Thanks for submitting the patch!

First thing that springs to mind after applying it:

gmake[2]: Entering directory '/home/joerg/src/avrdude'
  CC       libavrdude_a-jtag3.o
jtag3.c:1031:12: warning: 'jtag3_edbg_recv_frame' defined but not used [-Wunused-function]
 static int jtag3_edbg_recv_frame(PROGRAMMER * pgm, unsigned char **msg) {
            ^~~~~~~~~~~~~~~~~~~~~

So why is this function no longer going to be used?

After some closer inspection, I have a number of stylistic issues with the patch before I'd like to apply it.

 programmer
+  id    = "atmelice_avr32";
+  desc  = "Atmel-ICE (ARM/AVR) in ISP mode";

Really in ISP mode?

+        if (data[0]==SCOPE_AVR32)
+        {
+            char *str=0;
...
+              case AVR32_FAILURE_RECEIVE_LENGTH: str="Incorrect packet length"; break;
...
+  if (len > (((USBDEV_MAX_XFER_3-4)*16)-4))

I would appreciate if the patch keeps the existing style rule of having white space around operators (=, == etc.).

-         avrdude_message(MSG_INFO, "\\%03o", data[i]);
+          //avrdude_message(MSG_INFO, "\\%03o", data[i]);
+          avrdude_message(MSG_INFO, "\\%02x", data[i]);
...
+#if 0
   /* 4 bytes overhead for CMD, fragment #, and length info */
   int max_xfer = pgm->fd.usb.max_xfer;
   int nfragments = (len + max_xfer - 1) / max_xfer;
@@ -507,6 +647,7 @@
         }
       data += this_len;
       len -= this_len;
+#endif
     }

Source code comments and conditional compilation is not the way to keep track of history. This is what SVN (or whatever VCS is used) is for.

Also, as you can see in the first snippet, make sure to not artificially alter the indentation style. Unfortunately, AVRDUDE's code base was never very consistent on whether to use TAB or spaces, but if you change a single existing line, do not convert a TAB to spaces or vice versa.

+  //if ((buf = malloc(USBDEV_MAX_XFER_3)) == NULL) {
+  //  avrdude_message(MSG_INFO, "%s: jtag3_edbg_recv(): out of memory\n",
+       //    progname);
+  //  return -1;
+  //}

See above. If some piece of code isn't needed anymore, remove it, rather than commenting it out.

     if ((c == RSP3_FAILED) && ((*resp)[3] == RSP3_FAIL_OCD_LOCKED)) {
       avrdude_message(MSG_INFO,
-                     "%s: Device is locked! Chip erase required to unlock.\n",
-                     progname);
+          "%s: Device is locked! Chip erase required to unlock.\n",
+          progname);
     } else {
       avrdude_message(MSG_INFO, "%s: bad response to %s command: 0x%02x\n",
-                     progname, descr, c);
+          progname, descr, c);

Again, random indentation changes, please avoid this. Tell you editor to not touch anything else but what you're actually editing.

-  free(resp);
+  if (resp) free(resp);

Unneeded. The C standard guarantees free(0) does nothing.

The entire patch looks quite intrusive to the existing jtag3.c file. Thus, it imposes quite a bit of risk of breaking things that are unrelated to the AVR32 addition (see the first warning about a now unused function).

For that reason, I'd like to defer inclusion of the patch until after the next AVRDUDE release. Sorry, I don't have the time and energy of maintaining a separate -stable branch.

However, even though I'm setting it to "postponed" now, please feel encouraged to re-submit the patch with the above things fixed.