A user reported a very strange issue - when a document with a date field is saved, it changes the value one day to the past. With every save. But only for some dates, not all. It turned out to be a mystery that goes deep into XPages and Notes/Java APIs.
I've posted a sample on OpenNTF Discord and Serdar tried it on his server - no issue. But he uses the GMT zone and I have CET (Windows set to UTC+1 - Amsterdam, Berlin... to be precise).
To cut it short, the issue is caused by daylight saving interpretation between Notes and Java. The date fields (because XPages have no notion of real date-only fields) are stored with 00:00 time component and for some dates the conversion back to Java Date resulted in 23:00 on the previous day. XPages that get the date component as String for the input field, which is then saved back as a previous day during document save.
The app is full of date fields and I couldn't add custom logic to every save operation, so I tried to fix it at XPages converter level. The problem actually happens before that as the Date object in the converter is already wrong, but if I could intercept had massage the value, the rest of the application can remain unaffected.
API test
But before fixing the problem, we need to see how big the problem is. I was able to create a Java agent with a loop that reports all dates that "jump". You can try it on your machine (and your time-zone).
import java.text.SimpleDateFormat;
import java.util.Calendar;
import java.util.Date;
import java.util.GregorianCalendar;
import lotus.domino.AgentBase;
import lotus.domino.AgentContext;
import lotus.domino.DateTime;
import lotus.domino.Session;
public class JavaAgent extends AgentBase {
public void NotesMain() {
try {
SimpleDateFormat sdf = new SimpleDateFormat("dd.MM.yyyy");
SimpleDateFormat sdf2y = new SimpleDateFormat("dd.MM.yy");
Session session = getSession();
AgentContext agentContext = session.getAgentContext();
Calendar cal = new GregorianCalendar(2024, 7, 21);
for (int i = 0; i < 50000; i++) {
// create a new date and let Notes decide about the DST
DateTime dt = session.createDateTime(sdf.format(cal.getTime()) + " 00:00:00");
Date date = dt.toJavaDate();
String notesDate = dt.toString().split(" ")[0];
String javaDate = sdf.format(date).split(" ")[0];
String javaDate2y = sdf2y.format(date).split(" ")[0];
if (!notesDate.equals(javaDate) && !notesDate.equals(javaDate2y)) {
System.out.println("Notes: " + dt);
System.out.println("Java: " + date);
}
cal.add(Calendar.DAY_OF_MONTH, -1);
}
} catch (Exception e) {
e.printStackTrace();
}
}
}
15.02.2024 10:13:07 Agent Manager: Agent printing: Notes: 29.10.95 00:00:00 CEDT
15.02.2024 10:13:07 Agent Manager: Agent printing: Java: Sat Oct 28 23:00:00 CET 1995
15.02.2024 10:13:07 Agent Manager: Agent printing: Notes: 28.10.95 00:00:00 CEDT
15.02.2024 10:13:07 Agent Manager: Agent printing: Java: Fri Oct 27 23:00:00 CET 1995
15.02.2024 10:13:07 Agent Manager: Agent printing: Notes: 27.10.95 00:00:00 CEDT
15.02.2024 10:13:07 Agent Manager: Agent printing: Java: Thu Oct 26 23:00:00 CET 1995
So the problem is bigger that just a few problematic dates.
Fix routine
The only way how I was able to get the correct date was from DominoDocument.getDocument().getItemValueDateTimeArray(..). But I can't do that in a XPages converter. With some experiments, I have found that I can use different ways to convert the Date back to DateTime and then it can give me a hint that the problem occurred. And then I can adjust the date.
It turned out that if I create the DateTime from a Calendar instead, I get a different DateTime. Strangely, if I convert that one back to Java Date, I get the wrong date again. But we can get the difference and adjust the original date.
private static Date fixDate(Session s, Date origDate) {
Date res = origDate;
try {
DateTime dateTime = s.createDateTime(origDate);
Calendar c = Calendar.getInstance();
c.setTime(origDate);
DateTime calDateTime = s.createDateTime(c);
if (!StringUtil.equals(calDateTime.getDateOnly(), dateTime.getDateOnly())) {
int diff = dateTime.timeDifference(calDateTime);
c.add(Calendar.SECOND, -diff);
res = c.getTime();
}
} catch (NotesException e) {
e.printStackTrace();
}
return res;
}Taking this to XPages
Ok, we have a snippet that can save the situation, but injecting it into XPages was harder than expected. The standard converter generates a client-side converter object, which doesn't work with custom converters, so I can't just configure a custom converter on the field (and replacing the converter globally doesn't work as XPages add their one for javax.facesDateTime after processing the app faces config - I learned this the hard way in the past).
Luckily, Sven Hasselbach blogged about a similar problem for validators - http://hasselba.ch/blog/?p=764. So I can inject the converter via code, so all the instanceof checks then work nicely.
And those few lines seem to get the situation under control. Even if HCL fixes the problem in the API, the code can stay in the app as it does nothing to dates that work already correctly. It doesn't have any hardcoded offset of format, so it can help in other similar situations too. I ran the test agent with the fix routine and no dates were reported, so it should not break other dates.
<xp:this.afterPageLoad>
<![CDATA[#{javascript:getComponent(compositeData.inputId).setConverter(
new com.pradny.domino.converter.CustomDateConverter()
);}]]>
</xp:this.afterPageLoad>The converter itself just extends the standard one and uses a pattern defined in the app. It's designed just for this single scenario, so it can't be used for other formats, but it's fine as it's applied only to components in custom controls that use this format.
package comm.pradny.domino.converter;
import java.text.SimpleDateFormat;
import java.util.Calendar;
import java.util.Date;
import javax.faces.component.UIComponent;
import javax.faces.context.FacesContext;
import com.ibm.commons.util.StringUtil;
import com.ibm.xsp.convert.DateTimeConverter;
import com.ibm.xsp.extlib.util.ExtLibUtil;
import icrc.openLog.OpenLogItem;
import icrc.welcome.Settings;
import lotus.domino.DateTime;
import lotus.domino.NotesException;
import lotus.domino.Session;
public class CustomDateConverter extends DateTimeConverter {
private static final SimpleDateFormat formatter = new java.text.SimpleDateFormat(Settings.DATE_PATTERN);
public CustomDateConverter() {
super();
setPattern(Settings.DATE_PATTERN);
}
private Date fixDate(Session s, Date origDate) {
Date res = origDate;
try {
DateTime dateTime = s.createDateTime(origDate);
Calendar c = Calendar.getInstance();
c.setTime(origDate);
DateTime calDateTime = s.createDateTime(c);
if (!StringUtil.equals(calDateTime.getDateOnly(), dateTime.getDateOnly())) {
int diff = dateTime.timeDifference(calDateTime);
c.add(Calendar.SECOND, -diff);
res = c.getTime();
}
} catch (NotesException e) {
OpenLogItem.logError(e);
}
return res;
}
@Override
public String getAsString(FacesContext context, UIComponent component, Object value) {
if (value instanceof Date) {
Date origDate = fixDate(ExtLibUtil.getCurrentSession(), (Date) value);
String s = formatter.format(origDate);
return s;
} else {
throw new IllegalArgumentException("Value is not a java.util.Date");
}
}
}Conclusion
The most surprising thing about this problem is that I haven't seen it before. I assume it was always in the API, but I probably never worked intensively with dates before 1995 in an app. In this case, it was detected with date of birth fields.
I hate time-zone issues and daylight issues even more. The discussion about stopping of EU went completely silent now, but it would be great to get rid of it. Of course, the change would break many systems and cause headaches for many IT people.
I hope that HCL will find a cure at the low API level for this issue as it's not just an XPages issue. Obviously, the proper solution is to get LocalDate implementation in the API and XPages, but the code heavily relies on the old Date class and it would be challenging to replace it everywhere.
Comments
Post a Comment