From 96cebe806a12af1160ff086a3d7eb35bfd4ad261 Mon Sep 17 00:00:00 2001 From: Alexis La Goutte Date: Wed, 24 Oct 2012 15:52:08 +0000 Subject: [PATCH] Fix indent and add Modelines info for new common ui source file(s) svn path=/trunk/; revision=45763 --- ui/text_import.c | 209 +++++++++++++++++++++++++---------------------- ui/text_import.h | 13 +++ 2 files changed, 124 insertions(+), 98 deletions(-) diff --git a/ui/text_import.c b/ui/text_import.c index c8a122f4ba..45d48bae8c 100644 --- a/ui/text_import.c +++ b/ui/text_import.c @@ -181,7 +181,7 @@ static unsigned long packet_start = 0; static void start_new_packet (void); /* This buffer contains strings present before the packet offset 0 */ -#define PACKET_PREAMBLE_MAX_LEN 2048 +#define PACKET_PREAMBLE_MAX_LEN 2048 static unsigned char packet_preamble[PACKET_PREAMBLE_MAX_LEN+1]; static int packet_preamble_len = 0; @@ -524,12 +524,12 @@ write_current_packet (void) if (ts_fmt == NULL) { ts_usec++; } /* fake packet counter */ pkthdr.caplen = pkthdr.len = prefix_length + curr_offset + eth_trailer_length; pkthdr.pkt_encap = pcap_link_type; - pkthdr.interface_id = 0; - pkthdr.presence_flags = 0; - pkthdr.opt_comment = NULL; - pkthdr.drop_count = 0; - pkthdr.pack_flags = 0; - pkthdr.presence_flags = WTAP_HAS_CAP_LEN|WTAP_HAS_INTERFACE_ID|WTAP_HAS_TS; + pkthdr.interface_id = 0; + pkthdr.presence_flags = 0; + pkthdr.opt_comment = NULL; + pkthdr.drop_count = 0; + pkthdr.pack_flags = 0; + pkthdr.presence_flags = WTAP_HAS_CAP_LEN|WTAP_HAS_INTERFACE_ID|WTAP_HAS_TS; wtap_dump(wdh, &pkthdr, packet_buf, &err); } @@ -550,14 +550,14 @@ append_to_preamble(char *str) if (packet_preamble_len != 0) { if (packet_preamble_len == PACKET_PREAMBLE_MAX_LEN) - return; /* no room to add more preamble */ + return; /* no room to add more preamble */ /* Add a blank separator between the previous token and this token. */ packet_preamble[packet_preamble_len++] = ' '; } toklen = strlen(str); if (toklen != 0) { if (packet_preamble_len + toklen > PACKET_PREAMBLE_MAX_LEN) - return; /* no room to add the token to the preamble */ + return; /* no room to add the token to the preamble */ g_strlcpy(&packet_preamble[packet_preamble_len], str, PACKET_PREAMBLE_MAX_LEN); packet_preamble_len += (int) toklen; if (debug >= 2) { @@ -577,103 +577,103 @@ append_to_preamble(char *str) static void parse_preamble (void) { - struct tm timecode; - char *subsecs; - char *p; - int subseclen; - int i; + struct tm timecode; + char *subsecs; + char *p; + int subseclen; + int i; - /* - * If no "-t" flag was specified, don't attempt to parse a packet - * preamble to extract a time stamp. - */ - if (ts_fmt == NULL) - return; + /* + * If no "-t" flag was specified, don't attempt to parse a packet + * preamble to extract a time stamp. + */ + if (ts_fmt == NULL) + return; - /* - * Initialize to today localtime, just in case not all fields - * of the date and time are specified. - */ + /* + * Initialize to today localtime, just in case not all fields + * of the date and time are specified. + */ - timecode = timecode_default; - ts_usec = 0; + timecode = timecode_default; + ts_usec = 0; - /* - * Null-terminate the preamble. - */ - packet_preamble[packet_preamble_len] = '\0'; + /* + * Null-terminate the preamble. + */ + packet_preamble[packet_preamble_len] = '\0'; - /* Ensure preamble has more than two chars before atempting to parse. - * This should cover line breaks etc that get counted. - */ - if ( strlen(packet_preamble) > 2 ) { - /* Get Time leaving subseconds */ - subsecs = strptime( packet_preamble, ts_fmt, &timecode ); - if (subsecs != NULL) { - /* Get the long time from the tm structure */ - /* (will return -1 if failure) */ - ts_sec = mktime( &timecode ); - } else - ts_sec = -1; /* we failed to parse it */ + /* Ensure preamble has more than two chars before atempting to parse. + * This should cover line breaks etc that get counted. + */ + if ( strlen(packet_preamble) > 2 ) { + /* Get Time leaving subseconds */ + subsecs = strptime( packet_preamble, ts_fmt, &timecode ); + if (subsecs != NULL) { + /* Get the long time from the tm structure */ + /* (will return -1 if failure) */ + ts_sec = mktime( &timecode ); + } else + ts_sec = -1; /* we failed to parse it */ - /* This will ensure incorrectly parsed dates get set to zero */ - if ( -1 == ts_sec ) - { - /* Sanitize - remove all '\r' */ - char *c; - while ((c = strchr(packet_preamble, '\r')) != NULL) *c=' '; - fprintf (stderr, "Failure processing time \"%s\" using time format \"%s\"\n (defaulting to Jan 1,1970 00:00:00 GMT)\n", - packet_preamble, ts_fmt); - if (debug >= 2) { - fprintf(stderr, "timecode: %02d/%02d/%d %02d:%02d:%02d %d\n", - timecode.tm_mday, timecode.tm_mon, timecode.tm_year, - timecode.tm_hour, timecode.tm_min, timecode.tm_sec, timecode.tm_isdst); - } - ts_sec = 0; /* Jan 1,1970: 00:00 GMT; tshark/wireshark will display date/time as adjusted by timezone */ - ts_usec = 0; - } - else - { - /* Parse subseconds */ - ts_usec = strtol(subsecs, &p, 10); - if (subsecs == p) { - /* Error */ - ts_usec = 0; - } else { - /* - * Convert that number to a number - * of microseconds; if it's N digits - * long, it's in units of 10^(-N) seconds, - * so, to convert it to units of - * 10^-6 seconds, we multiply by - * 10^(6-N). - */ - subseclen = (int) (p - subsecs); - if (subseclen > 6) { - /* - * *More* than 6 digits; 6-N is - * negative, so we divide by - * 10^(N-6). - */ - for (i = subseclen - 6; i != 0; i--) - ts_usec /= 10; - } else if (subseclen < 6) { - for (i = 6 - subseclen; i != 0; i--) - ts_usec *= 10; - } - } - } - } - if (debug >= 2) { - char *c; - while ((c = strchr(packet_preamble, '\r')) != NULL) *c=' '; - fprintf(stderr, "[[parse_preamble: \"%s\"]]\n", packet_preamble); - fprintf(stderr, "Format(%s), time(%u), subsecs(%u)\n", ts_fmt, (guint32)ts_sec, ts_usec); - } + /* This will ensure incorrectly parsed dates get set to zero */ + if ( -1 == ts_sec ) + { + /* Sanitize - remove all '\r' */ + char *c; + while ((c = strchr(packet_preamble, '\r')) != NULL) *c=' '; + fprintf (stderr, "Failure processing time \"%s\" using time format \"%s\"\n (defaulting to Jan 1,1970 00:00:00 GMT)\n", + packet_preamble, ts_fmt); + if (debug >= 2) { + fprintf(stderr, "timecode: %02d/%02d/%d %02d:%02d:%02d %d\n", + timecode.tm_mday, timecode.tm_mon, timecode.tm_year, + timecode.tm_hour, timecode.tm_min, timecode.tm_sec, timecode.tm_isdst); + } + ts_sec = 0; /* Jan 1,1970: 00:00 GMT; tshark/wireshark will display date/time as adjusted by timezone */ + ts_usec = 0; + } + else + { + /* Parse subseconds */ + ts_usec = strtol(subsecs, &p, 10); + if (subsecs == p) { + /* Error */ + ts_usec = 0; + } else { + /* + * Convert that number to a number + * of microseconds; if it's N digits + * long, it's in units of 10^(-N) seconds, + * so, to convert it to units of + * 10^-6 seconds, we multiply by + * 10^(6-N). + */ + subseclen = (int) (p - subsecs); + if (subseclen > 6) { + /* + * *More* than 6 digits; 6-N is + * negative, so we divide by + * 10^(N-6). + */ + for (i = subseclen - 6; i != 0; i--) + ts_usec /= 10; + } else if (subseclen < 6) { + for (i = 6 - subseclen; i != 0; i--) + ts_usec *= 10; + } + } + } + } + if (debug >= 2) { + char *c; + while ((c = strchr(packet_preamble, '\r')) != NULL) *c=' '; + fprintf(stderr, "[[parse_preamble: \"%s\"]]\n", packet_preamble); + fprintf(stderr, "Format(%s), time(%u), subsecs(%u)\n", ts_fmt, (guint32)ts_sec, ts_usec); + } - /* Clear Preamble */ - packet_preamble_len = 0; + /* Clear Preamble */ + packet_preamble_len = 0; } /*---------------------------------------------------------------------- @@ -978,3 +978,16 @@ text_import_cleanup(void) { g_free(packet_buf); } + +/* + * Editor modelines + * + * Local Variables: + * c-basic-offset: 4 + * tab-width: 8 + * indent-tabs-mode: nil + * End: + * + * ex: set shiftwidth=4 tabstop=8 expandtab: + * :indentSize=4:tabSize=8:noTabs=true: + */ diff --git a/ui/text_import.h b/ui/text_import.h index fc69db9c3c..1f789e3f27 100644 --- a/ui/text_import.h +++ b/ui/text_import.h @@ -94,3 +94,16 @@ void text_import_cleanup(void); #endif /* __cplusplus */ #endif /* __TEXT_IMPORT_H__ */ + +/* + * Editor modelines + * + * Local Variables: + * c-basic-offset: 4 + * tab-width: 8 + * indent-tabs-mode: nil + * End: + * + * ex: set shiftwidth=4 tabstop=8 expandtab: + * :indentSize=4:tabSize=8:noTabs=true: + */