Add label/value option under format

[justgrimes]

It would be nice to add label and value under format. This would be useful if certain data files are zipped. If csv files are zipped then technically the format is zip i.e. "application/zip". With a label/value option under format you could identify and display the actual format of the contents of the file.

For example:

"distribution": [ {
    "accessURL": "http://www.foo.gov/csv.zip",
    "format": [{
        "label": "csv",
        "value": "application/zip"
        }]
    },{
    "accessURL": "http://www.foo.gov/shapefiles.zip",
    "format": [{
        "label": "shapefiles",
        "value": "application/zip"
        }]
    }

[mhogeweg]

I propose continuing discussion aruond this as part of #291.

[philipashlock]

I think that label would be misleading actually. It seems like the ideal scenario would be to make it clear that it's a zipped csv or a zipped shapefile. CKAN actually supports this natively with an additional field, but I don't know what convention would make sense in the context of POD/DCAT other than possibly introducing a new field.

For reference, in CKAN it's like this:

mimetype: standard mimetype (e.g. for zipped csv would be application/zip)
mimetype-inner: mimetype of innermost object (so for example would be text/csv)

@mhogeweg While there are parallels to the discussion about APIs and the different layers of media types there, I think this one use case for downloadable zipped files is common enough that it might make sense to address here separately.

[konklone]

I had assumed that posting zip files as data was a bug, not something to be resolved via mimetype.

I couldn't find any discussion of the matter here on POD about posting links to ZIP files instead of the actual data. It seems like it falls into the same category as including a link to a landing page (HTML) where data is linked to somewhere. Under either case, you can no longer automatically download and index the data itself - you have to make some sort of intelligent (or crude) guess at what to do with the HTML or ZIP file you've been given.

[mhogeweg]

The DOI data.json file is uncompressed about 110MB, while zipped up it is about 14% of that (15MB). zipping files will be a common practice for quite some time and it would be good to have a mechanism to describe the file and what's inside (which could be more than one file of different types).

I suggested combining it with the other discussion as I see this very similar to the distinction between a web service and the response you get back (a SOAP service returning a KMZ which is zipped KML file(s), or any other sample of service that's out there). something could return the results of a study as a pdf and one or more csv, kml, xls, shp files that all are zipped up in a convenient package.

[waldoj]

FWIW, the sitemap.xml spec allows for gzipping the file, which is simply indicated by renaming the file to sitemap.xml.gz. How the client knows to find it there without having to check two locations every time, I don't have the faintest idea.

[smrgeoinfo]

so one distribution could be something like and open data package (http://dataprotocols.org/data-packages/). A MIME type for this might be sufficient, but if it were possible to add some properties on the link to help the user know a-priori what they're going to get inside, it would be a good thing.

[konklone]

If we're just talking about zipping a large file, HTTP has compression negotiation built in to the transport layer.

So you can get the benefits of compression without needing to involve that optimization step in the data catalog at all, and without having to list or guess at multiple URLs.

[justgrimes]

@konklone There are a couple of reasons why zip files might be used:

• To collocate related data files where the content is split across multiple files (e.g. relational database extract; each table a csv)
• To collocate data files where the data file format involves more than one file (e.g. "shapefiles" - shp, dbx, prj, dbf)
• Compression to reduce file size
• Compression as a form of data integrity (ensure file isn't corrupted)
• Existing policies and practices (that supersede data.gov)

@philipashlock - A quick scan of data.gov shows that a number of datasets don’t resolve to their expected format. It seems that some people assumed format to be the contents and not the linked data file. Moreover, the use of zip files is pretty pervasive. They are the most popular format on data.gov (32% of all listed formats). Displaying this format does little to inform the user; if anything it obfuscates holdings. It’s particularly bad for data sets that have multiple formats, use distribution, and more than one file resolve to zip files. At best the user would have to mouse over and hope the agency incorporated content file type in the actual naming of the file.

I'm just wondering what might be the most helpful to display to the user. Not sure what the optimal solution is: add label/format; add mimetype-inner; better educate agencies as to what the format field should contain & recommend agencies don't use zip files.

[waldoj]

@konklone, I think that servers will only send gzipped or deflated content if the client indicates that it can receive those. cURL, if I understand its defaults properly, requires a --compress flag to send gzip,deflate as the Accept-Encoding value. This means that a human requesting data with a proper browser would get the automatically compressed version, but any cURL-based client would have to transfer the uncompressed data unless the implementor knew to specify --compress.

[konklone]

The basic principle that I've assumed is at work here, is that someone (be it data.gov, or anyone) should be able to use an agency's data catalog to download each dataset and know, programmatically, how to point the contained data at a program that knows how to handle that mime-type.

At the very least, I'd assume the mime-type of the dataset in the catalog would be that dataset's semantic mime-type (what's inside the zip file). Zip is just an implementation detail. In other words, instead of making a new field, mimetype-inner, for the contents, make a new field or fields that describe whatever implementation details the publisher has used that's in the way of accessing the real data.