diff options
author | Hiltjo Posthuma <hiltjo@codemadness.org> | 2014-04-01 00:38:06 +0200 |
---|---|---|
committer | Hiltjo Posthuma <hiltjo@codemadness.org> | 2014-04-01 00:38:06 +0200 |
commit | 9ef8a4cb0ec0f8a101928097a6a2028c3448a0c2 (patch) | |
tree | 37b6604410785d077e1d6d1a1aaf97fce98c3fe2 | |
parent | 8e0b6ac2eb4295eb48fd740518bdb5b97ddfc228 (diff) |
cleanup some more, wrap some lines
Signed-off-by: Hiltjo Posthuma <hiltjo@codemadness.org>
-rw-r--r-- | sfeed.c | 36 |
1 files changed, 23 insertions, 13 deletions
@@ -11,12 +11,12 @@ #define ISWSNOSPACE(c) (((unsigned)c - '\t') < 5) /* isspace(c) && c != ' ' */ enum { FeedTypeNone = 0, FeedTypeRSS = 1, FeedTypeAtom = 2 }; -const char *feedtypes[] = { "", "rss", "atom" }; +static const char *feedtypes[] = { "", "rss", "atom" }; enum { ContentTypeNone = 0, ContentTypePlain = 1, ContentTypeHTML = 2 }; -const char *contenttypes[] = { "", "plain", "html" }; +static const char *contenttypes[] = { "", "plain", "html" }; -const int FieldSeparator = '\t'; /* output field seperator character */ +static const int FieldSeparator = '\t'; /* output field seperator character */ enum { TagUnknown = 0, @@ -54,7 +54,7 @@ typedef struct feedtag { } FeedTag; static void die(const char *s); -static void cleanup(void); +/*static void cleanup(void);*/ static String *currentfield = NULL; /* pointer to current FeedItem field String */ static FeedItem feeditem; /* data for current feed item */ @@ -411,7 +411,7 @@ parsetime(const char *s, char *buf) { buf[0] = '\0'; if(parsetimeformat(s, &tm, &end)) { offset = gettimetz(end, tz, sizeof(tz) - 1); - /* TODO: make sure snprintf cant overflow */ + /* TODO: use snprintf(): make sure *printf can't overflow */ if(buf) sprintf(buf, "%04d-%02d-%02d %02d:%02d:%02d %-.16s", tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, @@ -472,7 +472,8 @@ isattr(const char *name, size_t len, const char *name2, size_t len2) { static void xml_handler_data(XMLParser *p, const char *s, size_t len) { if(currentfield) { - if(feeditemtagid != AtomTagAuthor || !strcmp(p->tag, "name")) /* author>name */ + /* author>name */ + if(feeditemtagid != AtomTagAuthor || !strcmp(p->tag, "name")) string_append(currentfield, s, len); } } @@ -484,7 +485,8 @@ xml_handler_cdata(XMLParser *p, const char *s, size_t len) { } static void -xml_handler_attr_start(struct xmlparser *p, const char *tag, size_t taglen, const char *name, size_t namelen) { +xml_handler_attr_start(struct xmlparser *p, const char *tag, size_t taglen, + const char *name, size_t namelen) { if(iscontent && !iscontenttag) { if(!attrcount) xml_handler_data(p, " ", 1); @@ -496,7 +498,8 @@ xml_handler_attr_start(struct xmlparser *p, const char *tag, size_t taglen, cons } static void -xml_handler_attr_end(struct xmlparser *p, const char *tag, size_t taglen, const char *name, size_t namelen) { +xml_handler_attr_end(struct xmlparser *p, const char *tag, size_t taglen, + const char *name, size_t namelen) { if(iscontent && !iscontenttag) { xml_handler_data(p, "\"", 1); attrcount = 0; @@ -504,7 +507,8 @@ xml_handler_attr_end(struct xmlparser *p, const char *tag, size_t taglen, const } static void -xml_handler_start_element_parsed(XMLParser *p, const char *tag, size_t taglen, int isshort) { +xml_handler_start_element_parsed(XMLParser *p, const char *tag, size_t taglen, + int isshort) { if(iscontent && !iscontenttag) { if(isshort) xml_handler_data(p, "/>", 2); @@ -525,8 +529,11 @@ xml_handler_attr(XMLParser *p, const char *tag, size_t taglen, /*if(feeditemtagid == AtomTagContent || feeditemtagid == AtomTagSummary) {*/ if(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")))) { + (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; /* p->xmldataentity = NULL;*/ @@ -670,8 +677,11 @@ xml_handler_end_element(XMLParser *p, const char *name, size_t namelen, int issh if(feeditem.feedtype != FeedTypeNone) { /* end of RSS or Atom entry / item */ /* TODO: optimize, use gettag() ? to tagid? */ - if((feeditem.feedtype == FeedTypeAtom && istag(name, namelen, "entry", strlen("entry"))) || /* Atom */ - (feeditem.feedtype == FeedTypeRSS && istag(name, namelen, "item", strlen("item")))) { /* RSS */ + if((feeditem.feedtype == FeedTypeAtom && + istag(name, namelen, "entry", strlen("entry"))) || /* Atom */ + (feeditem.feedtype == FeedTypeRSS && + istag(name, namelen, "item", strlen("item")))) /* RSS */ + { printf("%ld", (long)parsetime((&feeditem.timestamp)->data, timebuf)); putchar(FieldSeparator); fputs(timebuf, stdout); |