zrythm-serd/src/writer.c

/*
  Copyright 2011-2020 David Robillard <http://drobilla.net>

  Permission to use, copy, modify, and/or distribute this software for any
  purpose with or without fee is hereby granted, provided that the above
  copyright notice and this permission notice appear in all copies.

  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/

#include "byte_sink.h"
#include "serd_internal.h"
#include "stack.h"
#include "string_utils.h"
#include "uri_utils.h"

#include "serd/serd.h"

#include <assert.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

typedef enum {
  FIELD_NONE,
  FIELD_SUBJECT,
  FIELD_PREDICATE,
  FIELD_OBJECT,
  FIELD_GRAPH
} Field;

typedef struct {
  SerdNode graph;
  SerdNode subject;
  SerdNode predicate;
} WriteContext;

static const WriteContext WRITE_CONTEXT_NULL = {{0, 0, 0, 0, SERD_NOTHING},
                                                {0, 0, 0, 0, SERD_NOTHING},
                                                {0, 0, 0, 0, SERD_NOTHING}};

typedef enum {
  SEP_NONE,
  SEP_END_S,       ///< End of a subject ('.')
  SEP_END_P,       ///< End of a predicate (';')
  SEP_END_O,       ///< End of an object (',')
  SEP_S_P,         ///< Between a subject and predicate (whitespace)
  SEP_P_O,         ///< Between a predicate and object (whitespace)
  SEP_ANON_BEGIN,  ///< Start of anonymous node ('[')
  SEP_ANON_END,    ///< End of anonymous node (']')
  SEP_LIST_BEGIN,  ///< Start of list ('(')
  SEP_LIST_SEP,    ///< List separator (whitespace)
  SEP_LIST_END,    ///< End of list (')')
  SEP_GRAPH_BEGIN, ///< Start of graph ('{')
  SEP_GRAPH_END,   ///< End of graph ('}')
  SEP_URI_BEGIN,   ///< URI start quote ('<')
  SEP_URI_END      ///< URI end quote ('>')
} Sep;

typedef struct {
  const char* str;              ///< Sep string
  uint8_t     len;              ///< Length of sep string
  uint8_t     space_before;     ///< Newline before sep
  uint8_t     space_after_node; ///< Newline after sep if after node
  uint8_t     space_after_sep;  ///< Newline after sep if after sep
} SepRule;

static const SepRule rules[] = {{NULL, 0, 0, 0, 0},
                                {" .\n\n", 4, 0, 0, 0},
                                {" ;", 2, 0, 1, 1},
                                {" ,", 2, 0, 1, 0},
                                {NULL, 0, 0, 1, 0},
                                {" ", 1, 0, 0, 0},
                                {"[", 1, 0, 1, 1},
                                {"]", 1, 1, 0, 0},
                                {"(", 1, 0, 0, 0},
                                {NULL, 0, 0, 1, 0},
                                {")", 1, 1, 0, 0},
                                {" {", 2, 0, 1, 1},
                                {" }", 2, 0, 1, 1},
                                {"<", 1, 0, 0, 0},
                                {">", 1, 0, 0, 0},
                                {"\n", 1, 0, 1, 0}};

struct SerdWriterImpl {
  SerdSyntax    syntax;
  SerdStyle     style;
  SerdEnv*      env;
  SerdNode      root_node;
  SerdURI       root_uri;
  SerdURI       base_uri;
  SerdStack     anon_stack;
  SerdByteSink  byte_sink;
  SerdErrorSink error_sink;
  void*         error_handle;
  WriteContext  context;
  SerdNode      list_subj;
  unsigned      list_depth;
  unsigned      indent;
  uint8_t*      bprefix;
  size_t        bprefix_len;
  Sep           last_sep;
  bool          empty;
};

typedef enum { WRITE_STRING, WRITE_LONG_STRING } TextContext;

static bool
write_node(SerdWriter*        writer,
           const SerdNode*    node,
           const SerdNode*    datatype,
           const SerdNode*    lang,
           Field              field,
           SerdStatementFlags flags);

static bool
supports_abbrev(const SerdWriter* writer)
{
  return writer->syntax == SERD_TURTLE || writer->syntax == SERD_TRIG;
}

static bool
supports_uriref(const SerdWriter* writer)
{
  return writer->syntax == SERD_TURTLE || writer->syntax == SERD_TRIG;
}

static void
w_err(SerdWriter* writer, SerdStatus st, const char* fmt, ...)
{
  /* TODO: This results in errors with no file information, which is not
     helpful when re-serializing a file (particularly for "undefined
     namespace prefix" errors.  The statement sink API needs to be changed to
     add a Cursor parameter so the source can notify the writer of the
     statement origin for better error reporting. */

  va_list args;
  va_start(args, fmt);
  const SerdError e = {st, (const uint8_t*)"", 0, 0, fmt, &args};
  serd_error(writer->error_sink, writer->error_handle, &e);
  va_end(args);
}

static inline WriteContext*
anon_stack_top(SerdWriter* writer)
{
  assert(!serd_stack_is_empty(&writer->anon_stack));
  return (WriteContext*)(writer->anon_stack.buf + writer->anon_stack.size -
                         sizeof(WriteContext));
}

static void
copy_node(SerdNode* dst, const SerdNode* src)
{
  if (src) {
    dst->buf     = (uint8_t*)realloc((char*)dst->buf, src->n_bytes + 1);
    dst->n_bytes = src->n_bytes;
    dst->n_chars = src->n_chars;
    dst->flags   = src->flags;
    dst->type    = src->type;
    memcpy((char*)dst->buf, src->buf, src->n_bytes + 1);
  } else {
    dst->type = SERD_NOTHING;
  }
}

static inline size_t
sink(const void* buf, size_t len, SerdWriter* writer)
{
  return serd_byte_sink_write(buf, len, &writer->byte_sink);
}

// Write a single character, as an escape for single byte characters
// (Caller prints any single byte characters that don't need escaping)
static size_t
write_character(SerdWriter* writer, const uint8_t* utf8, size_t* size)
{
  char           escape[11] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  const uint32_t c          = parse_utf8_char(utf8, size);
  switch (*size) {
  case 0:
    w_err(writer, SERD_ERR_BAD_ARG, "invalid UTF-8: %X\n", utf8[0]);
    return sink(replacement_char, sizeof(replacement_char), writer);
  case 1:
    snprintf(escape, sizeof(escape), "\\u%04X", utf8[0]);
    return sink(escape, 6, writer);
  default:
    break;
  }

  if (!(writer->style & SERD_STYLE_ASCII)) {
    // Write UTF-8 character directly to UTF-8 output
    return sink(utf8, *size, writer);
  }

  if (c <= 0xFFFF) {
    snprintf(escape, sizeof(escape), "\\u%04X", c);
    return sink(escape, 6, writer);
  }

  snprintf(escape, sizeof(escape), "\\U%08X", c);
  return sink(escape, 10, writer);
}

static inline bool
uri_must_escape(const uint8_t c)
{
  switch (c) {
  case ' ':
  case '"':
  case '<':
  case '>':
  case '\\':
  case '^':
  case '`':
  case '{':
  case '|':
  case '}':
    return true;
  default:
    return !in_range(c, 0x20, 0x7E);
  }
}

static size_t
write_uri(SerdWriter* writer, const uint8_t* utf8, size_t n_bytes)
{
  size_t len = 0;
  for (size_t i = 0; i < n_bytes;) {
    size_t j = i; // Index of next character that must be escaped
    for (; j < n_bytes; ++j) {
      if (uri_must_escape(utf8[j])) {
        break;
      }
    }

    // Bulk write all characters up to this special one
    len += sink(&utf8[i], j - i, writer);
    if ((i = j) == n_bytes) {
      break; // Reached end
    }

    // Write UTF-8 character
    size_t size = 0;
    len += write_character(writer, utf8 + i, &size);
    i += size;
    if (size == 0) {
      // Corrupt input, scan to start of next character
      for (++i; i < n_bytes && (utf8[i] & 0x80); ++i) {
      }
    }
  }

  return len;
}

static bool
lname_must_escape(const uint8_t c)
{
  /* This arbitrary list of characters, most of which have nothing to do with
     Turtle, must be handled as special cases here because the RDF and SPARQL
     WGs are apparently intent on making the once elegant Turtle a baroque
     and inconsistent mess, throwing elegance and extensibility completely
     out the window for no good reason.

     Note '-', '.', and '_' are also in PN_LOCAL_ESC, but are valid unescaped
     in local names, so they are not escaped here. */

  switch (c) {
  case '\'':
  case '!':
  case '#':
  case '$':
  case '%':
  case '&':
  case '(':
  case ')':
  case '*':
  case '+':
  case ',':
  case '/':
  case ';':
  case '=':
  case '?':
  case '@':
  case '~':
    return true;
  default:
    break;
  }
  return false;
}

static size_t
write_lname(SerdWriter* writer, const uint8_t* utf8, size_t n_bytes)
{
  size_t len = 0;
  for (size_t i = 0; i < n_bytes; ++i) {
    size_t j = i; // Index of next character that must be escaped
    for (; j < n_bytes; ++j) {
      if (lname_must_escape(utf8[j])) {
        break;
      }
    }

    // Bulk write all characters up to this special one
    len += sink(&utf8[i], j - i, writer);
    if ((i = j) == n_bytes) {
      break; // Reached end
    }

    // Write escape
    len += sink("\\", 1, writer);
    len += sink(&utf8[i], 1, writer);
  }

  return len;
}

static size_t
write_text(SerdWriter*    writer,
           TextContext    ctx,
           const uint8_t* utf8,
           size_t         n_bytes)
{
  size_t len = 0;
  for (size_t i = 0; i < n_bytes;) {
    // Fast bulk write for long strings of printable ASCII
    size_t j = i;
    for (; j < n_bytes; ++j) {
      if (utf8[j] == '\\' || utf8[j] == '"' ||
          (!in_range(utf8[j], 0x20, 0x7E))) {
        break;
      }
    }

    len += sink(&utf8[i], j - i, writer);
    if ((i = j) == n_bytes) {
      break; // Reached end
    }

    const uint8_t in = utf8[i++];
    if (ctx == WRITE_LONG_STRING) {
      switch (in) {
      case '\\':
        len += sink("\\\\", 2, writer);
        continue;
      case '\b':
        len += sink("\\b", 2, writer);
        continue;
      case '\n':
      case '\r':
      case '\t':
      case '\f':
        len += sink(&in, 1, writer); // Write character as-is
        continue;
      case '\"':
        if (i == n_bytes) { // '"' at string end
          len += sink("\\\"", 2, writer);
        } else {
          len += sink(&in, 1, writer);
        }
        continue;
      default:
        break;
      }
    } else if (ctx == WRITE_STRING) {
      switch (in) {
      case '\\':
        len += sink("\\\\", 2, writer);
        continue;
      case '\n':
        len += sink("\\n", 2, writer);
        continue;
      case '\r':
        len += sink("\\r", 2, writer);
        continue;
      case '\t':
        len += sink("\\t", 2, writer);
        continue;
      case '"':
        len += sink("\\\"", 2, writer);
        continue;
      default:
        break;
      }
      if (writer->syntax == SERD_TURTLE) {
        switch (in) {
        case '\b':
          len += sink("\\b", 2, writer);
          continue;
        case '\f':
          len += sink("\\f", 2, writer);
          continue;
        default:
          break;
        }
      }
    }

    // Write UTF-8 character
    size_t size = 0;
    len += write_character(writer, utf8 + i - 1, &size);
    if (size == 0) {
      // Corrupt input, scan to start of next character
      for (; i < n_bytes && (utf8[i] & 0x80); ++i) {
      }
    } else {
      i += size - 1;
    }
  }

  return len;
}

static size_t
uri_sink(const void* buf, size_t len, void* stream)
{
  return write_uri((SerdWriter*)stream, (const uint8_t*)buf, len);
}

static void
write_newline(SerdWriter* writer)
{
  sink("\n", 1, writer);
  for (unsigned i = 0; i < writer->indent; ++i) {
    sink("\t", 1, writer);
  }
}

static bool
write_sep(SerdWriter* writer, const Sep sep)
{
  const SepRule* rule = &rules[sep];
  if (rule->space_before) {
    write_newline(writer);
  }

  if (rule->str) {
    sink(rule->str, rule->len, writer);
  }

  if ((writer->last_sep && rule->space_after_sep) ||
      (!writer->last_sep && rule->space_after_node)) {
    write_newline(writer);
  } else if (writer->last_sep && rule->space_after_node) {
    sink(" ", 1, writer);
  }

  writer->last_sep = sep;
  return true;
}

static SerdStatus
reset_context(SerdWriter* writer, bool graph)
{
  if (graph) {
    writer->context.graph.type = SERD_NOTHING;
  }

  writer->context.subject.type   = SERD_NOTHING;
  writer->context.predicate.type = SERD_NOTHING;
  writer->empty                  = false;
  return SERD_SUCCESS;
}

static SerdStatus
free_context(SerdWriter* writer)
{
  serd_node_free(&writer->context.graph);
  serd_node_free(&writer->context.subject);
  serd_node_free(&writer->context.predicate);
  return reset_context(writer, true);
}

static bool
is_inline_start(const SerdWriter* writer, Field field, SerdStatementFlags flags)
{
  return (supports_abbrev(writer) &&
          ((field == FIELD_SUBJECT && (flags & SERD_ANON_S_BEGIN)) ||
           (field == FIELD_OBJECT && (flags & SERD_ANON_O_BEGIN))));
}

static bool
write_literal(SerdWriter*        writer,
              const SerdNode*    node,
              const SerdNode*    datatype,
              const SerdNode*    lang,
              SerdStatementFlags flags)
{
  if (supports_abbrev(writer) && datatype && datatype->buf) {
    const char* type_uri = (const char*)datatype->buf;
    if (!strncmp(type_uri, NS_XSD, sizeof(NS_XSD) - 1) &&
        (!strcmp(type_uri + sizeof(NS_XSD) - 1, "boolean") ||
         !strcmp(type_uri + sizeof(NS_XSD) - 1, "integer"))) {
      sink(node->buf, node->n_bytes, writer);
      return true;
    }

    if (!strncmp(type_uri, NS_XSD, sizeof(NS_XSD) - 1) &&
        !strcmp(type_uri + sizeof(NS_XSD) - 1, "decimal") &&
        strchr((const char*)node->buf, '.') &&
        node->buf[node->n_bytes - 1] != '.') {
      /* xsd:decimal literals without trailing digits, e.g. "5.", can
         not be written bare in Turtle.  We could add a 0 which is
         prettier, but changes the text and breaks round tripping.
      */
      sink(node->buf, node->n_bytes, writer);
      return true;
    }
  }

  if (supports_abbrev(writer) &&
      (node->flags & (SERD_HAS_NEWLINE | SERD_HAS_QUOTE))) {
    sink("\"\"\"", 3, writer);
    write_text(writer, WRITE_LONG_STRING, node->buf, node->n_bytes);
    sink("\"\"\"", 3, writer);
  } else {
    sink("\"", 1, writer);
    write_text(writer, WRITE_STRING, node->buf, node->n_bytes);
    sink("\"", 1, writer);
  }
  if (lang && lang->buf) {
    sink("@", 1, writer);
    sink(lang->buf, lang->n_bytes, writer);
  } else if (datatype && datatype->buf) {
    sink("^^", 2, writer);
    return write_node(writer, datatype, NULL, NULL, FIELD_NONE, flags);
  }
  return true;
}

// Return true iff `buf` is a valid prefixed name suffix
static inline bool
is_name(const uint8_t* buf, const size_t len)
{
  // TODO: This is more strict than it should be
  for (size_t i = 0; i < len; ++i) {
    if (!(is_alpha(buf[i]) || is_digit(buf[i]))) {
      return false;
    }
  }

  return true;
}

static bool
write_uri_node(SerdWriter* const        writer,
               const SerdNode*          node,
               const Field              field,
               const SerdStatementFlags flags)
{
  SerdNode  prefix;
  SerdChunk suffix;

  if (is_inline_start(writer, field, flags)) {
    ++writer->indent;
    write_sep(writer, SEP_ANON_BEGIN);
    sink("== ", 3, writer);
  }

  const bool has_scheme = serd_uri_string_has_scheme(node->buf);
  if (supports_abbrev(writer)) {
    if (field == FIELD_PREDICATE &&
        !strcmp((const char*)node->buf, NS_RDF "type")) {
      return sink("a", 1, writer) == 1;
    }

    if (!strcmp((const char*)node->buf, NS_RDF "nil")) {
      return sink("()", 2, writer) == 2;
    }

    if (has_scheme && (writer->style & SERD_STYLE_CURIED) &&
        serd_env_qualify(writer->env, node, &prefix, &suffix) &&
        is_name(suffix.buf, suffix.len)) {
      write_uri(writer, prefix.buf, prefix.n_bytes);
      sink(":", 1, writer);
      write_uri(writer, suffix.buf, suffix.len);
      return true;
    }
  }

  if (!has_scheme && !supports_uriref(writer) &&
      !serd_env_get_base_uri(writer->env, NULL)->buf) {
    w_err(writer,
          SERD_ERR_BAD_ARG,
          "syntax does not support URI reference <%s>\n",
          node->buf);
    return false;
  }

  write_sep(writer, SEP_URI_BEGIN);
  if (writer->style & SERD_STYLE_RESOLVED) {
    SerdURI in_base_uri;
    SerdURI uri;
    SerdURI abs_uri;
    serd_env_get_base_uri(writer->env, &in_base_uri);
    serd_uri_parse(node->buf, &uri);
    serd_uri_resolve(&uri, &in_base_uri, &abs_uri);
    bool     rooted = uri_is_under(&writer->base_uri, &writer->root_uri);
    SerdURI* root   = rooted ? &writer->root_uri : &writer->base_uri;
    if (!uri_is_under(&abs_uri, root) || writer->syntax == SERD_NTRIPLES ||
        writer->syntax == SERD_NQUADS) {
      serd_uri_serialise(&abs_uri, uri_sink, writer);
    } else {
      serd_uri_serialise_relative(
        &uri, &writer->base_uri, root, uri_sink, writer);
    }
  } else {
    write_uri(writer, node->buf, node->n_bytes);
  }

  write_sep(writer, SEP_URI_END);
  if (is_inline_start(writer, field, flags)) {
    sink(" ;", 2, writer);
    write_newline(writer);
  }

  return true;
}

static bool
write_curie(SerdWriter* const        writer,
            const SerdNode*          node,
            const Field              field,
            const SerdStatementFlags flags)
{
  SerdChunk  prefix = {NULL, 0};
  SerdChunk  suffix = {NULL, 0};
  SerdStatus st     = SERD_SUCCESS;

  switch (writer->syntax) {
  case SERD_NTRIPLES:
  case SERD_NQUADS:
    if ((st = serd_env_expand(writer->env, node, &prefix, &suffix))) {
      w_err(writer, st, "undefined namespace prefix `%s'\n", node->buf);
      return false;
    }
    write_sep(writer, SEP_URI_BEGIN);
    write_uri(writer, prefix.buf, prefix.len);
    write_uri(writer, suffix.buf, suffix.len);
    write_sep(writer, SEP_URI_END);
    break;
  case SERD_TURTLE:
  case SERD_TRIG:
    if (is_inline_start(writer, field, flags)) {
      ++writer->indent;
      write_sep(writer, SEP_ANON_BEGIN);
      sink("== ", 3, writer);
    }
    write_lname(writer, node->buf, node->n_bytes);
    if (is_inline_start(writer, field, flags)) {
      sink(" ;", 2, writer);
      write_newline(writer);
    }
  }

  return true;
}

static bool
write_blank(SerdWriter* const        writer,
            const SerdNode*          node,
            const Field              field,
            const SerdStatementFlags flags)
{
  if (supports_abbrev(writer)) {
    if (is_inline_start(writer, field, flags)) {
      ++writer->indent;
      return write_sep(writer, SEP_ANON_BEGIN);
    }

    if (field == FIELD_SUBJECT && (flags & SERD_LIST_S_BEGIN)) {
      assert(writer->list_depth == 0);
      copy_node(&writer->list_subj, node);
      ++writer->list_depth;
      ++writer->indent;
      return write_sep(writer, SEP_LIST_BEGIN);
    }

    if (field == FIELD_OBJECT && (flags & SERD_LIST_O_BEGIN)) {
      ++writer->indent;
      ++writer->list_depth;
      return write_sep(writer, SEP_LIST_BEGIN);
    }

    if ((field == FIELD_SUBJECT && (flags & SERD_EMPTY_S)) ||
        (field == FIELD_OBJECT && (flags & SERD_EMPTY_O))) {
      return sink("[]", 2, writer) == 2;
    }
  }

  sink("_:", 2, writer);
  if (writer->bprefix && !strncmp((const char*)node->buf,
                                  (const char*)writer->bprefix,
                                  writer->bprefix_len)) {
    sink(node->buf + writer->bprefix_len,
         node->n_bytes - writer->bprefix_len,
         writer);
  } else {
    sink(node->buf, node->n_bytes, writer);
  }

  return true;
}

static bool
write_node(SerdWriter*        writer,
           const SerdNode*    node,
           const SerdNode*    datatype,
           const SerdNode*    lang,
           Field              field,
           SerdStatementFlags flags)
{
  bool ret = false;
  switch (node->type) {
  case SERD_NOTHING:
    break;
  case SERD_LITERAL:
    ret = write_literal(writer, node, datatype, lang, flags);
    break;
  case SERD_URI:
    ret = write_uri_node(writer, node, field, flags);
    break;
  case SERD_CURIE:
    ret = write_curie(writer, node, field, flags);
    break;
  case SERD_BLANK:
    ret = write_blank(writer, node, field, flags);
    break;
  }

  writer->last_sep = SEP_NONE;
  return ret;
}

static inline bool
is_resource(const SerdNode* node)
{
  return node && node->buf && node->type > SERD_LITERAL;
}

static void
write_pred(SerdWriter* writer, SerdStatementFlags flags, const SerdNode* pred)
{
  write_node(writer, pred, NULL, NULL, FIELD_PREDICATE, flags);
  write_sep(writer, SEP_P_O);
  copy_node(&writer->context.predicate, pred);
}

static bool
write_list_obj(SerdWriter*        writer,
               SerdStatementFlags flags,
               const SerdNode*    predicate,
               const SerdNode*    object,
               const SerdNode*    datatype,
               const SerdNode*    lang)
{
  if (!strcmp((const char*)object->buf, NS_RDF "nil")) {
    --writer->indent;
    write_sep(writer, SEP_LIST_END);
    return true;
  }

  if (!strcmp((const char*)predicate->buf, NS_RDF "first")) {
    write_sep(writer, SEP_LIST_SEP);
    write_node(writer, object, datatype, lang, FIELD_OBJECT, flags);
  }

  return false;
}

SerdStatus
serd_writer_write_statement(SerdWriter*        writer,
                            SerdStatementFlags flags,
                            const SerdNode*    graph,
                            const SerdNode*    subject,
                            const SerdNode*    predicate,
                            const SerdNode*    object,
                            const SerdNode*    datatype,
                            const SerdNode*    lang)
{
  if (!is_resource(subject) || !is_resource(predicate) || !object ||
      !object->buf) {
    return SERD_ERR_BAD_ARG;
  }

#define TRY(write_result)      \
  do {                         \
    if (!(write_result)) {     \
      return SERD_ERR_UNKNOWN; \
    }                          \
  } while (0)

  if (writer->syntax == SERD_NTRIPLES || writer->syntax == SERD_NQUADS) {
    TRY(write_node(writer, subject, NULL, NULL, FIELD_SUBJECT, flags));
    sink(" ", 1, writer);
    TRY(write_node(writer, predicate, NULL, NULL, FIELD_PREDICATE, flags));
    sink(" ", 1, writer);
    TRY(write_node(writer, object, datatype, lang, FIELD_OBJECT, flags));
    if (writer->syntax == SERD_NQUADS && graph) {
      sink(" ", 1, writer);
      TRY(write_node(writer, graph, datatype, lang, FIELD_GRAPH, flags));
    }
    sink(" .\n", 3, writer);
    return SERD_SUCCESS;
  }

  if ((graph && !serd_node_equals(graph, &writer->context.graph)) ||
      (!graph && writer->context.graph.type)) {
    writer->indent = 0;

    if (writer->context.subject.type) {
      write_sep(writer, SEP_END_S);
    }

    if (writer->context.graph.type) {
      write_sep(writer, SEP_GRAPH_END);
    }

    reset_context(writer, true);
    if (graph) {
      TRY(write_node(writer, graph, datatype, lang, FIELD_GRAPH, flags));
      ++writer->indent;
      write_sep(writer, SEP_GRAPH_BEGIN);
      copy_node(&writer->context.graph, graph);
    }
  }

  if ((flags & SERD_LIST_CONT)) {
    if (write_list_obj(writer, flags, predicate, object, datatype, lang)) {
      // Reached end of list
      if (--writer->list_depth == 0 && writer->list_subj.type) {
        reset_context(writer, false);
        serd_node_free(&writer->context.subject);
        writer->context.subject = writer->list_subj;
        writer->list_subj       = SERD_NODE_NULL;
      }
      return SERD_SUCCESS;
    }
  } else if (serd_node_equals(subject, &writer->context.subject)) {
    if (serd_node_equals(predicate, &writer->context.predicate)) {
      // Abbreviate S P
      if (!(flags & SERD_ANON_O_BEGIN)) {
        ++writer->indent;
      }
      write_sep(writer, SEP_END_O);
      write_node(writer, object, datatype, lang, FIELD_OBJECT, flags);
      if (!(flags & SERD_ANON_O_BEGIN)) {
        --writer->indent;
      }
    } else {
      // Abbreviate S
      Sep sep = writer->context.predicate.type ? SEP_END_P : SEP_S_P;
      write_sep(writer, sep);
      write_pred(writer, flags, predicate);
      write_node(writer, object, datatype, lang, FIELD_OBJECT, flags);
    }
  } else {
    // No abbreviation
    if (writer->context.subject.type) {
      assert(writer->indent > 0);
      --writer->indent;
      if (serd_stack_is_empty(&writer->anon_stack)) {
        write_sep(writer, SEP_END_S);
      }
    } else if (!writer->empty) {
      write_sep(writer, SEP_S_P);
    }

    if (!(flags & SERD_ANON_CONT)) {
      write_node(writer, subject, NULL, NULL, FIELD_SUBJECT, flags);
      ++writer->indent;
      write_sep(writer, SEP_S_P);
    } else {
      ++writer->indent;
    }

    reset_context(writer, false);
    copy_node(&writer->context.subject, subject);

    if (!(flags & SERD_LIST_S_BEGIN)) {
      write_pred(writer, flags, predicate);
    }

    write_node(writer, object, datatype, lang, FIELD_OBJECT, flags);
  }

  if (flags & (SERD_ANON_S_BEGIN | SERD_ANON_O_BEGIN)) {
    WriteContext* ctx =
      (WriteContext*)serd_stack_push(&writer->anon_stack, sizeof(WriteContext));
    *ctx                     = writer->context;
    WriteContext new_context = {
      serd_node_copy(graph), serd_node_copy(subject), SERD_NODE_NULL};
    if ((flags & SERD_ANON_S_BEGIN)) {
      new_context.predicate = serd_node_copy(predicate);
    }
    writer->context = new_context;
  } else {
    copy_node(&writer->context.graph, graph);
    copy_node(&writer->context.subject, subject);
    copy_node(&writer->context.predicate, predicate);
  }

  return SERD_SUCCESS;
}

SerdStatus
serd_writer_end_anon(SerdWriter* writer, const SerdNode* node)
{
  if (writer->syntax == SERD_NTRIPLES || writer->syntax == SERD_NQUADS) {
    return SERD_SUCCESS;
  }

  if (serd_stack_is_empty(&writer->anon_stack) || writer->indent == 0) {
    w_err(writer, SERD_ERR_UNKNOWN, "unexpected end of anonymous node\n");
    return SERD_ERR_UNKNOWN;
  }

  --writer->indent;
  write_sep(writer, SEP_ANON_END);
  free_context(writer);
  writer->context = *anon_stack_top(writer);
  serd_stack_pop(&writer->anon_stack, sizeof(WriteContext));
  const bool is_subject = serd_node_equals(node, &writer->context.subject);
  if (is_subject) {
    copy_node(&writer->context.subject, node);
    writer->context.predicate.type = SERD_NOTHING;
  }

  return SERD_SUCCESS;
}

SerdStatus
serd_writer_finish(SerdWriter* writer)
{
  if (writer->context.subject.type) {
    write_sep(writer, SEP_END_S);
  }

  if (writer->context.graph.type) {
    write_sep(writer, SEP_GRAPH_END);
  }

  serd_byte_sink_flush(&writer->byte_sink);
  writer->indent = 0;
  return free_context(writer);
}

SerdWriter*
serd_writer_new(SerdSyntax     syntax,
                SerdStyle      style,
                SerdEnv*       env,
                const SerdURI* base_uri,
                SerdSink       ssink,
                void*          stream)
{
  const WriteContext context = WRITE_CONTEXT_NULL;
  SerdWriter*        writer  = (SerdWriter*)calloc(1, sizeof(SerdWriter));

  writer->syntax     = syntax;
  writer->style      = style;
  writer->env        = env;
  writer->root_node  = SERD_NODE_NULL;
  writer->root_uri   = SERD_URI_NULL;
  writer->base_uri   = base_uri ? *base_uri : SERD_URI_NULL;
  writer->anon_stack = serd_stack_new(4 * sizeof(WriteContext));
  writer->context    = context;
  writer->list_subj  = SERD_NODE_NULL;
  writer->empty      = true;
  writer->byte_sink  = serd_byte_sink_new(
    ssink, stream, (style & SERD_STYLE_BULK) ? SERD_PAGE_SIZE : 1);

  return writer;
}

void
serd_writer_set_error_sink(SerdWriter*   writer,
                           SerdErrorSink error_sink,
                           void*         error_handle)
{
  writer->error_sink   = error_sink;
  writer->error_handle = error_handle;
}

void
serd_writer_chop_blank_prefix(SerdWriter* writer, const uint8_t* prefix)
{
  free(writer->bprefix);
  writer->bprefix_len = 0;
  writer->bprefix     = NULL;

  const size_t prefix_len = prefix ? strlen((const char*)prefix) : 0;
  if (prefix_len) {
    writer->bprefix_len = prefix_len;
    writer->bprefix     = (uint8_t*)malloc(writer->bprefix_len + 1);
    memcpy(writer->bprefix, prefix, writer->bprefix_len + 1);
  }
}

SerdStatus
serd_writer_set_base_uri(SerdWriter* writer, const SerdNode* uri)
{
  if (!serd_env_set_base_uri(writer->env, uri)) {
    serd_env_get_base_uri(writer->env, &writer->base_uri);

    if (writer->syntax == SERD_TURTLE || writer->syntax == SERD_TRIG) {
      if (writer->context.graph.type || writer->context.subject.type) {
        sink(" .\n\n", 4, writer);
        reset_context(writer, true);
      }
      sink("@base <", 7, writer);
      sink(uri->buf, uri->n_bytes, writer);
      sink("> .\n", 4, writer);
    }
    writer->indent = 0;
    return reset_context(writer, true);
  }

  return SERD_ERR_UNKNOWN;
}

SerdStatus
serd_writer_set_root_uri(SerdWriter* writer, const SerdNode* uri)
{
  serd_node_free(&writer->root_node);

  if (uri && uri->buf) {
    writer->root_node = serd_node_copy(uri);
    serd_uri_parse(uri->buf, &writer->root_uri);
  } else {
    writer->root_node = SERD_NODE_NULL;
    writer->root_uri  = SERD_URI_NULL;
  }

  return SERD_SUCCESS;
}

SerdStatus
serd_writer_set_prefix(SerdWriter*     writer,
                       const SerdNode* name,
                       const SerdNode* uri)
{
  if (!serd_env_set_prefix(writer->env, name, uri)) {
    if (writer->syntax == SERD_TURTLE || writer->syntax == SERD_TRIG) {
      if (writer->context.graph.type || writer->context.subject.type) {
        sink(" .\n\n", 4, writer);
        reset_context(writer, true);
      }
      sink("@prefix ", 8, writer);
      sink(name->buf, name->n_bytes, writer);
      sink(": <", 3, writer);
      write_uri(writer, uri->buf, uri->n_bytes);
      sink("> .\n", 4, writer);
    }
    writer->indent = 0;
    return reset_context(writer, true);
  }

  return SERD_ERR_UNKNOWN;
}

void
serd_writer_free(SerdWriter* writer)
{
  if (!writer) {
    return;
  }

  serd_writer_finish(writer);
  serd_stack_free(&writer->anon_stack);
  free(writer->bprefix);
  serd_byte_sink_free(&writer->byte_sink);
  serd_node_free(&writer->root_node);
  free(writer);
}

SerdEnv*
serd_writer_get_env(SerdWriter* writer)
{
  return writer->env;
}

size_t
serd_file_sink(const void* buf, size_t len, void* stream)
{
  return fwrite(buf, 1, len, (FILE*)stream);
}

size_t
serd_chunk_sink(const void* buf, size_t len, void* stream)
{
  SerdChunk* chunk = (SerdChunk*)stream;
  chunk->buf       = (uint8_t*)realloc((uint8_t*)chunk->buf, chunk->len + len);
  memcpy((uint8_t*)chunk->buf + chunk->len, buf, len);
  chunk->len += len;
  return len;
}

uint8_t*
serd_chunk_sink_finish(SerdChunk* stream)
{
  serd_chunk_sink("", 1, stream);
  return (uint8_t*)stream->buf;
}