TXT record not properly translated to textAttributes

[annovanvliet]

The key-values pairs in a TXT record are not always properly translated to a Map with the correct textAttributes, when the values contain white-spaces.

IN:
tester1@mymachine._presence._tcp.local. 4500 IN TXT "last=van Tester" "hash=SHA-1" "status=avail" "ver=NfJ3flxICtbypursw=" "msg=A Message with whitespace !" "node=http://www.igniterealtime.org/projects/smack" "email=null" "1st=Test1" "port.p2ppj=2300" "txtvers=1"

OUT:
registerService Services Successfully Registered: Service ("tester1@mymachine._presence._tcp.local." can be reached at "MYMACHINE.local." [MYMACHINE.local./192.168.2.9, MYMACHINE.local./fe80:0:0:0:e5c1:5739:49ad:3b0c] on port 2300
TXT: last="van", Tester="", hash="SHA-1", status="avail", ver="NfJ3flI83zSdUDzCEICtbypursw=", msg="A", Message="",

When browsing through the code, I would say that the error is in the split function of the ServiceInstance. A boolean 'inQuote' is declared and set, but not used.

[timmattison]

Likely the fix for this would go here:

mdnsjava/mdnsjava/src/main/java/net/posick/mDNS/ServiceInstance.java

Line 456 in 747725f

break;

Alternatively it could be done with a regex which might be slower but would be more concise. Any preference on how this is remedied @annovanvliet?

[timmattison]

Just verified that there's no test coverage of this code. If possible it'd be good to add a test in with your data so if I make a patch I can validate that I'm fixing exactly what you're seeing.

[annovanvliet]

A unit test is for sure useful.

IMHO the fix is as follows
if ( inQuote || !Character.isWhitespace(chars[index]))
instead of
if ( !Character.isWhitespace(chars[index]))

For sure a Regexp is possible also, but the current solution is fine and fast (when correctly implemented)

[ColdForest]

Ya, I just wanted to second that I'm seeing this issue also. I'll look at patching ServiceInstance::split() as above until a new version is released.

[ColdForest]

The suggested fix definitely doesn't work. I'll need to review this more closely...

[ColdForest]

From RFC 6763, section 6.5:

If there is an '=' in a DNS-SD TXT record string, then everything after the first '=' to the end of the string is the value. The value can contain any eight-bit values including '='. The value MUST NOT be enclosed in additional quotation marks or any similar punctuation; any quotation marks, or leading or trailing spaces, are part of the value.

So I'm not seeing why there should be any support for quoting, regardless of escaping. It seems like split() in this context should simply return the entire string (in this context whitespace is indeed significant). Perhaps this is uniquely related to parsing TXTRecords; for these object types the TXT response elements are each passed in as a single "key=value" string and the result of split should be to just return the identity (no splitting needs to occur).

Not sure what the correct change here is, considering the ServiceInstance object can be constructed with several different types of "text attributes".

Any other thoughts here?

BTW, this is super easy to demonstrate (as the OP indicates) with the following avahi service configuration and then dnssd -B _test._tcp local:

  <service>\
    <type>_test._tcp</type>  
    <port>9999</port>  
    <txt-record>x=1 2 3</txt-record>  
    <txt-record>y=ok</txt-record>  
  </service>

[posicks]

The original writing and examples of the DNS-SD standard required a txtvers value and multiple values per text RR. I will have to re-read those sections, add the face changed from draft.

[ColdForest]

Thx Steven. Here's my simple, working kludge that properly preserves the whitespace in these records:

diff --git a/mdnsjava/src/main/java/net/posick/mDNS/ServiceInstance.java b/mdnsjava/src/main/java/net/posick/mDNS/ServiceInstance.java
index b838383..963a29f 100644
--- a/mdnsjava/src/main/java/net/posick/mDNS/ServiceInstance.java
+++ b/mdnsjava/src/main/java/net/posick/mDNS/ServiceInstance.java
@@ -388,7 +388,13 @@ public class ServiceInstance implements Serializable
         } else
         {
             Map textAttributes = new LinkedHashMap();
-            String[] pairs = split(rawText.toString());
+//            System.out.println("text: " + rawText.toString());
+            //String[] pairs = split(rawText.toString());
+            // (don't attempt any split() since:
+            //    1. split() is broken
+            //    2. each string is a full txt record so just treat this as "key=value"
+            String[] pairs =  { rawText.toString() };
+//            System.out.println("split output: " + Arrays.toString(pairs));
             for (String pair : pairs)
             {
                 if ((pair != null) && (pair.length() > 0))
@@ -461,6 +467,7 @@ public class ServiceInstance implements Serializable
                 builder.setLength(0);
             }
         }
+        list.add(text);
         
         return (String[]) list.toArray(new String[list.size()]);
     }

[deakjahn]

Still a problem after so many years. I have a printer that returns ty=Xerox Phaser 6130N and it will be parsed into three pairs:

ty="Xerox", Phaser="", 6130N=""

Could you, please, accept that simple modification above that actually works?