3aaa52af8d
X-SVN-Rev: 596
337 lines
12 KiB
C++
337 lines
12 KiB
C++
/********************************************************************
|
|
* COPYRIGHT:
|
|
* Copyright (c) 1997-1999, International Business Machines Corporation and
|
|
* others. All Rights Reserved.
|
|
********************************************************************/
|
|
|
|
#include "dtfmtrtts.h"
|
|
|
|
#include <stdio.h>
|
|
|
|
#include "unicode/datefmt.h"
|
|
#include "unicode/smpdtfmt.h"
|
|
#include "unicode/gregocal.h"
|
|
|
|
|
|
// *****************************************************************************
|
|
// class DateFormatRoundTripTest
|
|
// *****************************************************************************
|
|
|
|
// Useful for turning up subtle bugs: Change the following to TRUE, recompile,
|
|
// and run while at lunch.
|
|
bool_t DateFormatRoundTripTest::INFINITE = FALSE; // Warning -- makes test run infinite loop!!!
|
|
|
|
// If SPARSENESS is > 0, we don't run each exhaustive possibility.
|
|
// There are 24 total possible tests per each locale. A SPARSENESS
|
|
// of 12 means we run half of them. A SPARSENESS of 23 means we run
|
|
// 1 of them. SPARSENESS _must_ be in the range 0..23.
|
|
int32_t DateFormatRoundTripTest::SPARSENESS = 18;
|
|
int32_t DateFormatRoundTripTest::TRIALS = 4;
|
|
int32_t DateFormatRoundTripTest::DEPTH = 5;
|
|
|
|
|
|
#define CASE(id,test) case id: name = #test; if (exec) { logln(#test "---"); logln((UnicodeString)""); test(); } break;
|
|
|
|
void
|
|
DateFormatRoundTripTest::runIndexedTest( int32_t index, bool_t exec, char* &name, char* par )
|
|
{
|
|
// if (exec) logln((UnicodeString)"TestSuite NumberFormatRegressionTest");
|
|
switch (index) {
|
|
CASE(0,TestDateFormatRoundTrip)
|
|
default: name = ""; break;
|
|
}
|
|
}
|
|
|
|
bool_t
|
|
DateFormatRoundTripTest::failure(UErrorCode status, const char* msg)
|
|
{
|
|
if(U_FAILURE(status)) {
|
|
errln(UnicodeString("FAIL: ") + msg + " failed, error " + u_errorName(status));
|
|
return TRUE;
|
|
}
|
|
|
|
return FALSE;
|
|
}
|
|
|
|
// ==
|
|
|
|
void DateFormatRoundTripTest::TestDateFormatRoundTrip()
|
|
{
|
|
UErrorCode status = U_ZERO_ERROR;
|
|
dateFormat = new SimpleDateFormat("EEE MMM dd HH:mm:ss.SSS zzz yyyy G", status);
|
|
failure(status, "new SimpleDateFormat");
|
|
|
|
getFieldCal = Calendar::createInstance(status);
|
|
failure(status, "Calendar::createInstance");
|
|
|
|
|
|
int32_t locCount = 0;
|
|
const Locale *avail = DateFormat::getAvailableLocales(locCount);
|
|
logln("DateFormat available locales: " + locCount);
|
|
if(quick) {
|
|
if(locCount > 5)
|
|
locCount = 5;
|
|
logln("Quick mode: only testing first 5 Locales");
|
|
}
|
|
TimeZone *tz = TimeZone::createDefault();
|
|
UnicodeString temp;
|
|
logln("Default TimeZone: " + tz->getID(temp));
|
|
delete tz;
|
|
|
|
if (INFINITE) {
|
|
// Special infinite loop test mode for finding hard to reproduce errors
|
|
Locale loc = Locale::getDefault();
|
|
logln("ENTERING INFINITE TEST LOOP FOR Locale: " + loc.getDisplayName(temp));
|
|
for(;;)
|
|
test(loc);
|
|
}
|
|
else {
|
|
test(Locale::getDefault());
|
|
|
|
for (int i=0; i < locCount; ++i) {
|
|
test(avail[i]);
|
|
}
|
|
}
|
|
|
|
delete dateFormat;
|
|
delete getFieldCal;
|
|
}
|
|
|
|
void DateFormatRoundTripTest::test(const Locale& loc)
|
|
{
|
|
UnicodeString temp;
|
|
if( ! INFINITE)
|
|
logln("Locale: " + loc.getDisplayName(temp));
|
|
|
|
// Total possibilities = 24
|
|
// 4 date
|
|
// 4 time
|
|
// 16 date-time
|
|
bool_t TEST_TABLE [24];//= new boolean[24];
|
|
int32_t i = 0;
|
|
for(i = 0; i < 24; ++i)
|
|
TEST_TABLE[i] = TRUE;
|
|
|
|
// If we have some sparseness, implement it here. Sparseness decreases
|
|
// test time by eliminating some tests, up to 23.
|
|
for(i = 0; i < SPARSENESS; ) {
|
|
int random = (int)(randFraction() * 24);
|
|
if (random >= 0 && random < 24 && TEST_TABLE[i]) {
|
|
TEST_TABLE[i] = FALSE;
|
|
++i;
|
|
}
|
|
}
|
|
|
|
int32_t itable = 0;
|
|
int32_t style = 0;
|
|
for(style = DateFormat::FULL; style <= DateFormat::SHORT; ++style) {
|
|
if(TEST_TABLE[itable++]) {
|
|
DateFormat *df = DateFormat::createDateInstance((DateFormat::EStyle)style, loc);
|
|
test(df);
|
|
delete df;
|
|
}
|
|
}
|
|
|
|
for(style = DateFormat::FULL; style <= DateFormat::SHORT; ++style) {
|
|
if (TEST_TABLE[itable++]) {
|
|
DateFormat *df = DateFormat::createTimeInstance((DateFormat::EStyle)style, loc);
|
|
test(df, TRUE);
|
|
delete df;
|
|
}
|
|
}
|
|
|
|
for(int32_t dstyle = DateFormat::FULL; dstyle <= DateFormat::SHORT; ++dstyle) {
|
|
for(int32_t tstyle = DateFormat::FULL; tstyle <= DateFormat::SHORT; ++tstyle) {
|
|
if(TEST_TABLE[itable++]) {
|
|
DateFormat *df = DateFormat::createDateTimeInstance((DateFormat::EStyle)dstyle, (DateFormat::EStyle)tstyle, loc);
|
|
test(df);
|
|
delete df;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void DateFormatRoundTripTest::test(DateFormat *fmt, bool_t timeOnly)
|
|
{
|
|
UnicodeString pat;
|
|
if(fmt->getDynamicClassID() != SimpleDateFormat::getStaticClassID()) {
|
|
errln("DateFormat wasn't a SimpleDateFormat");
|
|
return;
|
|
}
|
|
|
|
pat = ((SimpleDateFormat*)fmt)->toPattern(pat);
|
|
|
|
// NOTE TO MAINTAINER
|
|
// This indexOf check into the pattern needs to be refined to ignore
|
|
// quoted characters. Currently, this isn't a problem with the locale
|
|
// patterns we have, but it may be a problem later.
|
|
|
|
bool_t hasEra = (pat.indexOf(UnicodeString("G")) != -1);
|
|
bool_t hasZone = (pat.indexOf(UnicodeString("z")) != -1);
|
|
|
|
// Because patterns contain incomplete data representing the Date,
|
|
// we must be careful of how we do the roundtrip. We start with
|
|
// a randomly generated Date because they're easier to generate.
|
|
// From this we get a string. The string is our real starting point,
|
|
// because this string should parse the same way all the time. Note
|
|
// that it will not necessarily parse back to the original date because
|
|
// of incompleteness in patterns. For example, a time-only pattern won't
|
|
// parse back to the same date.
|
|
|
|
//try {
|
|
for(int i = 0; i < TRIALS; ++i) {
|
|
UDate *d = new UDate [DEPTH];
|
|
UnicodeString *s = new UnicodeString[DEPTH];
|
|
|
|
d[0] = generateDate();
|
|
|
|
UErrorCode status = U_ZERO_ERROR;
|
|
|
|
// We go through this loop until we achieve a match or until
|
|
// the maximum loop count is reached. We record the points at
|
|
// which the date and the string starts to match. Once matching
|
|
// starts, it should continue.
|
|
int loop;
|
|
int dmatch = 0; // d[dmatch].getTime() == d[dmatch-1].getTime()
|
|
int smatch = 0; // s[smatch].equals(s[smatch-1])
|
|
for(loop = 0; loop < DEPTH; ++loop) {
|
|
if (loop > 0) {
|
|
d[loop] = fmt->parse(s[loop-1], status);
|
|
failure(status, "fmt->parse");
|
|
}
|
|
|
|
s[loop] = fmt->format(d[loop], s[loop]);
|
|
|
|
if(loop > 0) {
|
|
if(smatch == 0) {
|
|
bool_t match = s[loop] == s[loop-1];
|
|
if(smatch == 0) {
|
|
if(match)
|
|
smatch = loop;
|
|
}
|
|
else if( ! match)
|
|
errln("FAIL: String mismatch after match");
|
|
}
|
|
|
|
if(dmatch == 0) {
|
|
// {sfb} watch out here, this might not work
|
|
bool_t match = d[loop]/*.getTime()*/ == d[loop-1]/*.getTime()*/;
|
|
if(dmatch == 0) {
|
|
if(match)
|
|
dmatch = loop;
|
|
}
|
|
else if( ! match)
|
|
errln("FAIL: Date mismatch after match");
|
|
}
|
|
|
|
if(smatch != 0 && dmatch != 0)
|
|
break;
|
|
}
|
|
}
|
|
// At this point loop == DEPTH if we've failed, otherwise loop is the
|
|
// max(smatch, dmatch), that is, the index at which we have string and
|
|
// date matching.
|
|
|
|
// Date usually matches in 2. Exceptions handled below.
|
|
int maxDmatch = 2;
|
|
int maxSmatch = 1;
|
|
if (dmatch > maxDmatch) {
|
|
// Time-only pattern with zone information and a starting date in PST.
|
|
if(timeOnly && hasZone && fmt->getTimeZone().inDaylightTime(d[0], status) && ! failure(status, "TimeZone::inDST()")) {
|
|
maxDmatch = 3;
|
|
maxSmatch = 2;
|
|
}
|
|
}
|
|
|
|
// String usually matches in 1. Exceptions are checked for here.
|
|
if(smatch > maxSmatch) { // Don't compute unless necessary
|
|
// Starts in BC, with no era in pattern
|
|
if( ! hasEra && getField(d[0], Calendar::ERA) == GregorianCalendar::BC)
|
|
maxSmatch = 2;
|
|
// Starts in DST, no year in pattern
|
|
else if(fmt->getTimeZone().inDaylightTime(d[0], status) && ! failure(status, "foo") &&
|
|
pat.indexOf(UnicodeString("yyyy")) == -1)
|
|
maxSmatch = 2;
|
|
// Two digit year with zone and year change and zone in pattern
|
|
else if (hasZone &&
|
|
fmt->getTimeZone().inDaylightTime(d[0], status) !=
|
|
fmt->getTimeZone().inDaylightTime(d[dmatch], status) && ! failure(status, "foo") &&
|
|
getField(d[0], Calendar::YEAR) !=
|
|
getField(d[dmatch], Calendar::YEAR) &&
|
|
pat.indexOf(UnicodeString("y")) != -1 &&
|
|
pat.indexOf(UnicodeString("yyyy")) == -1)
|
|
maxSmatch = 2;
|
|
}
|
|
|
|
if(dmatch > maxDmatch || smatch > maxSmatch) {
|
|
errln(UnicodeString("Pattern: ") + pat);
|
|
logln(UnicodeString(" Date ") + dmatch + " String " + smatch);
|
|
|
|
for(int j = 0; j <= loop && j < DEPTH; ++j) {
|
|
UnicodeString temp;
|
|
FieldPosition pos(FieldPosition::DONT_CARE);
|
|
logln((j>0?" P> ":" ") + dateFormat->format(d[j], temp, pos) + " F> " +
|
|
escape(s[j], temp) +
|
|
(j > 0 && d[j]/*.getTime()*/==d[j-1]/*.getTime()*/?" d==":"") +
|
|
(j > 0 && s[j] == s[j-1]?" s==":""));
|
|
}
|
|
}
|
|
}
|
|
/*}
|
|
catch (ParseException e) {
|
|
errln("Exception: " + e.getMessage());
|
|
logln(e.toString());
|
|
}*/
|
|
}
|
|
|
|
/**
|
|
* Return a field of the given date
|
|
*/
|
|
int32_t DateFormatRoundTripTest::getField(UDate d, int32_t f) {
|
|
// Should be synchronized, but we're single threaded so it's ok
|
|
UErrorCode status = U_ZERO_ERROR;
|
|
getFieldCal->setTime(d, status);
|
|
failure(status, "getfieldCal->setTime");
|
|
int32_t ret = getFieldCal->get((Calendar::EDateFields)f, status);
|
|
failure(status, "getfieldCal->get");
|
|
return ret;
|
|
}
|
|
|
|
UnicodeString& DateFormatRoundTripTest::escape(const UnicodeString& src, UnicodeString& dst )
|
|
{
|
|
dst.remove();
|
|
for (int32_t i = 0; i < src.length(); ++i) {
|
|
UChar c = src[i];
|
|
if(c < 0x0080)
|
|
dst += c;
|
|
else {
|
|
dst += UnicodeString("[");
|
|
char buf [8];
|
|
sprintf(buf, "%#x", c);
|
|
dst += UnicodeString(buf);
|
|
dst += UnicodeString("]");
|
|
}
|
|
}
|
|
|
|
return dst;
|
|
}
|
|
|
|
UDate DateFormatRoundTripTest::generateDate()
|
|
{
|
|
double a = randFraction();
|
|
|
|
// Now 'a' ranges from 0..1; scale it to range from 0 to 8000 years
|
|
a *= 8000;
|
|
|
|
// Range from (4000-1970) BC to (8000-1970) AD
|
|
a -= 4000;
|
|
|
|
// Now scale up to ms
|
|
a *= 365.25 * 24 * 60 * 60 * 1000;
|
|
|
|
//return new Date((long)a);
|
|
return a;
|
|
}
|
|
|
|
//eof
|