Exiv2

Tobias,

That doesn't sound right and it certainly shouldn't crash. Can you provide an image or xmp file that I could play with? Please attach it here or send it to ahuggel <at> gmx <dot> net

Andreas

Hi Andreas,

here is a small XMP file and test program.

Tobias

I would like to make it clear that I don't think that crashing is the bug here (well, a minor one maybe), but that the empty element is not accounted for. And most importantly, I would like to know if there is any way at all to get that empty rdf:li element.

Is there anyone who can help? This issue is causing major problems in darktable and I would like to find a way to get that solved.

I'll help you with this, Tobias. I'm not aware of the issue. I'm not sure what I'm looking for as I don't get a crash (on current trunk build). It looks as though you have 4 elements in Seq:

685 rmills@rmillsmbp:~/Downloads $ xmllint -pretty 1 crash.xmp 
<?xml version="1.0"?>
<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
  <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
    <rdf:Description xmlns:darktable="http://darktable.sf.net/" rdf:about="">
      <darktable:multi_name>
        <rdf:Seq>
          <rdf:li>foo</rdf:li>
          <rdf:li>bar</rdf:li>
          <rdf:li/>
          <rdf:li>baz</rdf:li>
        </rdf:Seq>
      </darktable:multi_name>
    </rdf:Description>
  </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>
686 rmills@rmillsmbp:~/Downloads $ 

683 rmills@rmillsmbp:~/Downloads $ exiv2 -pa crash.xmp 
Xmp.darktable.multi_name                     XmpSeq      3  foo, bar, baz
684 rmills@rmillsmbp:~/Downloads $

// g++ -W -Wall -g `pkg-config --cflags --libs exiv2` -o crash crash.cc
#include <exiv2/easyaccess.hpp>
#include <exiv2/xmp.hpp>
#include <exiv2/error.hpp>
#include <exiv2/image.hpp>
#include <exiv2/exif.hpp>

#include <stdio.h>

int main()
{
  const char* filename = "crash.xmp";
  Exiv2::Image::AutoPtr image = Exiv2::ImageFactory::open(filename);
  if(image.get() == 0) {
      printf("unable to open file %s\n",filename);
      return -1 ;
  }

  int result = 0 ;
  image->readMetadata();

  Exiv2::XmpData                 &xmpData = image->xmpData();
  Exiv2::XmpData::const_iterator  pos     = xmpData.findKey(Exiv2::XmpKey("Xmp.darktable.multi_name"));
  if(pos != xmpData.end()) {
    printf("found %ld entries\n", pos->count());
    for ( int i = 0 ; i < pos->count() ; i++ ) {
        std::cout << i <<" : " << pos->toString(i) << std::endl;
    }
  } else {
      printf("not found\n");
      result = 1;
  }
  return result;
}

689 rmills@rmillsmbp:~/Downloads $ g++ -W -Wall -g `pkg-config --cflags --libs exiv2` -o crash crash.cc
690 rmills@rmillsmbp:~/Downloads $ ./crash
found 3 entries
0 : foo
1 : bar
2 : baz
691 rmills@rmillsmbp:~/Downloads $ 

1) There is no crash here. Correct?
2) We are reporting 3 li elements in Seq and you are expecting 4. Is that correct?

I don't know anything much about XMP. I'm a C++ build engineer. Is there a rule in XMP that says "ignore empty Seq/li elements?"

I've changed crash.xml. First, I changed <rdf:li/> to <rdf:li></rdf:li> and that made no difference. Then I changed it to <rdf:li>bullshit</rdf:li>. Good result:

693 rmills@rmillsmbp:~/Downloads $ ./crash
found 4 entries
0 : foo
1 : bar
2 : bullshit
3 : baz
694 rmills@rmillsmbp:~/Downloads $ 

Curiously, if I change "bullshit" in crash.xml to " " (a single space), the count == 4 and element 3 is a space. When I change "bullshit" in crash.xml to the empty string, the count returns to 3. There's something determined to hide empty list elements.

There's a discussion here about something similar: http://u88.n24.queensu.ca/exiftool/forum/index.php?topic=5237.msg25351#msg25351

Phil (Harvey) says:

The empty lists were removed, but no actual metadata was lost.

Darktable should not be writing empty lists. It is a waste of space.

More amazing discoveries after attempting to understand Adobe's and W3C's documents about XMP and RDF.

<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
 <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
  <rdf:Description rdf:about="" xmlns:darktable="http://darktable.sf.net/" >
   <darktable:multi_name>
    <rdf:Seq>
     <rdf:li>foo</rdf:li>
     <rdf:li>bar</rdf:li>
     <rdf:li rdf:resource=""/>
     <rdf:li>baz</rdf:li>
    </rdf:Seq>
   </darktable:multi_name>
  </rdf:Description>
 </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>

713 rmills@rmillsmbp:~/Downloads $ ./crash
found 3 entries
0 : foo
1 : bar
2 : baz
714 rmills@rmillsmbp:~/Downloads $ 

./crash
found 4 entries
0 : foo
1 : bar
2 : 911
3 : baz
715 rmills@rmillsmbp:~/Downloads $ 

Hello Robin,

thanks for looking into this. I am still on 0.25 which does crash. But as I said, that's only a side issue. The real thing is, as you correctly understood, the missing empty element. I haven't tried compiling libexiv2 myself, but I can give it a try to see how that works.

I did some debugging on this last night. There is no doubt in my mind that XMPsdk removes empty list elements. However, I don't know why. I'm not an expert in metadata.

I've searched the Adobe/XMP specs, the W3C/RDF spec and the comments in the code in xmpsdk. I haven't found anything to say that an empty li element is legal or illegal. For sure, it seems an odd thing to put into XMP. There are three very similar RDF containers. Bag, Seq, and Alt. For a 'Bag', you should think of a mathematical set of entities. A 'Seq' is an ordered set of entities. An 'Alt' is an unordered set from which you are expected to choose one. So, it's kind of odd to say "I want an empty element" in a 'Seq'. It's like saying "Waiter! I want a black hole in my soup!".

I've also looked at your file crash.xmp with the latest version of Adobe's XMPsdk (released July 2016).

553 rmills@rmillsmbp:~/gnu/xmpsdk/XMP-Toolkit-SDK-CC201607 $ samples/target/macintosh/intel_64/Release/ReadingXMP ~/Downloads/crash.xmp 
No smart handler available for /Users/rmills/Downloads/crash.xmp
Trying packet scanning.

/Users/rmills/Downloads/crash.xmp is opened successfully
dc:title in English = 
dc:title in French = 

XMP dumped to XMPDump.txt
554 rmills@rmillsmbp:~/gnu/xmpsdk/XMP-Toolkit-SDK-CC201607 $ cat XMPDump.txt

Dumping XMPMeta object ""  (0x0)

   darktable:  http://darktable.sf.net/  (0x80000000 : schema)
      darktable:multi_name  (0x600 : isOrdered isArray)
         [1] = "foo" 
         [2] = "bar" 
         [3] = "baz" 
555 rmills@rmillsmbp:~/gnu/xmpsdk/XMP-Toolkit-SDK-CC201607 $ 

I'd like to know more about why this is causing you to be unhappy. Why is DarkTable generating this XMP? Have you spoken to our friends at DT about this?

I've raised a topic on the Adobe XMPsdk Forum about this matter. https://forums.adobe.com/message/8968850#8968850

Would updating the XMPSDK help with this issue?

http://dev.exiv2.org/issues/941

Regrettably not. As you can see above, I built the latest XMPsdk and ran the sample application ReadingXMP on Tobias' code. The <rdf:li/> tag was silently ignored.

I think we'll have to get input from an XMP "language lawyer" at Adobe to say why this is ignored. Equally, we need to hear from DarkTable about the meaning of this code. I'm not sure the XMPsdk (and therefore libexiv2) can create that XMP. So, lots of strange things about this. I believe exiv2 is blameless in this situation.

Incidentally, I have not been able to reproduce the crash that is alleged with v0.25. However, Tobias has been explicit in asking us to focus on the empty element. So the crash is of no importance.

BTW, I downloaded and built the latest Adobe XMPsdk about a week ago. I have started looking at that in anticipation of integrating it into Exiv2 in v0.27.

The XMP files with the empty elements were created by libexiv2. In my sample I just deleted some unrelated things to make it more obvious what is going on. To explain why darktable relies on those empty elements I have to show a more complete example. I still removed some unrelated elements, but the important parts are there:

<?xml version="1.0" encoding="UTF-8"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
 <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
  <rdf:Description rdf:about="" 
    xmlns:darktable="http://darktable.sf.net/">
   <darktable:history_operation>
    <rdf:Seq>
     <rdf:li>sharpen</rdf:li>
     <rdf:li>flip</rdf:li>
     <rdf:li>basecurve</rdf:li>
     <rdf:li>tonecurve</rdf:li>
     <rdf:li>tonecurve</rdf:li>
     <rdf:li>tonecurve</rdf:li>
    </rdf:Seq>
   </darktable:history_operation>
   <darktable:history_params>
    <rdf:Seq>
     <rdf:li>000000400000003f0000003f</rdf:li>
     <rdf:li>ffffffff</rdf:li>
     <rdf:li>gz09eJxjYIAAM6vnNnqyn22E9n235b6aa3cy6rVdRaK9/Y970fYf95bbMzA0QPEoGEqADYnNhEUeAHH1EJo=</rdf:li>
     <rdf:li>gz06eJxjYICAcvFa669/NltdSOewY77w1NY+e5ndCreddgESofZT1qXYMzA0QPHAAK7ri21BeNbMmXYgDBQCucXe2NgYjOVbs8F4oNw5WN3HBsTsUIwMGJFoAHvBKzs=</rdf:li>
     <rdf:li>gz12eJxjYIAA6/BO2x0r620YGBrsIXjQAXsIHnUfKYAZCTMhYUaoPIgGALM+CPw=</rdf:li>
     <rdf:li>gz13eJxjYICAe7b29vrVpfYMDA1QPOiAPQSPuo8UwIyEmZAwI1QeRAMAvvwIfw==</rdf:li>
    </rdf:Seq>
   </darktable:history_params>
   <darktable:multi_name>
    <rdf:Seq>
     <rdf:li/>
     <rdf:li/>
     <rdf:li/>
     <rdf:li/>
     <rdf:li>1</rdf:li>
     <rdf:li>2</rdf:li>
    </rdf:Seq>
   </darktable:multi_name>
  </rdf:Description>
 </rdf:RDF>
</x:xmpmeta>

In darktable:history_operation there is an ordered list of image editing operations. In darktable:history_params there are the settings for those, and darktable:multi_name holds the instance names for those to make the multiple applications of the tonecurve distinguishable. If the empty elements are kept darktable can assign an empty name to the first 4 operations and "1" and "2" to the last two. When empty elements are skipped however, the non-empty names are applied to sharpen and flip which is wrong.

Of course that layout isn't ideal and some day™ the whole schema should be redesigned to have one set of settings per operation which includes everything, but as there are millions of those XMP files out in the wild it would be great if reading them back as intended was somehow possible.

Thanks for the clarification. I'll investigate how to create such a pattern of XMP with libexiv2. I hope somebody on the Adobe Forum provides some insight concerning this matter as this behaviour of XMPsdk seems quite intentional.

Here's a proposal about how to fix this. We can read/rewrite the XMP using expat to convert the empty tags from <rdf:li/> to <rdf:li>X</rdf:li>. The content of X is any string except the empty string. It could be a (space) , 0 or any string such as empty. So the following transform would occur:

   FROM                    ->   TO
   <darktable:multi_name>       <darktable:multi_name>                                               
    <rdf:Seq>                    <rdf:Seq>                 
     <rdf:li/>                    <rdf:li>empty</rdf:li>                                     
     <rdf:li/>                    <rdf:li>empty</rdf:li>                
     <rdf:li/>                    <rdf:li>empty</rdf:li>                
     <rdf:li/>                    <rdf:li>empty</rdf:li>                
     <rdf:li>1</rdf:li>           <rdf:li>1</rdf:li>                         
     <rdf:li>2</rdf:li>           <rdf:li>2</rdf:li>                         
    </rdf:Seq>                   </rdf:Seq>

Let's talk some more about this. For sure, I'm confident that we can achieve your goal to respect the millions of existing DT files with this troublesome XMP. We'll raise a Redmine Feature Request on libexiv2 when we have agreed on the appropriate remedy.

Thank you, that helped a lot. Using image->xmpPacket() and image->setXmpPacket() I can manually fix the XML data just fine. No need for any extra libexiv2 API I guess.

However, this leads me to a second question: In order to avoid problems like this in the future I would like to change the data layout to something like this:

<darktable:history>
  <rdf:Seq>
    <rdf:li>
      <darktable:modversion></darktable:modversion>
      <darktable:enabled></darktable:enabled>
      <darktable:operation></darktable:operation>
      <darktable:params></darktable:params>
      <darktable:blendop_params></darktable:blendop_params>
      <darktable:blendop_version></darktable:blendop_version>
      <darktable:multi_priority></darktable:multi_priority>
        <darktable:multi_name></darktable:multi_name>
    </rdf:li>
  <rdf:Seq>
</darktable:history

Following the examples from the API docs I am able to put everything into attributes of the <rdf:li> using code like this

  Exiv2::XmpProperties::registerNs("http://darktable.sf.net/", "darktable");

  Exiv2::XmpData xmpData;

  Exiv2::XmpTextValue tv("");

  tv.setXmpArrayType(Exiv2::XmpValue::xaSeq);
  xmpData.add(Exiv2::XmpKey("Xmp.darktable.history"), &tv); // Set the array type.
  tv.setXmpArrayType(Exiv2::XmpValue::xaNone);

  tv.read("flip");
  xmpData.add(Exiv2::XmpKey("Xmp.darktable.history[1]/darktable:operation"), &tv);

  tv.read("1");
  xmpData.add(Exiv2::XmpKey("Xmp.darktable.history[1]/darktable:priority"), &tv);

  tv.read("spot");
  xmpData.add(Exiv2::XmpKey("Xmp.darktable.history[2]/darktable:operation"), &tv);

  tv.read("2");
  xmpData.add(Exiv2::XmpKey("Xmp.darktable.history[2]/darktable:priority"), &tv);

Is there any way to get the format I want, i.e., having proper XML tags in the <rdf:li>?

Tobias

I'm a little lost. Can we divide your question into parts:

1 What XML do you want?
2 Is it legal XMP?
3 How do you get the exiv2 API to build it?

Example: DT.xmp:

617 rmills@rmillsmbp:~/gnu/exiv2/trunk $ xmllint --pretty 1 DT.xmp 
<?xml version="1.0"?>
<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
  <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
    <rdf:Description xmlns:darktable="http://darktable.sf.net/" rdf:about="">
      <darktable:multi_name>
        <rdf:Seq>
          <rdf:li>
            <rdf:Bag>
              <darktable:modversion>xxx</darktable:modversion>
              <darktable:enabled>yyy</darktable:enabled>
            </rdf:Bag>
          </rdf:li>
        </rdf:Seq>
      </darktable:multi_name>
    </rdf:Description>
  </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>
618 rmills@rmillsmbp:~/gnu/exiv2/trunk $ xmllint --pretty 1 DT.xmp  | wc
      19      30     659
619 rmills@rmillsmbp:~/gnu/exiv2/trunk $ exiv2 -pa DT.xmp 
Error: XMP Toolkit error 102: Named children only allowed for schemas and structs
Warning: Failed to decode XMP metadata.
620 rmills@rmillsmbp:~/gnu/exiv2/trunk $ 

However before we discuss the API, can you define the XMP/XML please.

Using exiv2 to have a look at the actual Xmp path to be used was a good hint, it seems that Xmp.darktable.history[1]/?darktable:mod version does the trick. However, the final XML is quite bloated so I guess I will just keep everything in attributes.

Like this?

629 rmills@rmillsmbp:~/gnu/exiv2/trunk $ cat DT.xmp 
<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
 <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
  <rdf:Description rdf:about="" xmlns:darktable="http://darktable.sf.net/" >
   <darktable:multi_name>
    <rdf:Seq>
     <rdf:li rdf:resource="1" darktable:modversion="abc" darktable:parameters="ABC"/>
     <rdf:li rdf:resource="2" darktable:modversion="xyz" darktable:parameters="XYZ"/>
    </rdf:Seq>
   </darktable:multi_name>
  </rdf:Description>
 </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>630 rmills@rmillsmbp:~/gnu/exiv2/trunk $ exiv2 -pa DT.xmp 
Xmp.darktable.multi_name                     XmpText     0  type="Seq" 
Xmp.darktable.multi_name[1]                  XmpText     1  1
Xmp.darktable.multi_name[1]/?darktable:modversion XmpText     3  abc
Xmp.darktable.multi_name[1]/?darktable:parameters XmpText     3  ABC
Xmp.darktable.multi_name[2]                  XmpText     1  2
Xmp.darktable.multi_name[2]/?darktable:modversion XmpText     3  xyz
Xmp.darktable.multi_name[2]/?darktable:parameters XmpText     3  XYZ
631 rmills@rmillsmbp:~/gnu/exiv2/trunk $ 

Yes, like that.

Good. I think this has a happy ending with no action required by Exiv2. Do you agree?

Indeed. Thank you for your help!

Sorry to bother you again. I didn't want to start a new topic for this follow up question.

I am now writing XMP files like this:

<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="XMP Core 4.4.0-Exiv2">
 <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
  <rdf:Description rdf:about="" 
    xmlns:darktable="http://darktable.sf.net/" 
   darktable:xmp_version="2">
   <darktable:history>
    <rdf:Seq>
     <rdf:li>
      <rdf:Description
       darktable:operation="sharpen" 
       darktable:enabled="1">
      <darktable:settings>
       <rdf:Seq>
        <rdf:li
         darktable:name="width" 
         darktable:type="int" 
         darktable:value="23"/>
        <rdf:li
         darktable:name="height" 
         darktable:type="int" 
         darktable:value="42"/>
       </rdf:Seq>
      </darktable:settings>
      </rdf:Description>
     </rdf:li>
     <rdf:li
      darktable:operation="flip" 
      darktable:enabled="1"/>
     <rdf:li
      darktable:operation="tonecurve" 
      darktable:enabled="1"/>
    </rdf:Seq>
   </darktable:history>
  </rdf:Description>
 </rdf:RDF>
</x:xmpmeta>

Using libexiv2 I can get an iterator to Xmp.darktable.history and iterate that, however since that isn't a simple array I get XmpText values only, containing everything in that subtree:

Xmp.darktable.history = type="Seq" 
Xmp.darktable.history[1] = type="Struct" 
Xmp.darktable.history[1]/darktable:operation = sharpen
Xmp.darktable.history[1]/darktable:enabled = 1
Xmp.darktable.history[1]/darktable:settings = type="Seq" 
Xmp.darktable.history[1]/darktable:settings[1] = type="Struct" 
Xmp.darktable.history[1]/darktable:settings[1]/darktable:name = width
Xmp.darktable.history[1]/darktable:settings[1]/darktable:type = int
Xmp.darktable.history[1]/darktable:settings[1]/darktable:value = 23
Xmp.darktable.history[1]/darktable:settings[2] = type="Struct" 
Xmp.darktable.history[1]/darktable:settings[2]/darktable:name = height
Xmp.darktable.history[1]/darktable:settings[2]/darktable:type = int
Xmp.darktable.history[1]/darktable:settings[2]/darktable:value = 42
Xmp.darktable.history[2] = type="Struct" 
Xmp.darktable.history[2]/darktable:operation = flip
Xmp.darktable.history[2]/darktable:enabled = 1
Xmp.darktable.history[3] = type="Struct" 
Xmp.darktable.history[3]/darktable:operation = tonecurve
Xmp.darktable.history[3]/darktable:enabled = 1

Is there any way to access that in a more structured way, without the type attributes and as arrays per level? Or am I required to do string matching against the keys and parse that myself?

You're not bothering me. I'm always happy to help you and all our friends at Dark Table (and GIMP, digiKam and all users of Exiv2 technology).

I'm not sure I've really understood your question, however I think it's involved with the rather awkward syntax for querying attributes. I attach the file Tobias.jpg which I made from test/data/Reagan.jpg and inserting your XMP with the v0.26 command $ exiv2 -iXX Tobias.jpg which reads Tobias.xmp and replaces the XMP.

Run sample/exiv2json.cpp on Tobias.jpg and the XMP will be converted to JSON which you can read with your favourite JSON parser.

579 rmills@rmillsmbp:~/gnu/exiv2/trunk $ bin/exiv2json -x Tobias.jpg  
{
    "Xmp": {
        "darktable": {
            "xmp_version": "2",
            "history": [
                {
                    "darktable": {
                        "operation": "sharpen",
                        "enabled": "1",
                        "settings": [
                            {
                                "darktable": {
                                    "name": "width",
                                    "type": "int",
                                    "value": "23" 
                                }
                            },
                            {
                                "darktable": {
                                    "name": "height",
                                    "type": "int",
                                    "value": "42" 
                                }
                            }
                        ]
                    }
                },
                {
                    "darktable": {
                        "operation": "flip",
                        "enabled": "1" 
                    }
                },
                {
                    "darktable": {
                        "operation": "tonecurve",
                        "enabled": "1" 
                    }
                }
            ]
        },
        "xmlns": {
            "darktable": "http:\/\/darktable.sf.net\/" 
        }
    }
}
580 rmills@rmillsmbp:~/gnu/exiv2/trunk $

Another approach is to extract the XMP with image->printStructure() or image->xmpPacket() and manipulate it with an XML library such as expat or Sax.

If you don't have exiv2json available, you can build it from source, or download it for your platform from our buildserver: http://exiv2.dyndns.org:8080/userContent/builds/Categorized/Latest/

I should have pointed out another couple of things:

1 samples/exiv2json.cpp is a deep recursive parser of the Exiv2/XMP Syntax. It rebuilds the data into a tree of JSON objects and arrays. The code is non-trivial.
2 you can use your favourite JSON parser to decode the JSON or you can use the api in Jzon.h/Jzon.cpp to navigate the tree.

//// I'm a little bored and have wondered "Is there a tool like XSLT or SQL for JSON?" Yes. jq (Json Query, I think).

706 rmills@rmillsmbp:~/gnu/exiv2/trunk $ jq ".Xmp.darktable.history|length" Tobias.json
3
707 rmills@rmillsmbp:~/gnu/exiv2/trunk $ jq ".Xmp.darktable.history[2]" Tobias.json
{
  "darktable": {
    "operation": "tonecurve",
    "enabled": "1" 
  }
}
708 rmills@rmillsmbp:~/gnu/exiv2/trunk $ for i in $(seq 0 1 $(( $(jq ".Xmp.darktable.history|length" Tobias.json) -1)));do echo operation: $(jq ".Xmp.darktable.history[$i].darktable.operation" Tobias.json); done
operation: "sharpen" 
operation: "flip" 
operation: "tonecurve" 
709 rmills@rmillsmbp:~/gnu/exiv2/trunk $ 

722 rmills@rmillsmbp:~/gnu/exiv2/trunk $ json=$(bin/exiv2json -x http://dev.exiv2.org/attachments/download/1048/Tobias.jpg)
723 rmills@rmillsmbp:~/gnu/exiv2/trunk $ for i in $(seq 0 1 $(( $(echo $json|jq ".Xmp.darktable.history|length") -1)));do echo operation: $(echo $json|jq ".Xmp.darktable.history[$i].darktable.operation"); done
operation: "sharpen" 
operation: "flip" 
operation: "tonecurve" 
724 rmills@rmillsmbp:~/gnu/exiv2/trunk $