mirror of
https://github.com/saitohirga/WSJT-X.git
synced 2024-11-29 15:48:38 -05:00
fb1b2e7687
-----BEGIN PGP SIGNATURE----- iQEzBAABCAAdFiEEUELwYiy7vDRM+hTD2GSwbR6BYY8FAl3a8/QACgkQ2GSwbR6B YY8y+ggAqwaaCRMCSwDzStooYjStJErVoZ/yB//MmoF+798TBb+3MkiT+KclW7aN rxTOmEJ+a84PmsiW6Qz/AAcHPnbKDhtwWASZooopgNYDfgyaUib3HfIa+cFJO6my OUzlTsH94yDxeBCWOtf1dLiP0YwJKfp/vxzbfDAWXTd4dx3Ok/7KwghYocmP3eCq ruZ+dWojQzwE+h50P6B9gTFxGWMHUDDNeZ/c8u494wqDQvQaRoM9oOnQWzlGiMis JaerraI52JEdFLnXjtLYeyXtfg+zA0KvRY2DWy1aCPpGkNjy3DPNlOA+cDQpzBru NIqsUF+GplgkVs3sKNLXq36ZgaBQBg== =5w4B -----END PGP SIGNATURE----- gpgsig -----BEGIN PGP SIGNATURE----- iQEzBAABCAAdFiEEUELwYiy7vDRM+hTD2GSwbR6BYY8FAl3a9R8ACgkQ2GSwbR6B YY9SWQf+JdfIvlHy8ExSs9ImCLck/e66V2OAqaEONQX1bEC4QiG3M/rVJGgf2iFz ZFshtilsajlpsYZQIFToTmJmI76eGLAmyNIcZN2WpcJQMNWZnxqklMj2wIHbRh+g Pvw3FC+lc4LUjZCCMYrVjNXxsRmQtjqV1lVFBnmrAWCdrgHmhyrVbPZT4B+fRTXL ikw/BiRgxhI0aoIWKzNrtztMYUBjanKLTHE81AVgsgZkBDNybOjR7mE2E/Jwc4Qx mVu+Z0uI+bVuBlMOrqp0K+rrUz7qMZe5/AMSRbgzwfueEOIShgu2wrfV++3JDXLH AhtIAcL/b39RBx0jkFE8rBKaZpkoAw== =DNp6 -----END PGP SIGNATURE----- Merge tag 'wsjtx-2.1.1' into develop Tagging the WSJT-X v2.1.1 release
212 lines
6.2 KiB
C++
212 lines
6.2 KiB
C++
#include "decodedtext.h"
|
|
|
|
#include <QStringList>
|
|
#include <QRegularExpression>
|
|
#include <QDebug>
|
|
|
|
extern "C" {
|
|
bool stdmsg_(char const * msg, fortran_charlen_t);
|
|
}
|
|
|
|
namespace
|
|
{
|
|
QRegularExpression words_re {R"(^(?:(?<word1>(?:CQ|DE|QRZ)(?:\s?DX|\s(?:[A-Z]{1,4}|\d{3}))|[A-Z0-9/]+|\.{3})\s)(?:(?<word2>[A-Z0-9/]+)(?:\s(?<word3>[-+A-Z0-9]+)(?:\s(?<word4>(?:OOO|(?!RR73)[A-R]{2}[0-9]{2})))?)?)?)"};
|
|
}
|
|
|
|
DecodedText::DecodedText (QString const& the_string)
|
|
: string_ {the_string.left (the_string.indexOf (QChar::Nbsp))} // discard appended info
|
|
, padding_ {string_.indexOf (" ") > 4 ? 2 : 0} // allow for
|
|
// seconds
|
|
, message_ {string_.mid (column_qsoText + padding_).trimmed ()}
|
|
, is_standard_ {false}
|
|
{
|
|
// qDebug () << "DecodedText: the_string:" << the_string << "Nbsp pos:" << the_string.indexOf (QChar::Nbsp);
|
|
if (message_.length() >= 1)
|
|
{
|
|
message0_ = message_.left(36);
|
|
message_ = message_.left(36).remove (QRegularExpression {"[<>]"});
|
|
int i1 = message_.indexOf ('\r');
|
|
if (i1 > 0)
|
|
{
|
|
message_ = message_.left (i1 - 1);
|
|
}
|
|
if (message_.contains (QRegularExpression {"^(CQ|QRZ)\\s"}))
|
|
{
|
|
// TODO this magic position 16 is guaranteed to be after the
|
|
// last space in a decoded CQ or QRZ message but before any
|
|
// appended DXCC entity name or worked before information
|
|
auto eom_pos = message_.indexOf (' ', 16);
|
|
// we always want at least the characters to position 16
|
|
if (eom_pos < 16) eom_pos = message_.size () - 1;
|
|
// remove DXCC entity and worked B4 status. TODO need a better way to do this
|
|
message_ = message_.left (eom_pos + 1);
|
|
}
|
|
// stdmsg is a Fortran routine that packs the text, unpacks it
|
|
// and compares the result
|
|
auto message_c_string = message0_.toLocal8Bit ();
|
|
message_c_string += QByteArray {37 - message_c_string.size (), ' '};
|
|
is_standard_ = stdmsg_(message_c_string.constData(),37);
|
|
}
|
|
};
|
|
|
|
QStringList DecodedText::messageWords () const
|
|
{
|
|
if(is_standard_) {
|
|
// extract up to the first four message words
|
|
QString t=message_;
|
|
if(t.left(4)=="TU; ") t=message_.mid(4,-1);
|
|
return words_re.match(t).capturedTexts();
|
|
}
|
|
// simple word split for free text messages
|
|
auto words = message_.split (' ', QString::SkipEmptyParts);
|
|
// add whole message as item 0 to mimic RE capture list
|
|
words.prepend (message_);
|
|
return words;
|
|
}
|
|
|
|
QString DecodedText::CQersCall() const
|
|
{
|
|
QRegularExpression callsign_re {R"(^(CQ|DE|QRZ)(\s?DX|\s([A-Z]{1,4}|\d{3}))?\s(?<callsign>[A-Z0-9/]{2,})(\s[A-R]{2}[0-9]{2})?)"};
|
|
return callsign_re.match (message_).captured ("callsign");
|
|
}
|
|
|
|
|
|
bool DecodedText::isJT65() const
|
|
{
|
|
return string_.indexOf("#") == column_mode + padding_;
|
|
}
|
|
|
|
bool DecodedText::isJT9() const
|
|
{
|
|
return string_.indexOf("@") == column_mode + padding_;
|
|
}
|
|
|
|
bool DecodedText::isTX() const
|
|
{
|
|
int i = string_.indexOf("Tx");
|
|
return (i >= 0 && i < 15); // TODO guessing those numbers. Does Tx ever move?
|
|
}
|
|
|
|
bool DecodedText::isLowConfidence () const
|
|
{
|
|
return QChar {'?'} == string_.mid (padding_ + column_qsoText + 21, 1);
|
|
}
|
|
|
|
int DecodedText::frequencyOffset() const
|
|
{
|
|
return string_.mid(column_freq + padding_,4).toInt();
|
|
}
|
|
|
|
int DecodedText::snr() const
|
|
{
|
|
int i1=string_.indexOf(" ")+1;
|
|
return string_.mid(i1,3).toInt();
|
|
}
|
|
|
|
float DecodedText::dt() const
|
|
{
|
|
return string_.mid(column_dt + padding_,5).toFloat();
|
|
}
|
|
|
|
/*
|
|
2343 -11 0.8 1259 # YV6BFE F6GUU R-08
|
|
2343 -19 0.3 718 # VE6WQ SQ2NIJ -14
|
|
2343 -7 0.3 815 # KK4DSD W7VP -16
|
|
2343 -13 0.1 3627 @ CT1FBK IK5YZT R+02
|
|
|
|
0605 Tx 1259 # CQ VK3ACF QF22
|
|
*/
|
|
|
|
// find and extract any report. Returns true if this is a standard message
|
|
bool DecodedText::report(QString const& myBaseCall, QString const& dxBaseCall, /*mod*/QString& report) const
|
|
{
|
|
if (message_.size () < 1) return false;
|
|
|
|
QStringList const& w = message_.split(" ",QString::SkipEmptyParts);
|
|
if (w.size ()
|
|
&& is_standard_ && (w[0] == myBaseCall
|
|
|| w[0].endsWith ("/" + myBaseCall)
|
|
|| w[0].startsWith (myBaseCall + "/")
|
|
|| (w.size () > 1 && !dxBaseCall.isEmpty ()
|
|
&& (w[1] == dxBaseCall
|
|
|| w[1].endsWith ("/" + dxBaseCall)
|
|
|| w[1].startsWith (dxBaseCall + "/")))))
|
|
{
|
|
QString tt="";
|
|
if(w.size() > 2) tt=w[2];
|
|
bool ok;
|
|
auto i1=tt.toInt(&ok);
|
|
if (ok and i1>=-50 and i1<50)
|
|
{
|
|
report = tt;
|
|
}
|
|
else
|
|
{
|
|
if (tt.mid(0,1)=="R")
|
|
{
|
|
i1=tt.mid(1).toInt(&ok);
|
|
if(ok and i1>=-50 and i1<50)
|
|
{
|
|
report = tt.mid(1);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return is_standard_;
|
|
}
|
|
|
|
// get the first text word, usually the call
|
|
QString DecodedText::call() const
|
|
{
|
|
return words_re.match (message_).captured ("word1");
|
|
}
|
|
|
|
// get the second word, most likely the de call and the third word, most likely grid
|
|
void DecodedText::deCallAndGrid(/*out*/QString& call, QString& grid) const
|
|
{
|
|
auto const& match = words_re.match (message_);
|
|
call = match.captured ("word2");
|
|
grid = match.captured ("word3");
|
|
if ("R" == grid) grid = match.captured ("word4");
|
|
}
|
|
|
|
unsigned DecodedText::timeInSeconds() const
|
|
{
|
|
return 3600 * string_.mid (column_time, 2).toUInt ()
|
|
+ 60 * string_.mid (column_time + 2, 2).toUInt()
|
|
+ (padding_ ? string_.mid (column_time + 2 + padding_, 2).toUInt () : 0U);
|
|
}
|
|
|
|
/*
|
|
2343 -11 0.8 1259 # YV6BFE F6GUU R-08
|
|
2343 -19 0.3 718 # VE6WQ SQ2NIJ -14
|
|
2343 -7 0.3 815 # KK4DSD W7VP -16
|
|
2343 -13 0.1 3627 @ CT1FBK IK5YZT R+02
|
|
|
|
0605 Tx 1259 # CQ VK3ACF QF22
|
|
*/
|
|
|
|
QString DecodedText::report() const // returns a string of the SNR field with a leading + or - followed by two digits
|
|
{
|
|
int sr = snr();
|
|
if (sr<-50)
|
|
sr = -50;
|
|
else
|
|
if (sr > 49)
|
|
sr = 49;
|
|
|
|
QString rpt;
|
|
rpt.sprintf("%d",abs(sr));
|
|
if (sr > 9)
|
|
rpt = "+" + rpt;
|
|
else
|
|
if (sr >= 0)
|
|
rpt = "+0" + rpt;
|
|
else
|
|
if (sr >= -9)
|
|
rpt = "-0" + rpt;
|
|
else
|
|
rpt = "-" + rpt;
|
|
return rpt;
|
|
}
|