Roy Fielding:

There are probably other rules that I am forgetting, but the above are the rules related to the hypertext constraint that are most often violated within so-called REST APIs. Please try to adhere to them or choose some other buzzword for your API.

A must read for all "REST API" designers.

How can we advance arts and humanities research through the development of shared technology services?

This is the question that the Bamboo Project seeks to answer. Last week, colleagues and folks I met at THATCamp were at the second Bamboo workshop, and I've been following their comments on Twitter. Count me as a skeptic. Keep in mind that I'm also a bit of a carpetbagger: a programmer with a math and physics background working in the humanities, not a humanities researcher. Unlike most digital humanities researchers, however, I have had experience working with broken legacy architectures.

The creation of computer network architectures is fraught with social and technical difficulties and architects are almost certain to get it wrong the first time (Gopher, for example). They might even get it wrong the second time around (WS-*, for example). The GIS community's standard service architecture (OGC W*S) is terribly wrong for the Web, and I fear a repeat in the digital humanities. It's not that the designers lack for domain expertise or goodwill, just that it's a perilous enterprise. The architectures we design today will likely be wrong for, or feel wrong for, or smell wrong to, the digital humanities programmers and integrators of the future. There's a lot of risk for Bamboo here, especially if the disdain for the "Wild West" Web evident in the proposal leads to needless reinvention.

More important than service infrastructure for the advancement of these fields is education and skills. Let's graduate students and future scholars with competency in computing, including:

• XML
• High level programming languages (Python, Ruby, etc)
• Web architecture, HTTP, HTML
• Formal logic
• RDF (and RDFa)

They'll be perfectly equipped to design exactly the services they need based on proven architectures like the Web.

Update (2008-10-20): via Tom Elliot, I realize I've rediscovered some of the skill set proposed by Lisa Spiro.

Update (2008-10-21): Hugh Cayless reports on the second workshop.

One of the highlights of this trip was the chance to meet fellow Python and GIS programmer, and blogger, Jáchym Čepický.

I'm not sure how many readers of Planet OSGeo understand Czech and know that Jáchym went through a scary episode in life this past summer. He gave me a sobering recap and then we had some less sobering toasts to his good health at Pivovarský klub. In a city served almost exclusively by single-beer pubs, Pivovarský klub is an oasis for the beer enthusiast. I had several beers from the tap that I never saw anywhere else -- beers far more fresh and flavorful than the ubiquitous Gambrinus.

I've just seen mention of this via Twitter, but haven't discussed with or seen any statement from the developers, on advice of lawyers, no doubt: Reuters sues GMU for no real reason other than Zotero's competition with, and besting of EndNote. Is this the kind of company you want to support, Steve? One that seems ready to sue its users for trying to free their data from its proprietary format?

More from Bruce D'Arcus, and also Hugh Cayless.

Thanks to some hand holding from Chris and Josh, Pleiades now has Spherical Mercator maps using the Google physical geography layer as a stand-in for our ideal ancient world base map. See http://pleiades.stoa.org/places/639166. Coordinate transformation is provided by the code I blogged yesterday.

Pyproj is Jeffrey Whitaker's Python interface to PROJ.4. I'm no longer interested in other Python projection packages. Its most interesting feature is interoperability with packages (such as Shapely) that use the Numpy array interface. It depends on Cython, which makes it a bit tricky to include in a buildout: you must install Cython into your buildout's python, not as an egg, and make the pyproj egg only after this step is finished. Like this:

.. sourcecode:: none

  [buildout]
  parts =
    cython-src
    cython-install
    pyproj

  [cython-src]
  recipe = hexagonit.recipe.download
  url = http://cython.org/Cython-0.9.8.1.1.tar.gz

  [cython-install]
  recipe = iw.recipe.cmd
  on_install = true
  cmds =
    cd ${buildout:directory}/parts/cython-src/Cython-0.9.8.1.1
    ${python:executable} setup.py install

  [pyproj]
  recipe = zc.recipe.egg:eggs
  index = http://atlantides.org/eggcarton/index
  eggs = pyproj

You might be able to pull pyproj off PyPI, but here I am using my own index. Once built, you can try it out using zopepy:

>>> from pyproj import Proj
>>> defn_900913 = """
... +proj=merc +a=6378137 +b=6378137
... +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0 +k=1.0
... +units=m +nadgrids=@null +no_defs
... """
>>> proj_900913 = Proj(defn)
>>> lonlat = (25.0, 25.0)
>>> proj_900913(*lonlat)
(2782987.269831839, 2875744.6243522423)

I like that projection definitions can be split across lines of the screen.

I've just finished writing a module that supports projection of objects that provide the geometry part of the Lab's geo interface: proj.py. Maybe you'll find it useful. What it does is take a geojson geometry (or Shapely geometry) and return a new projected geojson geometry object:

>>> from pleiades.openlayers.proj import Transform, PROJ_900913
>>> transform = Transform(PROJ_900913)

>>> # Point, forward
>>> from geojson import Point
>>> point = Point(coordinates=(25.0, 25.0))
>>> fwd = transform(point)
>>> fwd
Point(coordinates=(2782987.269831839, 2875744.6243522423))

>>> # Point, inverse
>>> inv = transform(fwd, inverse=True)
>>> inv
Point(coordinates=(24.999999999999996, 24.999999999999996))

>>> # Line, forward
>>> from geojson import LineString
>>> line = LineString(coordinates=((25.0, 25.0), (30.0, 30.0)))
>>> fwd = transform(line)
>>> fwd
LineString(coordinates=((2782987.269831839, 2875744.6243522423), ...))

>>> # Line, inverse
>>> inv = transform(fwd, inverse=True)
>>> inv
LineString(coordinates=((24.999999999999996, 24.999999999999996), ...))

There's also an object hook if you'd like the transform to yield instances of your own classes.

KML lacks two important elements. It has neither 1) a simple and universal element equivalent to Dublin Core description (such as {http://purl.org/dc/elements/1.1/}description), nor 2) an excellent element for encapsulating or linking to rich content accompanying placemarks. Instead of the first, we have kml:Snippet. Snippet! It's somewhere in between a title and a description and that's no place to be. Instead of the second, we have kml:description, which supports an under-specified subset of HTML and other media including video.

The kml:description element is not about metadata anymore, it's about adding rich content -- charts, photos, video -- to popup windows of a geographic browsers. Under-specification of the content of kml:description is a problem. Which HTML elements are valid? Why shouldn't all be valid? The choice only between escaping HTML and using CDATA blobs is another problem. XHTML in a description wouldn't break the KML document; why shouldn't we be able to use XHTML without CDATA? Lack of support for dynamic content is yet another problem. Why can't we specify rich content by a URL, to be fetched only as needed and refreshed only if modified? Limitations of KML's description element and its ties to the original implementation hold us back. We need something better.

The Atom community wrestled with the same issues and came up with the atom:content element. An atom:content element may contain text, HTML, XHTML, XML, encoded binary media, or provide a URL to another document or media resource. When prompted, Atom processors (such as a news reader) display this content to a user. All HTML or XHTML elements are valid. CDATA blobs are unnecessary and not allowed. Atom's content element comes with clear processing instructions: HTML must be escaped, and should be valid within a <div>. The immediate child element of XHTML content must be a <div>. While guided by implementations, none of this depends on unspecified behavior of a particular Atom processor.

The "src" attribute of the atom:content element provides developers with more options. It allows you to decouple semi-static data from dynamic data. The locations and identifiers of physical instruments deployed in the field (river gauges, ASOS, you name it) can be decoupled from their nearly real-time observations and measurements. Serve a representation of the deployed instruments as a static document updated only as are the physical objects. From each instrument entry, link to a specialized web resource that produces tables or graphs on demand.

Atom processors can also take advantage of the "src" attribute to improve performance perceived by a user. Fetch the feed, show users titles and summaries first, get rich content as the user asks for it -- or get the rich content in the background, or during lulls in user activity. Processors should also be able to exercise conditional GET to avoid reloading unchanged rich content.

The atom:content element appears to support all of the requirements of, and overcome all of the faults of, kml:description. It fully embraces and accommodates rich content and provides new design options via links to dynamic or massive content. KML is already open to Atom elements (version 2.2 already includes atom:author, atom:link, and atom:name) and so atom:content is a more natural replacement for kml:description than RSS 1.0's content element or RSS 2.0's enclosure element. I'm not in a position to offer any advice to Google, but -- considering its investment in Atom, to act on the commonalities between KML and Atom seems useful and even profitable.

I've heard from KML folks that nothing pushes the spec forward more than working code. I'm not interested in writing a new geographic browser, but I can start serving KML enhanced with atom:content (which meets needs of a user that just cropped up on Monday), support it in my own KML processing packages, and perhaps even exercise my withering C++ skills on a libkml patch. After that? Let's swap in atom:summary for kml:Snippet. Snippet!

Martin Daly:

The OGC Geospatial Search Summit was held here at the Atlanta TC yesterday. It was an invitation only, behind closed doors event - grumble, grumble, moan, moan - so I don’t know any details about what went on.

Inside the "Open" Geospatial Consortium there are yet more inner circles?

I don't follow the django-developers list so I might have missed Kevin Teague's great overview and analysis of Python packaging, dependencies, and builds (via Simon Willison). I've referred to setuptools, easy_install, and zc.buildout a number times, but never explained them quite as thoroughly or as well.

Setuptools is the not-quite-standard replacement for Python's distutils module. It has a big set of features, the one of which that helps me the most is versioning of dependencies. I can specify that zgeo.plone.kml 0.2 requires version 0.1 or higher of zgeo.plone.geographer, which in turn requires version 0.3 or higher of zgeo.geographer, and easy_install of the first one of those packages gets the proper dependencies in a cascade. I'm finding this to be better for my sanity than use of the svn:externals property to pull in tagged revisions as we used to do.

Despite what Teague says (update: I misinterpreted, see clarification in comments), I'm finding zc.buildout to be a big help with managing the C/C++ components in Pleiades. Because the GEOS project is -- how shall I say it? -- cautious about bug fix releases, and version 3.0 absent from Linux distros, I'm using Kai Lautaportti's hexagon.recipe.cmmi buildout recipe to download the GEOS 3.0.0 tarball, patch it, compile, make, and make install into an isolated environment for our web application with no more configuration than this:

[libgeos]
recipe = hexagonit.recipe.cmmi
url = http://geos.refractions.net/downloads/geos-3.0.0.tar.bz2
patch-options = -p1
patches =
  ${buildout:directory}/patches/geos-3.0.0-conditional.patch

Works like a charm.

To avoid polluting the Python package index with software of no use outside Pleiades, I've made our own specialized index. Easy_install of pleiades.policy (my obvious cargo cult of Martin Aspeli's example) pulls in pleiades.kml, and then the appropriate versions of zgeo.plone.kml and zgeo.kml. I'm happy to explain how to make such an index to anybody who's interested.

Do read Teague's post to the django-developers list if you're new to the Python software ecosystem, and if you develop any curiosity about how Zope works, read his excellent article about its component architecture.