Add rational for the functioning and opening of the amazon store.

This commit is contained in:
John Schember 2011-02-27 21:08:44 -05:00
parent 84211bae54
commit 4038afdec2

View File

@ -22,6 +22,89 @@ class AmazonKindleStore(StorePlugin):
description = _('Kindle books from Amazon')
def open(self, gui, parent=None, detail_item=None):
'''
Amazon comes with a number of difficulties.
QWebView has major issues with Amazon.com. The largest of
issues is it simply doesn't work on a number of pages.
When connecting to a number parts of Amazon.com (Kindle library
for instance) QNetworkAccessManager fails to connect with a
NetworkError of 399 - ProtocolFailure. The strange thing is,
when I check QNetworkRequest.HttpStatusCodeAttribute when the
399 error is returned the status code is 200 (Ok). However, once
the QNetworkAccessManager decides there was a NetworkError it
does not download the page from Amazon. So I can't even set the
HTML in the QWebView myself.
There is http://bugreports.qt.nokia.com/browse/QTWEBKIT-259 an
open bug about the issue but it is not correct. We can set the
useragent (Arora does) to something else and the above issue
will persist. This http://developer.qt.nokia.com/forums/viewthread/793
gives a bit more information about the issue but as of now (27/Feb/2011)
there is no solution or work around.
We cannot change the The linkDelegationPolicy to allow us to avoid
QNetworkAccessManager because it only works links. Forms aren't
included so the same issue persists on any part of the site (login)
that use a form to load a new page.
Using an aStore was evaluated but I've decided against using it.
There are three major issues with an aStore. Because checkout is
handled by sending the user to Amazon we can't put it in a QWebView.
If we're sending the user to Amazon sending them there directly is
nicer. Also, we cannot put the aStore in a QWebView and let it open the
redirection the users default browser because the cookies with the
shopping cart won't transfer.
Another issue with the aStore is how it handles the referral. It only
counts the referral for the items in the shopping card / the item
that directed the user to Amazon. Kindle books do not use the shopping
cart and send the user directly to Amazon for the purchase. In this
instance we would only get referral credit for the one book that the
aStore directs to Amazon that the user buys. Any other purchases we
won't get credit for.
The last issue with the aStore is performance. Even though it's an
Amazon site it's alow. So much slower than Amazon.com that it makes
me not want to browse books using it. The look and feel are lesser
issues. So is the fact that it almost seems like the purchase is
with calibre. This can cause some support issues because we can't
do much for issues with Amazon.com purchase hiccups.
Another option that was evaluated was the Product Advertising API.
The reasons against this are complexity. It would take a lot of work
to basically re-create Amazon.com within calibre. The Product
Advertising API is also designed with being run on a server not
in an app. The signing keys would have to be made avaliable to ever
calibre user which means bad things could be done with our account.
The Product Advertising API also assumes the same browser for easy
shopping cart transfer to Amazon. With QWebView not working and there
not being an easy way to transfer cookies between a QWebView and the
users default browser this won't work well.
We could create our own website on the calibre server and create an
Amazon Product Advertising API store. However, this goes back to the
complexity argument. Why spend the time recreating Amazon.com
The final and largest issue against using the Product Advertising API
is the Efficiency Guidelines:
"Each account used to access the Product Advertising API will be allowed
an initial usage limit of 2,000 requests per hour. Each account will
receive an additional 500 requests per hour (up to a maximum of 25,000
requests per hour) for every $1 of shipped item revenue driven per hour
in a trailing 30-day period. Usage thresholds are recalculated daily based
on revenue performance."
With over two million users a limit of 2,000 request per hour could
render our store unusable for no other reason than Amazon rate
limiting our traffic.
The best (I use the term lightly here) solution is to open Amazon.com
in the users default browser and set the affiliate id as part of the url.
'''
from calibre.gui2 import open_url
aff_id = {'tag': 'josbl0e-cpb-20'}
store_link = 'http://www.amazon.com/Kindle-eBooks/b/?ie=UTF&node=1286228011&ref_=%(tag)s&ref=%(tag)s&tag=%(tag)s&linkCode=ur2&camp=1789&creative=390957' % aff_id