/**
* Format data to stream.
*
* @return the amount of bytes written, -1 on error.
*/
ssize_t
ostream_printf(ostream_t *os, const char *fmt, ...)
{
va_list args, args2;
char buf[1024];
size_t len;
char *data;
ssize_t w;
ostream_check(os);
va_start(args, fmt);
VA_COPY(args2, args);
len = gm_vsnprintf(buf, sizeof buf, fmt, args2);
va_end(args2);
if (len >= sizeof buf - 1) {
data = h_strdup_len_vprintf(fmt, args, &len);
} else {
data = buf;
}
va_end(args);
w = ostream_write(os, data, len);
if (data != buf)
hfree(data);
return w;
}
/**
* Emit attributes.
*/
static void
xfmt_handle_pass2_attr(const char *uri,
const char *local, const char *value, void *data)
{
struct xfmt_pass2 *xp2 = data;
int c;
bool apos_escape = FALSE;
size_t len;
size_t overhead;
if (uri != NULL) {
ostream_printf(xp2->os, " %s:", xfmt_uri_to_prefix(xp2, uri));
} else {
ostream_putc(xp2->os, ' ');
}
/*
* Inspect value to select proper quoting.
*/
c = xfmt_quoting_char(value);
if ('\0' == c) {
apos_escape = TRUE;
c = '\''; /* We'll be quoting "'" so it's safe to use */
}
/*
* Now check for escaping of any '&', '<' or '>'.
*/
overhead = xfmt_text_escape_overhead(value, TRUE, apos_escape, &len);
ostream_printf(xp2->os, "%s=%c", local, c);
if (0 == overhead) {
ostream_write(xp2->os, value, len);
} else {
char *escaped = xfmt_text_escape(value, TRUE, apos_escape,
len + overhead);
ostream_write(xp2->os, escaped, len + overhead);
hfree(escaped);
}
ostream_putc(xp2->os, c);
}
/**
* Emit a single byte to the stream.
*
* @return the amount of bytes written, -1 on error.
*/
ssize_t
ostream_putc(ostream_t *os, int c)
{
char buf[1];
ostream_check(os);
buf[0] = c & 0xff;
return ostream_write(os, buf, sizeof buf);
}
inline EscapedMnemonic& operator<<(EscapedMnemonic& ostream, const T& t)
{
return ostream_write(ostream, t);
}
/**
* Pass 2 handler on each tree node entry.
*/
static bool
xfmt_handle_pass2_enter(const void *node, void *data)
{
const xnode_t *xn = node;
struct xfmt_pass2 *xp2 = data;
xp2->depth++;
if (xnode_is_element(xn)) {
GSList *ns = xfmt_ns_declarations(xp2, xn);
const char *nsuri = xnode_element_ns(xn);
if (!xp2->had_text && !xp2->last_was_nl) {
if (!(xp2->options & XFMT_O_SINGLE_LINE))
ostream_putc(xp2->os, '\n');
xp2->last_was_nl = TRUE;
}
xfmt_indent(xp2);
/*
* Look for the namespace matching the default namespace, in which
* case we don't have to emit it.
*/
if (
nsuri != NULL && xp2->default_ns != NULL &&
0 == strcmp(nsuri, xp2->default_ns)
) {
nsuri = NULL;
}
if (nsuri != NULL) {
const char *prefix = xfmt_uri_to_prefix(xp2, nsuri);
ostream_printf(xp2->os, "<%s:%s", prefix, xnode_element_name(xn));
} else {
ostream_printf(xp2->os, "<%s", xnode_element_name(xn));
}
/*
* Install default namespace on the root element, if any.
*/
if (1 == xp2->depth && xp2->default_ns != NULL) {
int c = xfmt_quoting_char(xp2->default_ns);
g_assert(c != '\0');
ostream_printf(xp2->os, " xmlns=%c%s%c", c, xp2->default_ns, c);
}
/*
* Declare namespaces for the element's scope.
*/
xfmt_pass2_declare_ns(xp2, ns);
g_slist_free(ns);
/*
* Emit attributes.
*/
xnode_prop_foreach(xn, xfmt_handle_pass2_attr, xp2);
/*
* Handle content-less elements specially: we don't let the
* "leave" callback run.
*
* We consider an element with a single empty text child as
* content-less, so we test with xnode_is_empty() instead of
* !xnode_has_content().
*/
xp2->had_text = FALSE;
if (xnode_is_empty(xn)) {
ostream_write(xp2->os, XFMT_EMPTY, CONST_STRLEN(XFMT_EMPTY));
if (!(xp2->options & XFMT_O_SINGLE_LINE))
ostream_putc(xp2->os, '\n');
xp2->last_was_nl = TRUE;
xfmt_pass2_leaving(xp2); /* No children, no "leave" callback */
return FALSE;
}
ostream_write(xp2->os, XFMT_GT, CONST_STRLEN(XFMT_GT));
xp2->last_was_nl = FALSE;
} else if (xnode_is_text(xn)) {
const char *text = xnode_text(xn);
size_t len;
size_t overhead;
bool amp;
if (xp2->options & XFMT_O_SKIP_BLANKS) {
const char *start;
size_t tlen;
start = xfmt_strip_blanks(text, &tlen);
if (0 == tlen)
goto ignore;
/* FIXME: handle blank collapsing */
(void) start;
}
/*
* If text is known to have entities, we must not escape the '&'.
* This means the generated XML must define that entity in the DTD
* part of the tree.
*
* Computes the required overhead to fully escape the text (0 meaning
* that no escaping is required). If the overhead is larger than
* a leading "<![CDATA[" and a closing ""]]>", we can emit a CDATA
* section instead, provided the text does not contain "]]>".
*/
amp = !xnode_text_has_entities(xn);
overhead = xfmt_text_escape_overhead(text, amp, FALSE, &len);
if (0 == overhead) {
ostream_write(xp2->os, text, len);
} else if (
overhead >= XFMT_CDATA_OVERHEAD &&
NULL == strstr(text, XFMT_CDATA_END)
) {
ostream_write(xp2->os,
XFMT_CDATA_START, CONST_STRLEN(XFMT_CDATA_START));
ostream_write(xp2->os, text, len);
ostream_write(xp2->os,
XFMT_CDATA_END, CONST_STRLEN(XFMT_CDATA_END));
} else {
char *escaped = xfmt_text_escape(text, amp, FALSE, len + overhead);
ostream_write(xp2->os, escaped, len + overhead);
hfree(escaped);
}
xp2->last_was_nl = FALSE;
xp2->had_text = TRUE;
}
ignore:
return TRUE;
}
/**
* Extended XML formatting of a tree.
*
* Namespaces, if any, are automatically assigned a prefix, whose format
* is "ns%u", the counter being incremented from 0.
*
* Users can supply a vector mapping namespaces to prefixes, so that they
* can force specific prefixes for a given well-known namespace.
*
* If there is a default namespace, all the tags belonging to that namespace
* are emitted without any prefix.
*
* The output stream must be explicitly closed by the user upon return.
*
* Options can be supplied to tune the output:
*
* - XFMT_O_SKIP_BLANKS will skip pure white space nodes.
* - XFMT_O_COLLAPSE_BLANKS will replace consecutive blanks with 1 space
* - XFMT_O_NO_INDENT requests that no indentation of the tree be made.
* - XFMT_O_PROLOGUE emits a leading <?xml?> prologue.
* - XFMT_O_FORCE_10 force generation of XML 1.0
* - XFMT_O_SINGLE_LINE emits XML as one big line (implies XFMT_O_NO_INDENT).
*
* @param root the root of the tree to dump
* @param os the output stream where tree is dumped
* @param options formatting options, as documented above
* @param pvec a vector of prefixes to be used for namespaces
* @param pvcnt amount of entries in vector
* @param default_ns default namespace to install at root element
*
* @return TRUE on success.
*/
bool
xfmt_tree_extended(const xnode_t *root, ostream_t *os, uint32 options,
const struct xfmt_prefix *pvec, size_t pvcnt, const char *default_ns)
{
struct xfmt_pass1 xp1;
struct xfmt_pass2 xp2;
struct xfmt_invert_ctx ictx;
const char *dflt_ns;
g_assert(root != NULL);
g_assert(os != NULL);
if (options & XFMT_O_COLLAPSE_BLANKS) {
/* FIXME */
g_carp("XFMT_O_COLLAPSE_BLANKS not supported yet");
stacktrace_where_print(stderr);
}
if (options & XFMT_O_SINGLE_LINE)
options |= XFMT_O_NO_INDENT;
/*
* First pass: look at namespaces and construct a table recording the
* earliest tree depth at which a namespace is used.
*/
ZERO(&xp1);
xp1.uri2node = htable_create(HASH_KEY_STRING, 0);
xp1.uri2prefix = nv_table_make(FALSE);
if (default_ns != NULL)
xp1.attr_uris = hset_create(HASH_KEY_STRING, 0);
htable_insert_const(xp1.uri2node, VXS_XML_URI, root);
xnode_tree_enter_leave(deconstify_pointer(root),
xfmt_handle_pass1_enter, xfmt_handle_pass1_leave, &xp1);
g_assert(0 == xp1.depth); /* Sound traversal */
/*
* If there was a default namespace, make sure it is used in the tree.
* Otherwise, discard it.
*/
if (default_ns != NULL) {
if (NULL == htable_lookup(xp1.uri2node, default_ns)) {
g_carp("XFMT default namespace '%s' is not needed", default_ns);
dflt_ns = NULL;
} else {
dflt_ns = default_ns;
}
} else {
dflt_ns = NULL;
}
/*
* Prepare context for second pass.
*/
ZERO(&xp2);
xp2.node2uri = htable_create(HASH_KEY_SELF, 0);
xp2.os = os;
xp2.options = options;
xp2.default_ns = dflt_ns;
xp2.attr_uris = xp1.attr_uris;
xp2.uri2prefix = xp1.uri2prefix;
xp2.uris = symtab_make();
xp2.prefixes = symtab_make();
xp2.depth = 0;
xp2.pcount = 0;
xp2.last_was_nl = TRUE;
/*
* Iterate over the hash table we've built to create a table indexed
* by tree node and listing the namespaces to declare for that node.
*/
ictx.uri2node = xp1.uri2node;
ictx.node2uri = xp2.node2uri;
htable_foreach(xp1.uri2node, xfmt_invert_uri_kv, &ictx);
htable_free_null(&xp1.uri2node);
/*
* Emit prologue if requested.
*/
if (options & XFMT_O_PROLOGUE) {
if (options & XFMT_O_FORCE_10) {
ostream_write(os, XFMT_DECL_10, CONST_STRLEN(XFMT_DECL_10));
} else {
ostream_write(os, XFMT_DECL, CONST_STRLEN(XFMT_DECL));
}
if (!(options & XFMT_O_SINGLE_LINE)) {
ostream_putc(os, '\n');
}
}
xfmt_prefix_declare(&xp2, VXS_XML_URI, VXS_XML);
/*
* Prepare user-defined URI -> prefix mappings.
*/
if (pvcnt != 0) {
size_t i;
for (i = 0; i < pvcnt; i++) {
const struct xfmt_prefix *p = &pvec[i];
xfmt_prefix_declare(&xp2, p->uri, p->prefix);
}
}
/*
* Second pass: generation.
*/
xnode_tree_enter_leave(deconstify_pointer(root),
xfmt_handle_pass2_enter, xfmt_handle_pass2_leave, &xp2);
g_assert(0 == xp2.depth); /* Sound traversal */
/*
* Done, cleanup.
*/
nv_table_free_null(&xp2.uri2prefix);
symtab_free_null(&xp2.prefixes);
symtab_free_null(&xp2.uris);
htable_free_null(&xp2.node2uri);
hset_free_null(&xp2.attr_uris);
return !ostream_has_ioerr(os);
}
