My earlier blog post on caching requires some follow up. In fact, the utility of httplib2 or any other HTTP cache hinges on whether or not the server hosting the wanted resources provides validating (ETag or Last-Modified) or freshness (Expires or Cache-Control) headers. In other words: the organization running the server has to want to help you save bandwidth. The original blog post says "That function caches the result of WMS requests for layer legends in a dedicated directory, assuming that the images are not changing over time."

WMS, eh? Do Web Map Services work with web caches? Not by design. How about in practice? I've plucked a WMS legend request from the USGS Framework WMS Viewer [1]:

If I read the Linfinti blog post right, this is the kind of image they'd like to avoid fetching unnecessarily. If you look closely at a HTTP request for the legend image (passing the -v option to curl, with [1] standing in for the request in the code below), you see:

krusty-2:~ seang$ curl -v [1] > /tmp/legend.png
* About to connect() to frameworkwfs.usgs.gov port 80 (#0)
*   Trying 76.10.128.162... connected
* Connected to frameworkwfs.usgs.gov (76.10.128.162) port 80 (#0)
> GET /framework/wms/wms.cgi?SERVICE=WMS&VERSION=1.3.2... HTTP/1.1
> User-Agent: curl/7.19.5 (i386-apple-darwin9.8.0) libcurl/7.19.5 OpenSSL/0.9.8k zlib/1.2.3
> Host: frameworkwfs.usgs.gov
> Accept: */*
>
< HTTP/1.1 200 OK
< Date: Thu, 18 Feb 2010 21:26:03 GMT
< Server: Apache/2.0.46 (Red Hat)
< Connection: close
< Transfer-Encoding: chunked
< Content-Type: image/png

No validator headers. No freshness headers. Neither httplib2 nor a HTTP caching proxy is going to be useful here. You're on your own. Hurrah for standards. Let a thousand ad-hoc caching schemes bloom.

Update (2010-02-28): 1.2b3 is out. URLs are changed below.

Update (2010-02-19): 1.2b2 is out. URLs are changed below.

This is the first 1.2 release uploaded to PyPI: http://pypi.python.org/pypi/Shapely/. I hope the new README introduces the package more clearly. The manual lags: it will be the last thing before a 1.2 final.

A consequence for downstream developers is that a package that depends on Shapely in the setuptools sense without specifying a specific version will now begin to pull in 1.2b* releases instead of 1.0.14. 1.2b1 seems to be working with GEOS 3.0 and we didn't intentionally break backwards compatibility, but you should be cautious about this. If you're depending on GEOS 2.2.3 and haven't already pinned your package to Shapely==1.0.14 (updating as you go), you should.

To install and try it out (in a virtualenv):

$ pip install http://gispython.org/dist/Shapely-1.2b3.tar.gz

or

$ easy_install http://gispython.org/dist/Shapely-1.2b3.tar.gz

I've fooled around with Jython, but this, via @erilem, is the first I've seen of it in real live GIS software:

>>> from geoserver import Catalog
>>> cat = Catalog('topp')
>>> print cat.stores()
[]
>>> st = cat['states_shapefile']
>>> print st.layers()
[]
>>> l = st['states']
>>> print l.count()
49
print l.bounds()
...

A little bit of its Java-ness leaks through, but at least one is spared the getters and setters. Cool, but then what's this smell at the end?

The following is the wish list based on feedback from teh community:

• add a new output format

• add a transaction listener

• add a dispatcher callback

• add a WPS process

• add a datastore

• a restful endpoint

Restful endpoint? Oh, for crying out loud. Head. Desk.

Fuzzy, the blogger with the not-Flash banner animation, writes:

Every time I hear about devs (some of my co-workers included) choosing the Flex API for AGS I have to bite my lip. With the life span of web mapping sites they’re probably more than safe going with Flash – it’s slide will likely be long and slow. But I think that slide is becoming inevitable.

Last week I commented that one upside of the Apple-Adobe Flash brouhaha was that it reduces all the necessarily complicated, technical, wonky arguments against Flash (which clients know, somehow, that they "need") to: "it won't work on your iPad". But I'm resisting because I'm almost certain that Apple will eventually win this game of chicken, extract whatever concessions or guarantees or fees or pounds of flesh from Adobe, and relent. I'd like to be wrong.

William Vambenepe points out some familiar bugs:

• The mailing lists of DMTF working groups are confidential. Even a DMTF member cannot see the message archive of a group unless he/she is a member of that specific group. The general public cannot see anything at all. And unless I missed it on the site, they cannot even know what DMTF working groups exist. It makes you wonder whether Dick Cheney decided to call his social club of energy company executives a “Task Force” because he was inspired by the secrecy of the DMTF (“Distributed Management Task Force”). Even when the work is finished and the standard published, the DMTF won’t release the mailing list archive, even though these discussions can be a great reference for people who later use the specification.

• Working documents are also confidential. Working groups can decide to publish some intermediate work, but this needs to be an explicit decision of the group, then approved by its parent group, and in practice it happens rarely (mileage varies depending on the groups).

• Even when a document is published, the process to provide feedback from the outside seems designed to thwart any attempt. Or at least that’s what it does in practice. Having blogged a fair amount on technical details of two DMTF standards (CMDBf and WS-Management) I often get questions and comments about these specifications from readers. I encourage them to bring their comments to the group and point them to the official feedback page. Not once have I, as a working group participant, seen the comments come out on the other end of the process.

GIS industry standards are made in just such a non-transparent members-only environment. I used to subscribe to the OGC's "mass market" (private archive, but open to subscription) list and tried to engage in some discussion there, but soon realized that although messages from the principals were being cross-posted there, they weren't subscribed themselves and didn't see any responses. I also tried to submit comments to the formal channel and found it to be broken (there's a year long gap in the archives: it could have broken for that length of time without anybody noticing). Now that it's fixed, you can see the public comment process doesn't get much use.

Despite this, the OGC's standards enjoy almost absolute buy-in from non-member GIS specialists, particularly those from the open source community who need something – anything – to counter de-facto standardization on ESRI products.

An ESRI user shares his software wish list:

3. Expose numpy in the geoprocessor: The geoprocessor as is right now (9.3.1) uses the excellent numpy module to perform matrix algebra (think of raster manipulation). Yet, when one wishes to run numpy commands, one needs to manually read raster files with GDAL, import them as numpy arrays (default), perform operations, and translate back to raster. ESRI must have modules for dealing with this, and we want them. Why would ESRI want to do this? Right now, raster manipulation through Python is done outside the geoprocessor. Most people turn to open source tools to manipulate data, which leads to less and less users relying on ESRI for this. Why pay when free software will do it? The capability is there, and we need to access it too.

Fewer and fewer users relying on ESRI for this? You say that like it's a bad thing.

I just stumbled onto this post at ReadWriteWeb:

Google Buzz data can be syndicated out to other services using the standard data formats called Atom, Activity Streams, MediaRSS and PubSubHubbub. That couldn't be more different from Facebook. Google has taken open data standards to battle against a marketplace of competitors that are closed and proprietary to varying degrees. This is a very big deal.

Maybe it is a big deal, and not for the best: Google Buzz the product is falling flat with my sources, who find it too raw, too unpredictable, too much. It takes some of the shine off the underlying architecture.

Here's another installment in my series (iterators, revenge of the iterators, features) that considers different Python GIS APIs and environments: plotting geometries with Matplotlib. Matplotlib has a number of functions for plotting and graphing, and they typically take as their first arguments sequences of x and y coordinate values. These sequences are immediately adapted to Numpy arrays. In some sense what we're actually considering here is how well different Python APIs integrate with Numpy.

In every code snippet below we import the Matplotlib pylab module and then obtained a native geometry object geom or feature object in some way (see previous post for details).

ESRI:

x = []
y = []
pnt = geom.Next()
while pnt is not None:
    x.append(pnt.x)
    y.append(pnt.y)
    pnt = geom.Next()
pylab.plot(x, y)

Wrapping the geometry up in a point or vertex iterator would help tidy this up.

FME:

x, y = zip(*feature.getCoordinates())
pylab.plot(x, y)

No actual geometry object in FME, but this isn't bad at all in combination with zip. Remember that zip(*x) unzips the sequence of items x. Within a function call, the * operator explodes a sequence.

OGR:

x, y = zip(*((geom.GetX(i), geom.GetY(i)) for i in range(geom.GetGeometryRef(0).GetPointCount()))
pylab.plot(x, y)

Note that using zip gets this done in one pass over the geometry instead of two as in the example linked through OGR above. I'm telling you: idiommatic Python wins.

QGIS:

x, y = zip(*((p.x, p.y) for p in geom.asPolyline()))
pylab.plot(x, y)

geojson:

x, y = zip(*geom.coordinates)
pylab.plot(x, y)

Easier and easier.

Shapely (1.2a6):

x, y = geom.xy
pylab.plot(x, y)

Easiest yet.

Where possible, I've used a Python idiom to compress the code to two lines for each API example. There are a lot of calls in the OGR and QGIS samples and efficiency and clarity are at odds. Since a list is being built no matter what, there's no harm in doing it with more lines of code for those APIs. With Shapely, I've deliberately made it hard to do it in more than two lines. In fact, it's easier to do it in one:

pylab.plot(*geom.xy)

Although I'm getting a little weary of "social media" products I'm very interested in the technology and architecture behind Google's Buzz. Once again we see how differently syndication and synchronization are conceived and engineered by the GIS Enterprise and the Google. Federated Geo-Synchronization on the one hand (developed in a clean room behind a license wall) and on the other: HTTP, Atom, Web Linking, and PubSubHubbub.

Overly disruptive to other APIs? Maybe. Certainly it's a reminder that the web is one API that really counts and that makes me a happy boy.

One thing that Shapely has lacked is one or two dirt simple example programs to keep the API real and help explain its use. I did something about this over the past couple of nights: 1.2a6 includes two easy to understand, easy to run scripts. I hope users profit from them. Myself, I found that they demanded a new and improved API feature. I'll explain.

First, here's an example of using Shapely to construct patches by growing buffer regions out from a set of points and dissolving those regions together as they intersect, and plotting the results with Matplotlib. This is run-of-the-mill GIS stuff, yes, but done in style.

A plate of blue-speckled brains splattered on the floor, or is it just me?

The interesting part of the complete, amply-documented dissolve.py script is here:

import pylab
from shapely.ops import cascaded_union

patches = cascaded_union(spots)

pylab.figure(num=None, figsize=(4, 4), dpi=180)

for patch in patches.geoms:
    x, y = patch.exterior.xy
    pylab.fill(x, y, color='#cccccc', aa=True)
    pylab.plot(x, y, color='#666666', aa=True, lw=1.0)
    for hole in patch.interiors:
        x, y = hole.xy
        pylab.fill(x, y, color='#ffffff', aa=True)
        pylab.plot(x, y, color='#999999', aa=True, lw=1.0)

pylab.text(-25, 25,
    "Patches: %d, total area: %.2f" % (len(patches.geoms), patches.area))

pylab.savefig('dissolve.png')

The xy property is completely new in 1.2a6, inspired by how awkwardly I had to slice and dice coordinates when writing this example against 1.2a5. It provides two Python arrays that are immediately usable with Numpy or Matplotlib. Speaking of Matplotlib: I'd love to know how to fill a patch but not its holes (you'll notice that I'm faking the emptiness of the holes in this example).

What would would you have to go through to pyplot ArcGIS scripting results?

Shapely doesn't just make grey matter go splat, it can also toss brains in the air and pierce them with lasers:

Or make a fair facsimile thereof. What's really going on in intersect.py is an analysis of a HTML5 geolocation (latitude, longitude, heading, and speed) trajectory's intersection with a cluster of patches. The intercepted patches are plotted in red and the intersecting segments of the trajectory itself are also plotted in red. Finally, scalar properties of different geometries are used in a text label. The example vector intercepts 2 of the 7 patches along 5 segments with a total length (to one decimal place) of 26.1:

import pylab
from shapely.geometry import LineString

# Represent the following geolocation parameters
#
# initial position: -25, -25
# heading: 45.0
# speed: 50*sqrt(2)
#
# as a line
vector = LineString(((-25.0, -25.0), (25.0, 25.0)))

# Find intercepted and missed patches. List the former so we can count them
intercepts = [patch for patch in patches.geoms if vector.intersects(patch)]
misses = (patch for patch in patches.geoms if not vector.intersects(patch))

pylab.figure(num=None, figsize=(4, 4), dpi=180)

for spot in misses:
    x, y = spot.exterior.xy
    pylab.fill(x, y, color='#cccccc', aa=True)
    pylab.plot(x, y, color='#999999', aa=True, lw=1.0)
    for hole in spot.interiors:
        x, y = hole.xy
        pylab.fill(x, y, color='#ffffff', aa=True)
        pylab.plot(x, y, color='#999999', aa=True, lw=1.0)

for spot in intercepts:
    x, y = spot.exterior.xy
    pylab.fill(x, y, color='red', alpha=0.25, aa=True)
    pylab.plot(x, y, color='red', alpha=0.5, aa=True, lw=1.0)
    for hole in spot.interiors:
        x, y = hole.xy
        pylab.fill(x, y, color='#ffffff', aa=True)
        pylab.plot(x, y, color='red', alpha=0.5, aa=True, lw=1.0)

pylab.arrow(-25, -25, 50, 50, color='#999999', aa=True,
    head_width=1.0, head_length=1.0)

intersection = vector.intersection(patches)
for segment in intersection.geoms:
    x, y = segment.xy
    pylab.plot(x, y, color='red', aa=True, lw=1.5)

pylab.text(-28, 25,
    "Patches: %d/%d (%d), total length: %.1f" \
     % (len(intercepts), len(patches.geoms),
        len(intersection.geoms), intersection.length))

pylab.savefig('intersect.png')

Install GEOS 3.2.0 (Windows users can get it from a PostGIS 1.5 installer, but will have to copy the DLLs to a location one can glean only from looking at shapely/geos.py. YMMV until we have Shapely 1.2 installers) then grab the new distribution with easy_install or pip (as well as Numpy and Matplotlib) and give them a try:

$ python /usr/local/bin/dissolve.py
$ python /usr/local/bin/intersect.py

I think this is pretty much the last 1.2 alpha.