summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--config.mk9
-rw-r--r--sfeed.c423
-rw-r--r--sfeed_frames.c8
3 files changed, 194 insertions, 246 deletions
diff --git a/config.mk b/config.mk
index 87c766f..ed0a879 100644
--- a/config.mk
+++ b/config.mk
@@ -12,12 +12,13 @@ INCS =
LIBS = -lc
# debug
-#CFLAGS = -fstack-protector-all -O0 -g -std=c99 -Wall -Wextra -pedantic -DVERSION=\"${VERSION}\" -D_POSIX_C_SOURCE=200809L
-#LDFLAGS = ${LIBS}
+CFLAGS = -fstack-protector-all -O0 -g -std=c99 -Wall -Wextra -pedantic \
+ -DVERSION=\"${VERSION}\" -D_POSIX_C_SOURCE=200809L -D_XOPEN_SOURCE
+LDFLAGS = ${LIBS}
# optimized
-CFLAGS = -O2 -std=c99 -DVERSION=\"${VERSION}\" -D_POSIX_C_SOURCE=200809L
-LDFLAGS = -s ${LIBS}
+#CFLAGS = -O2 -std=c99 -DVERSION=\"${VERSION}\" -D_POSIX_C_SOURCE=200809L -D_XOPEN_SOURCE
+#LDFLAGS = -s ${LIBS}
# Solaris
#CFLAGS = -fast ${INCS} -DVERSION=\"${VERSION}\"
diff --git a/sfeed.c b/sfeed.c
index bebd254..f739170 100644
--- a/sfeed.c
+++ b/sfeed.c
@@ -9,6 +9,7 @@
#include "xml.h"
#define ISWSNOSPACE(c) (((unsigned)c - '\t') < 5) /* isspace(c) && c != ' ' */
+#define LEN(x) (sizeof (x) / sizeof *(x))
enum { FeedTypeNone = 0, FeedTypeRSS = 1, FeedTypeAtom = 2 };
static const char *feedtypes[] = { "", "rss", "atom" };
@@ -55,17 +56,31 @@ typedef struct feedtag {
int id;
} FeedTag;
+typedef struct feedcontext {
+ char tag[256];
+ int tagid;
+ size_t taglen;
+ String *field;
+ FeedItem item;
+ int iscontent;
+ int iscontenttag;
+ int attrcount;
+} FeedContext;
+
static void die(const char *s);
static void cleanup(void);
-static String *currentfield = NULL; /* pointer to current FeedItem field String */
+#if 0
+static String *ctx.field = NULL; /* pointer to current FeedItem field String */
static FeedItem feeditem; /* data for current feed item */
-static char feeditemtag[256] = ""; /* current tag _inside_ a feeditem */
-static size_t feeditemtaglen = 0;
-static int feeditemtagid = 0; /* unique number for parsed tag (faster comparison) */
-static int iscontent = 0;
-static int iscontenttag = 0;
-static size_t attrcount = 0;
+static char ctx.tag[256] = ""; /* current tag _inside_ a feeditem */
+static size_t ctx.taglen = 0;
+static int ctx.tagid = 0; /* unique number for parsed tag (faster comparison) */
+static int ctx.iscontent = 0;
+static int ctx.iscontenttag = 0;
+static size_t ctx.attrcount = 0;
+#endif
+static FeedContext ctx;
static XMLParser parser; /* XML parser state */
static char *append = NULL;
@@ -255,12 +270,12 @@ string_append(String *s, const char *data, size_t len) {
/* cleanup, free allocated memory, etc */
static void
cleanup(void) {
- string_free(&feeditem.timestamp);
- string_free(&feeditem.title);
- string_free(&feeditem.link);
- string_free(&feeditem.content);
- string_free(&feeditem.id);
- string_free(&feeditem.author);
+ string_free(&ctx.item.timestamp);
+ string_free(&ctx.item.title);
+ string_free(&ctx.item.link);
+ string_free(&ctx.item.content);
+ string_free(&ctx.item.id);
+ string_free(&ctx.item.author);
}
static void /* print error message to stderr */
@@ -282,11 +297,11 @@ gettimetz(const char *s, char *buf, size_t bufsiz) {
char c;
buf[0] = '\0';
- if(bufsiz < sizeof(tzname) + 7)
+ if(bufsiz < sizeof(tzname) + strlen(" -00:00"))
return 0;
for(; *p && isspace((int)*p); p++); /* skip whitespace */
/* loop until some common timezone delimiters are found */
- for(;*p && (*p != '+' && *p != '-' && *p != 'Z' && *p != 'z'); p++);
+ for(; *p && (*p != '+' && *p != '-' && *p != 'Z' && *p != 'z'); p++);
/* TODO: cleanup / simplify */
if(isalpha((int)*p)) {
@@ -302,7 +317,7 @@ gettimetz(const char *s, char *buf, size_t bufsiz) {
} else
memcpy(tzname, "GMT", strlen("GMT") + 1);
if(!(*p)) {
- strlcpy(buf, tzname, bufsiz); /* TODO: dont depend on strlcpy? */
+ strlcpy(buf, tzname, bufsiz);
return 0;
}
if((sscanf(p, "%c%02d:%02d", &c, &tzhour, &tzmin)) > 0);
@@ -314,112 +329,37 @@ gettimetz(const char *s, char *buf, size_t bufsiz) {
return (tzhour * 3600) + (tzmin * 60) * (c == '-' ? -1 : 1);
}
-/* parses everything in a format similar to:
- * "%a, %d %b %Y %H:%M:%S" or "%Y-%m-%d %H:%M:%S" */
-/* TODO: calculate time offset (GMT only) from gettimetz ? */
-static int
-parsetimeformat(const char *s, struct tm *t, const char **end) {
- const char *months[] = {
- "Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct",
- "Nov", "Dec"
- };
- unsigned int i, fm;
- unsigned long l;
-
- memset(t, 0, sizeof(struct tm));
- if((l = strtoul(s, (void *)&s, 10))) {
- t->tm_year = abs(l) - 1900;
- if(!(l = strtoul(s, (void *)&s, 10)))
- return 0;
- t->tm_mon = abs(l) - 1;
- if(!(t->tm_mday = abs(strtoul(s, (void *)&s, 10))))
- return 0;
- } else {
- for(; *s && !isdigit((int)*s); s++);
- if(!(t->tm_mday = abs(strtoul(s, (void *)&s, 10))))
- return 0;
- for(; *s && !isalpha((int)*s); s++); /* skip non-alpha */
- for(fm = 0, i = 0; i < 12; i++) { /* parse month names */
- if(!strncasecmp(s, months[i], 3)) {
- t->tm_mon = i;
- fm = 1;
- break;
- }
- }
- if(!fm) /* can't find month */
- return 0;
- for(; *s && !isdigit((int)*s); s++); /* skip non-digit */
- if(!(l = strtoul(s, (void *)&s, 10)))
- return 0;
- t->tm_year = abs(l) - 1900;
- }
- for(; *s && !isdigit((int)*s); s++); /* skip non-digit */
- if((t->tm_hour = abs(strtoul(s, (void *)&s, 10))) > 23)
- return 0;
- for(; *s && !isdigit((int)*s); s++); /* skip non-digit */
- if((t->tm_min = abs(strtoul(s, (void *)&s, 10))) > 59)
- return 0;
- for(; *s && !isdigit((int)*s); s++); /* skip non-digit */
- if((t->tm_sec = abs(strtoul(s, (void *)&s, 10))) > 60)
- return 0;
- if(end)
- *end = s;
- return 1;
-}
-
-/* C defines the rounding for division in a nonsensical way */
-#define Q(a,b) ((a)>0 ? (a)/(b) : -(((b)-(a)-1)/(b)))
-
-/* copied from Musl C awesome small implementation, see LICENSE. */
-static time_t
-tmtotime(struct tm *tm) {
- time_t year = tm->tm_year - 100;
- int month = tm->tm_mon;
- int day = tm->tm_mday;
- int daysbeforemon[] = { 0,31,59,90,120,151,181,212,243,273,304,334 };
- int z4, z100, z400;
-
- /* normalize month */
- if(month >= 12) {
- year += month / 12;
- month %= 12;
- } else if(month < 0) {
- year += month / 12;
- month %= 12;
- if(month) {
- month += 12;
- year--;
- }
- }
- z4 = Q(year - (month < 2), 4); /* is leap? */
- z100 = Q(z4, 25);
- z400 = Q(z100, 4);
- day += year * 365 + z4 - z100 + z400 + daysbeforemon[month];
- return (time_t)day * 86400 +
- tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec +
- 946684800; /* the dawn of time, aka 1970 (30 years of seconds) :) */
-}
-
static time_t
parsetime(const char *s, char *buf, size_t bufsiz) {
+ time_t t = -1; /* can't parse */
+ char tz[64] = "";
struct tm tm;
- char tz[64];
- const char *end;
- int offset;
+ char *formats[] = {
+ "%a, %d %b %Y %H:%M:%S",
+ "%Y-%m-%d %H:%M:%S",
+ "%Y-%m-%dT%H:%M:%S",
+ NULL
+ };
+ char *p;
+ int offset, i;
if(buf)
buf[0] = '\0';
- if(parsetimeformat(s, &tm, &end)) {
- offset = gettimetz(end, tz, sizeof(tz) - 1);
- /* TODO: use snprintf(): make sure *printf can't overflow */
- if(buf)
- snprintf(buf, bufsiz, "%04d-%02d-%02d %02d:%02d:%02d %-.16s",
- tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday,
- tm.tm_hour, tm.tm_min, tm.tm_sec, tz);
- /* return UNIX time, reverse offset to GMT+0 */
- return tmtotime(&tm) - offset;
+ memset(&tm, 0, sizeof(tm));
+ for(i = 0; i < LEN(formats); i++) {
+ if((p = strptime(s, formats[i], &tm))) {
+ tm.tm_isdst = -1; /* don't use DST */
+ t = mktime(&tm);
+ offset = gettimetz(p, tz, sizeof(tz));
+ if(buf)
+ snprintf(buf, bufsiz, "%04d-%02d-%02d %02d:%02d:%02d %-.16s",
+ tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday,
+ tm.tm_hour, tm.tm_min, tm.tm_sec, tz);
+ t -= offset;
+ break;
+ }
}
- return -1; /* can't parse */
+ return t;
}
static void /* print text, escape tabs, newline and carriage return etc */
@@ -471,26 +411,26 @@ isattr(const char *name, size_t len, const char *name2, size_t len2) {
* block is bigger than the buffer */
static void
xml_handler_data(XMLParser *p, const char *s, size_t len) {
- if(currentfield) {
+ if(ctx.field) {
/* author>name */
- if(feeditemtagid != AtomTagAuthor || !strcmp(p->tag, "name"))
- string_append(currentfield, s, len);
+ if(ctx.tagid != AtomTagAuthor || !strcmp(p->tag, "name"))
+ string_append(ctx.field, s, len);
}
}
static void
xml_handler_cdata(XMLParser *p, const char *s, size_t len) {
- if(currentfield)
- string_append(currentfield, s, len);
+ if(ctx.field)
+ string_append(ctx.field, s, len);
}
static void
xml_handler_attr_start(struct xmlparser *p, const char *tag, size_t taglen,
const char *name, size_t namelen) {
- if(iscontent && !iscontenttag) {
- if(!attrcount)
+ if(ctx.iscontent && !ctx.iscontenttag) {
+ if(!ctx.attrcount)
xml_handler_data(p, " ", 1);
- attrcount++;
+ ctx.attrcount++;
xml_handler_data(p, name, namelen);
xml_handler_data(p, "=\"", 2);
return;
@@ -500,16 +440,16 @@ xml_handler_attr_start(struct xmlparser *p, const char *tag, size_t taglen,
static void
xml_handler_attr_end(struct xmlparser *p, const char *tag, size_t taglen,
const char *name, size_t namelen) {
- if(iscontent && !iscontenttag) {
+ if(ctx.iscontent && !ctx.iscontenttag) {
xml_handler_data(p, "\"", 1);
- attrcount = 0;
+ ctx.attrcount = 0;
}
}
static void
xml_handler_start_element_parsed(XMLParser *p, const char *tag, size_t taglen,
int isshort) {
- if(iscontent && !iscontenttag) {
+ if(ctx.iscontent && !ctx.iscontenttag) {
if(isshort)
xml_handler_data(p, "/>", 2);
else
@@ -521,112 +461,117 @@ static void
xml_handler_attr(XMLParser *p, const char *tag, size_t taglen,
const char *name, size_t namelen, const char *value,
size_t valuelen) {
- if(iscontent && !iscontenttag) {
+ if(ctx.iscontent && !ctx.iscontenttag) {
xml_handler_data(p, value, valuelen);
return;
}
- if(feeditem.feedtype == FeedTypeAtom) {
- /*if(feeditemtagid == AtomTagContent || feeditemtagid == AtomTagSummary) {*/
- if(iscontenttag) {
+ if(ctx.item.feedtype == FeedTypeAtom) {
+ /*if(ctx.tagid == AtomTagContent || ctx.tagid == AtomTagSummary) {*/
+ if(ctx.iscontenttag) {
if(isattr(name, namelen, "type", strlen("type")) &&
(isattr(value, valuelen, "xhtml", strlen("xhtml")) ||
isattr(value, valuelen, "text/xhtml", strlen("text/xhtml")) ||
isattr(value, valuelen, "html", strlen("html")) ||
isattr(value, valuelen, "text/html", strlen("text/html"))))
{
- feeditem.contenttype = ContentTypeHTML;
- iscontent = 1;
+ ctx.item.contenttype = ContentTypeHTML;
+ ctx.iscontent = 1;
/* p->xmldataentity = NULL;*/
p->xmlattrstart = xml_handler_attr_start;
p->xmlattrend = xml_handler_attr_end;
p->xmltagstartparsed = xml_handler_start_element_parsed;
}
- } else if(feeditemtagid == AtomTagLink && isattr(name, namelen, "href", strlen("href"))) /* link href attribute */
- string_append(&feeditem.link, value, valuelen);
+ } else if(ctx.tagid == AtomTagLink &&
+ isattr(name, namelen, "href", strlen("href")))
+ {
+ /* link href attribute */
+ string_append(&ctx.item.link, value, valuelen);
+ }
}
}
static void
xml_handler_start_element(XMLParser *p, const char *name, size_t namelen) {
- if(iscontenttag) {
+ if(ctx.iscontenttag) {
/* starts with div, handle as XML, dont convert entities */
/* TODO: test properly and do printf() to debug */
- if(feeditem.feedtype == FeedTypeAtom && !strncmp(name, "div", strlen("div")))
+ if(ctx.item.feedtype == FeedTypeAtom && !strncmp(name, "div", strlen("div")))
p->xmldataentity = NULL;
}
- if(iscontent) {
- attrcount = 0;
- iscontenttag = 0;
+ if(ctx.iscontent) {
+ ctx.attrcount = 0;
+ ctx.iscontenttag = 0;
xml_handler_data(p, "<", 1);
xml_handler_data(p, name, namelen);
return;
}
/* TODO: cleanup, merge with code below ?, return function if FeedTypeNone */
-/* iscontenttag = 0;*/
- if(feeditem.feedtype != FeedTypeNone) { /* in item */
- if(feeditemtag[0] == '\0') { /* set tag if not already set. */
- if(namelen >= sizeof(feeditemtag) - 2) /* check overflow */
+/* ctx.iscontenttag = 0;*/
+ if(ctx.item.feedtype != FeedTypeNone) { /* in item */
+ if(ctx.tag[0] == '\0') { /* set tag if not already set. */
+ if(namelen >= sizeof(ctx.tag) - 2) /* check overflow */
return;
- memcpy(feeditemtag, name, namelen + 1); /* copy including nul byte */
- feeditemtaglen = namelen;
- feeditemtagid = gettag(feeditem.feedtype, feeditemtag, feeditemtaglen);
- if(feeditem.feedtype == FeedTypeRSS) {
- if(feeditemtagid == TagUnknown)
- currentfield = NULL;
- else if(feeditemtagid == RSSTagPubdate || feeditemtagid == RSSTagDcdate)
- currentfield = &feeditem.timestamp;
- else if(feeditemtagid == RSSTagTitle)
- currentfield = &feeditem.title;
- else if(feeditemtagid == RSSTagLink)
- currentfield = &feeditem.link;
- else if(feeditemtagid == RSSTagDescription || feeditemtagid == RSSTagContentencoded) {
+ memcpy(ctx.tag, name, namelen + 1); /* copy including nul byte */
+ ctx.taglen = namelen;
+ ctx.tagid = gettag(ctx.item.feedtype, ctx.tag, ctx.taglen);
+ if(ctx.item.feedtype == FeedTypeRSS) {
+ if(ctx.tagid == TagUnknown)
+ ctx.field = NULL;
+ else if(ctx.tagid == RSSTagPubdate || ctx.tagid == RSSTagDcdate)
+ ctx.field = &ctx.item.timestamp;
+ else if(ctx.tagid == RSSTagTitle)
+ ctx.field = &ctx.item.title;
+ else if(ctx.tagid == RSSTagLink)
+ ctx.field = &ctx.item.link;
+ else if(ctx.tagid == RSSTagDescription ||
+ ctx.tagid == RSSTagContentencoded) {
/* clear previous summary, assumes previous content was not a summary text */
- if(feeditemtagid == RSSTagContentencoded && feeditem.content.len)
- string_clear(&feeditem.content);
+ if(ctx.tagid == RSSTagContentencoded && ctx.item.content.len)
+ string_clear(&ctx.item.content);
/* ignore, prefer content:encoded over description */
- if(!(feeditemtagid == RSSTagDescription && feeditem.content.len)) {
- iscontenttag = 1;
- currentfield = &feeditem.content;
+ if(!(ctx.tagid == RSSTagDescription && ctx.item.content.len)) {
+ ctx.iscontenttag = 1;
+ ctx.field = &ctx.item.content;
}
- } else if(feeditemtagid == RSSTagGuid)
- currentfield = &feeditem.id;
- else if(feeditemtagid == RSSTagAuthor || feeditemtagid == RSSTagDccreator)
- currentfield = &feeditem.author;
- } else if(feeditem.feedtype == FeedTypeAtom) {
- if(feeditemtagid == TagUnknown)
- currentfield = NULL;
- else if(feeditemtagid == AtomTagPublished || feeditemtagid == AtomTagUpdated)
- currentfield = &feeditem.timestamp;
- else if(feeditemtagid == AtomTagTitle)
- currentfield = &feeditem.title;
- else if(feeditemtagid == AtomTagSummary || feeditemtagid == AtomTagContent) {
+ } else if(ctx.tagid == RSSTagGuid)
+ ctx.field = &ctx.item.id;
+ else if(ctx.tagid == RSSTagAuthor || ctx.tagid == RSSTagDccreator)
+ ctx.field = &ctx.item.author;
+ } else if(ctx.item.feedtype == FeedTypeAtom) {
+ if(ctx.tagid == TagUnknown)
+ ctx.field = NULL;
+ else if(ctx.tagid == AtomTagPublished || ctx.tagid == AtomTagUpdated)
+ ctx.field = &ctx.item.timestamp;
+ else if(ctx.tagid == AtomTagTitle)
+ ctx.field = &ctx.item.title;
+ else if(ctx.tagid == AtomTagSummary || ctx.tagid == AtomTagContent) {
/* clear previous summary, assumes previous content was not a summary text */
- if(feeditemtagid == AtomTagContent && feeditem.content.len)
- string_clear(&feeditem.content);
+ if(ctx.tagid == AtomTagContent && ctx.item.content.len)
+ string_clear(&ctx.item.content);
/* ignore, prefer content:encoded over description */
- if(!(feeditemtagid == AtomTagSummary && feeditem.content.len)) {
- iscontenttag = 1;
- currentfield = &feeditem.content;
+ if(!(ctx.tagid == AtomTagSummary && ctx.item.content.len)) {
+ ctx.iscontenttag = 1;
+ ctx.field = &ctx.item.content;
}
- } else if(feeditemtagid == AtomTagId)
- currentfield = &feeditem.id;
- else if(feeditemtagid == AtomTagLink)
- currentfield = &feeditem.link;
- else if(feeditemtagid == AtomTagAuthor)
- currentfield = &feeditem.author;
+ } else if(ctx.tagid == AtomTagId)
+ ctx.field = &ctx.item.id;
+ else if(ctx.tagid == AtomTagLink)
+ ctx.field = &ctx.item.link;
+ else if(ctx.tagid == AtomTagAuthor)
+ ctx.field = &ctx.item.author;
}
/* TODO: prefer content encoded over content? test */
}
} else { /* start of RSS or Atom item / entry */
if(istag(name, namelen, "entry", strlen("entry"))) { /* Atom */
- feeditem.feedtype = FeedTypeAtom;
- feeditem.contenttype = ContentTypePlain; /* Default content type */
- currentfield = NULL; /* XXX: optimization */
+ ctx.item.feedtype = FeedTypeAtom;
+ ctx.item.contenttype = ContentTypePlain; /* Default content type */
+ ctx.field = NULL; /* XXX: optimization */
} else if(istag(name, namelen, "item", strlen("item"))) { /* RSS */
- feeditem.feedtype = FeedTypeRSS;
- feeditem.contenttype = ContentTypeHTML; /* Default content type */
- currentfield = NULL; /* XXX: optimization */
+ ctx.item.feedtype = FeedTypeRSS;
+ ctx.item.contenttype = ContentTypeHTML; /* Default content type */
+ ctx.field = NULL; /* XXX: optimization */
}
}
}
@@ -648,22 +593,22 @@ xml_handler_end_element(XMLParser *p, const char *name, size_t namelen, int issh
char timebuf[64];
int tagid;
- if(iscontent) {
- attrcount = 0;
+ if(ctx.iscontent) {
+ ctx.attrcount = 0;
/* TODO: optimize */
- tagid = gettag(feeditem.feedtype, name, namelen);
- if(feeditemtagid == tagid) { /* close content */
- iscontent = 0;
- iscontenttag = 0;
+ tagid = gettag(ctx.item.feedtype, name, namelen);
+ if(ctx.tagid == tagid) { /* close content */
+ ctx.iscontent = 0;
+ ctx.iscontenttag = 0;
p->xmldataentity = xml_handler_data_entity;
p->xmlattrstart = NULL;
p->xmlattrend = NULL;
p->xmltagstartparsed = NULL;
- feeditemtag[0] = '\0'; /* unset tag */
- feeditemtaglen = 0;
- feeditemtagid = TagUnknown;
+ ctx.tag[0] = '\0'; /* unset tag */
+ ctx.taglen = 0;
+ ctx.tagid = TagUnknown;
return; /* TODO: not sure if !isshort check below should be skipped */
}
@@ -674,32 +619,32 @@ xml_handler_end_element(XMLParser *p, const char *name, size_t namelen, int issh
}
return;
}
- if(feeditem.feedtype != FeedTypeNone) {
+ if(ctx.item.feedtype != FeedTypeNone) {
/* end of RSS or Atom entry / item */
/* TODO: optimize, use gettag() ? to tagid? */
- if((feeditem.feedtype == FeedTypeAtom &&
+ if((ctx.item.feedtype == FeedTypeAtom &&
istag(name, namelen, "entry", strlen("entry"))) || /* Atom */
- (feeditem.feedtype == FeedTypeRSS &&
+ (ctx.item.feedtype == FeedTypeRSS &&
istag(name, namelen, "item", strlen("item")))) /* RSS */
{
- printf("%ld", (long)parsetime((&feeditem.timestamp)->data,
+ printf("%ld", (long)parsetime((&ctx.item.timestamp)->data,
timebuf, sizeof(timebuf)));
putchar(FieldSeparator);
fputs(timebuf, stdout);
putchar(FieldSeparator);
- string_print(&feeditem.title);
+ string_print(&ctx.item.title);
putchar(FieldSeparator);
- string_print(&feeditem.link);
+ string_print(&ctx.item.link);
putchar(FieldSeparator);
- string_print(&feeditem.content);
+ string_print(&ctx.item.content);
putchar(FieldSeparator);
- fputs(contenttypes[feeditem.contenttype], stdout);
+ fputs(contenttypes[ctx.item.contenttype], stdout);
putchar(FieldSeparator);
- string_print(&feeditem.id);
+ string_print(&ctx.item.id);
putchar(FieldSeparator);
- string_print(&feeditem.author);
+ string_print(&ctx.item.author);
putchar(FieldSeparator);
- fputs(feedtypes[feeditem.feedtype], stdout);
+ fputs(feedtypes[ctx.item.feedtype], stdout);
if(append) {
putchar(FieldSeparator);
fputs(append, stdout);
@@ -707,30 +652,30 @@ xml_handler_end_element(XMLParser *p, const char *name, size_t namelen, int issh
putchar('\n');
/* clear strings */
- string_clear(&feeditem.timestamp);
- string_clear(&feeditem.title);
- string_clear(&feeditem.link);
- string_clear(&feeditem.content);
- string_clear(&feeditem.id);
- string_clear(&feeditem.author);
- feeditem.feedtype = FeedTypeNone;
- feeditem.contenttype = ContentTypePlain;
- feeditemtag[0] = '\0'; /* unset tag */
- feeditemtaglen = 0;
- feeditemtagid = TagUnknown;
+ string_clear(&ctx.item.timestamp);
+ string_clear(&ctx.item.title);
+ string_clear(&ctx.item.link);
+ string_clear(&ctx.item.content);
+ string_clear(&ctx.item.id);
+ string_clear(&ctx.item.author);
+ ctx.item.feedtype = FeedTypeNone;
+ ctx.item.contenttype = ContentTypePlain;
+ ctx.tag[0] = '\0'; /* unset tag */
+ ctx.taglen = 0;
+ ctx.tagid = TagUnknown;
/* not sure if needed */
- iscontenttag = 0;
- iscontent = 0;
- } else if(!strcmp(feeditemtag, name)) { /* clear */ /* XXX: optimize ? */
- currentfield = NULL;
- feeditemtag[0] = '\0'; /* unset tag */
- feeditemtaglen = 0;
- feeditemtagid = TagUnknown;
+ ctx.iscontenttag = 0;
+ ctx.iscontent = 0;
+ } else if(!strcmp(ctx.tag, name)) { /* clear */ /* XXX: optimize ? */
+ ctx.field = NULL;
+ ctx.tag[0] = '\0'; /* unset tag */
+ ctx.taglen = 0;
+ ctx.tagid = TagUnknown;
/* not sure if needed */
- iscontenttag = 0;
- iscontent = 0;
+ ctx.iscontenttag = 0;
+ ctx.iscontent = 0;
}
}
}
@@ -742,15 +687,17 @@ main(int argc, char **argv) {
if(argc > 1)
append = argv[1];
+ memset(&ctx, 0, sizeof(ctx));
+
/* init strings and initial memory pool size */
- string_buffer_init(&feeditem.timestamp, 64);
- string_buffer_init(&feeditem.title, 256);
- string_buffer_init(&feeditem.link, 1024);
- string_buffer_init(&feeditem.content, 4096);
- string_buffer_init(&feeditem.id, 1024);
- string_buffer_init(&feeditem.author, 256);
- feeditem.contenttype = ContentTypePlain;
- feeditem.feedtype = FeedTypeNone;
+ string_buffer_init(&ctx.item.timestamp, 64);
+ string_buffer_init(&ctx.item.title, 256);
+ string_buffer_init(&ctx.item.link, 1024);
+ string_buffer_init(&ctx.item.content, 4096);
+ string_buffer_init(&ctx.item.id, 1024);
+ string_buffer_init(&ctx.item.author, 256);
+ ctx.item.contenttype = ContentTypePlain;
+ ctx.item.feedtype = FeedTypeNone;
xmlparser_init(&parser, stdin);
parser.xmltagstart = xml_handler_start_element;
diff --git a/sfeed_frames.c b/sfeed_frames.c
index 64c5048..13e59d0 100644
--- a/sfeed_frames.c
+++ b/sfeed_frames.c
@@ -287,10 +287,10 @@ main(int argc, char **argv) {
} else {
fputs("<frameset framespacing=\"0\" cols=\"*\" frameborder=\"1\">\n", fpindex);
}
- fputs(" <frameset id=\"frameset\" framespacing=\"0\" cols=\"50%,50%\" frameborder=\"1\">\n"
- " <frame name=\"items\" src=\"items.html\" target=\"items\">\n"
- " <frame name=\"content\" target=\"content\">\n"
- " </frameset>\n"
+ fputs("\t<frameset id=\"frameset\" framespacing=\"0\" cols=\"50%,50%\" frameborder=\"1\">\n"
+ "\t\t<frame name=\"items\" src=\"items.html\" target=\"items\">\n"
+ "\t\t<frame name=\"content\" target=\"content\">\n"
+ "\t</frameset>\n"
"</frameset>\n"
"</html>", fpindex);