch * t;
+      t /= 10.0;
+    }
+    // printf(" --> [%g] -> %g %g (%d) [%.*s]\n", saved, d, t, n, s + n, buf);
+    if (n == 0) buf[s++] = '0';
+    while (t >= 1.0 && n + s < (int) sizeof(buf)) buf[n++] = '0', t /= 10.0;
+    if (s + n < (int) sizeof(buf)) buf[n + s++] = '.';
+    // printf(" 1--> [%g] -> [%.*s]\n", saved, s + n, buf);
+    for (i = 0, t = 0.1; s + n < (int) sizeof(buf) && n < width; i++) {
+      int ch = (int) (d / t);
+      buf[s + n++] = ch + '0';
+      d -= ch * t;
+      t /= 10.0;
+    }
+  }
+  while (n > 0 && buf[s + n - 1] == '0') n--;  // Trim trailing zeros
+  if (n > 0 && buf[s + n - 1] == '.') n--;     // Trim trailing dot
+  return fn(buf, s + n, fnd);
+}
+
+int mjson_print_str(mjson_print_fn_t fn, void *fnd, const char *s, int len) {
+  int i, n = fn("\"", 1, fnd);
+  for (i = 0; i < len; i++) {
+    char c = mjson_escape(s[i]);
+    if (c) {
+      n += fn("\\", 1, fnd);
+      n += fn(&c, 1, fnd);
+    } else {
+      n += fn(&s[i], 1, fnd);
+    }
+  }
+  return n + fn("\"", 1, fnd);
+}
+
+#if MJSON_ENABLE_BASE64
+int mjson_print_b64(mjson_print_fn_t fn, void *fnd, const unsigned char *s,
+                    int n) {
+  const char *t =
+      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+  int i, len = fn("\"", 1, fnd);
+  for (i = 0; i < n; i += 3) {
+    int a = s[i], b = i + 1 < n ? s[i + 1] : 0, c = i + 2 < n ? s[i + 2] : 0;
+    char buf[4] = {t[a >> 2], t[(a & 3) << 4 | (b >> 4)], '=', '='};
+    if (i + 1 < n) buf[2] = t[(b & 15) << 2 | (c >> 6)];
+    if (i + 2 < n) buf[3] = t[c & 63];
+    len += fn(buf, sizeof(buf), fnd);
+  }
+  return len + fn("\"", 1, fnd);
+}
+#endif /* MJSON_ENABLE_BASE64 */
+
+int mjson_vprintf(mjson_print_fn_t fn, void *fnd, const char *fmt,
+                  va_list xap) {
+  int i = 0, n = 0;
+  va_list ap;
+  va_copy(ap, xap);
+  while (fmt[i] != '\0') {
+    if (fmt[i] == '%') {
+      char fc = fmt[++i];
+      int is_long = 0;
+      if (fc == 'l') {
+        is_long = 1;
+        fc = fmt[i + 1];
+      }
+      if (fc == 'Q') {
+        char *buf = va_arg(ap, char *);
+        n += mjson_print_str(fn, fnd, buf ? buf : "",
+                             buf ? (int) strlen(buf) : 0);
+      } else if (strncmp(&fmt[i], ".*Q", 3) == 0) {
+        int len = va_arg(ap, int);
+        char *buf = va_arg(ap, char *);
+        n += mjson_print_str(fn, fnd, buf, len);
+        i += 2;
+      } else if (fc == 'd' || fc == 'u') {
+        int is_signed = (fc == 'd');
+        if (is_long) {
+          long val = va_arg(ap, long);
+          n += mjson_print_long(fn, fnd, val, is_signed);
+          i++;
+        } else {
+          int val = va_arg(ap, int);
+          n += mjson_print_int(fn, fnd, val, is_signed);
+        }
+      } else if (fc == 'B') {
+        const char *s = va_arg(ap, int) ? "true" : "false";
+        n += mjson_print_buf(fn, fnd, s, (int) strlen(s));
+      } else if (fc == 's') {
+        char *buf = va_arg(ap, char *);
+        n += mjson_print_buf(fn, fnd, buf, (int) strlen(buf));
+      } else if (strncmp(&fmt[i], ".*s", 3) == 0) {
+        int len = va_arg(ap, int);
+        char *buf = va_arg(ap, char *);
+        n += mjson_print_buf(fn, fnd, buf, len);
+        i += 2;
+      } else if (fc == 'g') {
+        n += mjson_print_dbl(fn, fnd, va_arg(ap, double), 6);
+      } else if (strncmp(&fmt[i], ".*g", 3) == 0) {
+        int width = va_arg(ap, int);
+        n += mjson_print_dbl(fn, fnd, va_arg(ap, double), width);
+        i += 2;
+#if MJSON_ENABLE_BASE64
+      } else if (fc == 'V') {
+        int len = va_arg(ap, int);
+        const char *buf = va_arg(ap, const char *);
+        n += mjson_print_b64(fn, fnd, (unsigned char *) buf, len);
+#endif
+      } else if (fc == 'H') {
+        const char *hex = "0123456789abcdef";
+        int i, len = va_arg(ap, int);
+        const unsigned char *p = va_arg(ap, const unsigned char *);
+        n += fn("\"", 1, fnd);
+        for (i = 0; i < len; i++) {
+          n += fn(&hex[(p[i] >> 4) & 15], 1, fnd);
+          n += fn(&hex[p[i] & 15], 1, fnd);
+        }
+        n += fn("\"", 1, fnd);
+      } else if (fc == 'M') {
+        mjson_vprint_fn_t vfn = va_arg(ap, mjson_vprint_fn_t);
+        n += vfn(fn, fnd, &ap);
+      }
+      i++;
+    } else {
+      n += mjson_print_buf(fn, fnd, &fmt[i++], 1);
+    }
+  }
+  va_end(xap);
+  va_end(ap);
+  return n;
+}
+
+int mjson_printf(mjson_print_fn_t fn, void *fnd, const char *fmt, ...) {
+  va_list ap;
+  int len;
+  va_start(ap, fmt);
+  len = mjson_vprintf(fn, fnd, fmt, ap);
+  va_end(ap);
+  return len;
+}
+#endif /* MJSON_ENABLE_PRINT */
+
+static int is_digit(int c) {
+  return c >= '0' && c <= '9';
+}
+
+/* NOTE: strtod() implementation by Yasuhiro Matsumoto. */
+static double mystrtod(const char *str, char **end) {
+  double d = 0.0;
+  int sign = 1, n = 0;
+  const char *p = str, *a = str;
+
+  /* decimal part */
+  if (*p == '-') {
+    sign = -1;
+    ++p;
+  } else if (*p == '+') {
+    ++p;
+  }
+  if (is_digit(*p)) {
+    d = (double) (*p++ - '0');
+    while (*p && is_digit(*p)) {
+      d = d * 10.0 + (double) (*p - '0');
+      ++p;
+      ++n;
+    }
+    a = p;
+  } else if (*p != '.') {
+    goto done;
+  }
+  d *= sign;
+
+  /* fraction part */
+  if (*p == '.') {
+    double f = 0.0;
+    double base = 0.1;
+    ++p;
+
+    if (is_digit(*p)) {
+      while (*p && is_digit(*p)) {
+        f += base * (*p - '0');
+        base /= 10.0;
+        ++p;
+        ++n;
+      }
+    }
+    d += f * sign;
+    a = p;
+  }
+
+  /* exponential part */
+  if ((*p == 'E') || (*p == 'e')) {
+    int i, e = 0, neg = 0;
+    p++;
+    if (*p == '-') p++, neg++;
+    if (*p == '+') p++;
+    while (is_digit(*p)) e = e * 10 + *p++ - '0';
+    if (neg) e = -e;
+#if 0
+    if (d == 2.2250738585072011 && e == -308) {
+      d = 0.0;
+      a = p;
+      goto done;
+    }
+    if (d == 2.2250738585072012 && e <= -308) {
+      d *= 1.0e-308;
+      a = p;
+      goto done;
+    }
+#endif
+    for (i = 0; i < e; i++) d *= 10;
+    for (i = 0; i < -e; i++) d /= 10;
+    a = p;
+  } else if (p > str && !is_digit(*(p - 1))) {
+    a = str;
+    goto done;
+  }
+
+done:
+  if (end) *end = (char *) a;
+  return d;
+}
+
+#if MJSON_ENABLE_MERGE
+int mjson_merge(const char *s, int n, const char *s2, int n2,
+                mjson_print_fn_t fn, void *userdata) {
+  int koff, klen, voff, vlen, t, t2, k, off = 0, len = 0, comma = 0;
+  if (n < 2) return len;
+  len += fn("{", 1, userdata);
+  while ((off = mjson_next(s, n, off, &koff, &klen, &voff, &vlen, &t)) != 0) {
+    char *path = (char *) alloca(klen + 1);
+    const char *val;
+    memcpy(path, "$.", 2);
+    memcpy(path + 2, s + koff + 1, klen - 2);
+    path[klen] = '\0';
+    if ((t2 = mjson_find(s2, n2, path, &val, &k)) != MJSON_TOK_INVALID) {
+      if (t2 == MJSON_TOK_NULL) continue;  // null deletes the key
+    } else {
+      val = s + voff;  // Key is not found in the update. Copy the old value.
+    }
+    if (comma) len += fn(",", 1, userdata);
+    len += fn(s + koff, klen, userdata);
+    len += fn(":", 1, userdata);
+    if (t == MJSON_TOK_OBJECT && t2 == MJSON_TOK_OBJECT) {
+      len += mjson_merge(s + voff, vlen, val, k, fn, userdata);
+    } else {
+      if (t2 != MJSON_TOK_INVALID) vlen = k;
+      len += fn(val, vlen, userdata);
+    }
+    comma = 1;
+  }
+  // Add missing keys
+  off = 0;
+  while ((off = mjson_next(s2, n2, off, &koff, &klen, &voff, &vlen, &t)) != 0) {
+    char *path = (char *) alloca(klen + 1);
+    const char *val;
+    if (t == MJSON_TOK_NULL) continue;
+    memcpy(path, "$.", 2);
+    memcpy(path + 2, s2 + koff + 1, klen - 2);
+    path[klen] = '\0';
+    if (mjson_find(s, n, path, &val, &vlen) != MJSON_TOK_INVALID) continue;
+    if (comma) len += fn(",", 1, userdata);
+    len += fn(s2 + koff, klen, userdata);
+    len += fn(":", 1, userdata);
+    len += fn(s2 + voff, vlen, userdata);
+    comma = 1;
+  }
+  len += fn("}", 1, userdata);
+  return len;
+}
+#endif  // MJSON_ENABLE_MERGE
+
+#if MJSON_ENABLE_PRETTY
+struct prettydata {
+  int level;
+  int len;
+  int prev;
+  const char *pad;
+  int padlen;
+  mjson_print_fn_t fn;
+  void *userdata;
+};
+
+static int pretty_cb(int ev, const char *s, int off, int len, void *ud) {
+  struct prettydata *d = (struct prettydata *) ud;
+  int i;
+  switch (ev) {
+    case '{':
+    case '[':
+      d->level++;
+      d->len += d->fn(s + off, len, d->userdata);
+      break;
+    case '}':
+    case ']':
+      d->level--;
+      if (d->prev != '[' && d->prev != '{' && d->padlen > 0) {
+        d->len += d->fn("\n", 1, d->userdata);
+        for (i = 0; i < d->level; i++)
+          d->len += d->fn(d->pad, d->padlen, d->userdata);
+      }
+      d->len += d->fn(s + off, len, d->userdata);
+      break;
+    case ',':
+      d->len += d->fn(s + off, len, d->userdata);
+      if (d->padlen > 0) {
+        d->len += d->fn("\n", 1, d->userdata);
+        for (i = 0; i < d->level; i++)
+          d->len += d->fn(d->pad, d->padlen, d->userdata);
+      }
+      break;
+    case ':':
+      d->len += d->fn(s + off, len, d->userdata);
+      if (d->padlen > 0) d->len += d->fn(" ", 1, d->userdata);
+      break;
+    case MJSON_TOK_KEY:
+      if (d->prev == '{' && d->padlen > 0) {
+        d->len += d->fn("\n", 1, d->userdata);
+        for (i = 0; i < d->level; i++)
+          d->len += d->fn(d->pad, d->padlen, d->userdata);
+      }
+      d->len += d->fn(s + off, len, d->userdata);
+      break;
+    default:
+      if (d->prev == '[' && d->padlen > 0) {
+        d->len += d->fn("\n", 1, d->userdata);
+        for (i = 0; i < d->level; i++)
+          d->len += d->fn(d->pad, d->padlen, d->userdata);
+      }
+      d->len += d->fn(s + off, len, d->userdata);
+      break;
+  }
+  d->prev = ev;
+  return 0;
+}
+
+int mjson_pretty(const char *s, int n, const char *pad, mjson_print_fn_t fn,
+                 void *userdata) {
+  struct prettydata d = {0, 0, 0, pad, (int) strlen(pad), fn, userdata};
+  if (mjson(s, n, pretty_cb, &d) < 0) return -1;
+  return d.len;
+}
+#endif  // MJSON_ENABLE_PRETTY
+
+#if MJSON_ENABLE_RPC
+struct jsonrpc_ctx jsonrpc_default_context;
+
+int mjson_globmatch(const char *s1, int n1, const char *s2, int n2) {
+  int i = 0, j = 0, ni = 0, nj = 0;
+  while (i < n1 || j < n2) {
+    if (i < n1 && j < n2 && (s1[i] == '?' || s2[j] == s1[i])) {
+      i++, j++;
+    } else if (i < n1 && (s1[i] == '*' || s1[i] == '#')) {
+      ni = i, nj = j + 1, i++;
+    } else if (nj > 0 && nj <= n2 && (s1[i - 1] == '#' || s2[j] != '/')) {
+      i = ni, j = nj;
+    } else {
+      return 0;
+    }
+  }
+  return 1;
+}
+
+void jsonrpc_return_errorv(struct jsonrpc_request *r, int code,
+                           const char *message, const char *data_fmt,
+                           va_list ap) {
+  if (r->id_len == 0) return;
+  mjson_printf(r->fn, r->fndata,
+               "{\"id\":%.*s,\"error\":{\"code\":%d,\"message\":%Q", r->id_len,
+               r->id, code, message == NULL ? "" : message);
+  if (data_fmt != NULL) {
+    mjson_printf(r->fn, r->fndata, ",\"data\":");
+    mjson_vprintf(r->fn, r->fndata, data_fmt, ap);
+  }
+  mjson_printf(r->fn, r->fndata, "}}\n");
+}
+
+void jsonrpc_return_error(struct jsonrpc_request *r, int code,
+                          const char *message, const char *data_fmt, ...) {
+  va_list ap;
+  va_start(ap, data_fmt);
+  jsonrpc_return_errorv(r, code, message, data_fmt, ap);
+  va_end(ap);
+}
+
+void jsonrpc_return_successv(struct jsonrpc_request *r, const char *result_fmt,
+                             va_list ap) {
+  if (r->id_len == 0) return;
+  mjson_printf(r->fn, r->fndata, "{\"id\":%.*s,\"result\":", r->id_len, r->id);
+  if (result_fmt != NULL) {
+    mjson_vprintf(r->fn, r->fndata, result_fmt, ap);
+  } else {
+    mjson_printf(r->fn, r->fndata, "%s", "null");
+  }
+  mjson_printf(r->fn, r->fndata, "}\n");
+}
+
+void jsonrpc_return_success(struct jsonrpc_request *r, const char *result_fmt,
+                            ...) {
+  va_list ap;
+  va_start(ap, result_fmt);
+  jsonrpc_return_successv(r, result_fmt, ap);
+  va_end(ap);
+}
+
+void jsonrpc_ctx_process(struct jsonrpc_ctx *ctx, const char *buf, int len,
+                         mjson_print_fn_t fn, void *fndata, void *ud) {
+  const char *result = NULL, *error = NULL;
+  int result_sz = 0, error_sz = 0;
+  struct jsonrpc_method *m = NULL;
+  struct jsonrpc_request r = {ctx, buf, len, 0, 0, 0, 0, 0, 0, fn, fndata, ud};
+
+  // Is is a response frame?
+  mjson_find(buf, len, "$.result", &result, &result_sz);
+  if (result == NULL) mjson_find(buf, len, "$.error", &error, &error_sz);
+  if (result_sz > 0 || error_sz > 0) {
+    if (ctx->response_cb) ctx->response_cb(buf, len, ctx->response_cb_data);
+    return;
+  }
+
+  // Method must exist and must be a string
+  if (mjson_find(buf, len, "$.method", &r.method, &r.method_len) !=
+      MJSON_TOK_STRING) {
+    mjson_printf(fn, fndata, "{\"error\":{\"code\":-32700,\"message\":%.*Q}}\n",
+                 len, buf);
+    return;
+  }
+
+  // id and params are optional
+  mjson_find(buf, len, "$.id", &r.id, &r.id_len);
+  mjson_find(buf, len, "$.params", &r.params, &r.params_len);
+
+  for (m = ctx->methods; m != NULL; m = m->next) {
+    if (mjson_globmatch(m->method, m->method_sz, r.method + 1,
+                        r.method_len - 2) > 0) {
+      if (r.params == NULL) r.params = "";
+      m->cb(&r);
+      break;
+    }
+  }
+  if (m == NULL) {
+    jsonrpc_return_error(&r, JSONRPC_ERROR_NOT_FOUND, "method not found", NULL);
+  }
+}
+
+static int jsonrpc_print_methods(mjson_print_fn_t fn, void *fndata,
+                                 va_list *ap) {
+  struct jsonrpc_ctx *ctx = va_arg(*ap, struct jsonrpc_ctx *);
+  struct jsonrpc_method *m;
+  int len = 0;
+  for (m = ctx->methods; m != NULL; m = m->next) {
+    if (m != ctx->methods) len += mjson_print_buf(fn, fndata, ",", 1);
+    len += mjson_print_str(fn, fndata, m->method, (int) strlen(m->method));
+  }
+  return len;
+}
+
+static void rpclist(struct jsonrpc_request *r) {
+  jsonrpc_return_success(r, "[%M]", jsonrpc_print_methods, r->ctx);
+}
+
+void jsonrpc_ctx_init(struct jsonrpc_ctx *ctx, mjson_print_fn_t response_cb,
+                      void *response_cb_data) {
+  ctx->response_cb = response_cb;
+  ctx->response_cb_data = response_cb_data;
+  jsonrpc_ctx_export(ctx, MJSON_RPC_LIST_NAME, rpclist);
+}
+
+void jsonrpc_init(mjson_print_fn_t response_cb, void *userdata) {
+  jsonrpc_ctx_init(&jsonrpc_default_context, response_cb, userdata);
+}
+#endif  // MJSON_ENABLE_RPC