1. What’s New?

To improve PEP-8 compliance a number of name changes are being made to methods and class attributes with each release. There is a module, pyslet.pep8, which contains a compatibility class for remapping missing class attribute names to their new forms and generating deprecation warnings, run your code with “python -Wd” to force these warnings to appear. As Pyslet makes the transition to Python 3 some of the old names may go away completely. The warning messages explain any changes you need to make. Although backwards compatible, using the new names is slightly faster as they don’t go through the extra deprecation wrapper.

It is still possible that some previously documented names could now fail (module level functions, function arguments, etc.) but I’ve tried to include wrappers or aliases so please raise an issue on Github if you discover a bug caused by the renaming. I’ll restore any missing old-style names to improve backwards compatibility on request.

1.1. Version Numbering

Pyslet version numbers use the check-in date as their last component so you can always tell if one build is newer than another. At the moment there is only one actively maintained branch of the code: version ‘0”. Changes are reported against the versions released to PyPi. The main version number increases with each PyPi release. Starting with version 0.7 changes are documented in order of build date to make it easier to see what is changing in the master branch on GitHub.

Not sure which version you are using? Try:

from pyslet.info import version
print version

1.2. Version 0.7.20170805

1.2.1. Summary of new features

Pyslet now supports Python 3, all tests are passing in Python 3.

Travis now builds and tests Python 2.7 and Python 3.5, I’ve dropped 2.6 from the continuous integration testing because the latest Ubuntu images have dropped Python2.6 but you can still run tox on your own environments as it includes 2.6 in tox.ini.

Various bug fixes in OData and HTTP code.

Warning: for future compatibility with Python 3 you should ensure that you use the bytes type (and the ‘b’ prefix on any string constants) when initialising OData entity properties of type Edm.Binary. Failure to do so will raise an error in Python 3.

1.2.2. Tracked issues

The following issues are resolved (or substantially resolved) in this release.

#3 PEP-8 Compliance

The pep8-regression.py script now checks all source files using flake8; all reported errors have been resolved

Added a new metaclass-based solution to enable method renaming while maintaining support for derived classes that override using the old names. Crazy I know, but it works.

#12 Bug in odata2/sqlds.py

Bug when using numeric or named parameters in DB API. Added support for pyformat in DB APIs as part of enabling support for PyMySQL.

#23 Framework for WSGI-based LTI Applications (beta quality)

Re-engineered Session support in the wsgi module to reduce database load, replacing the Session table completely with signed cookies. If you have used the wsgi.SessionApp class directly this will be a breaking change but these classes will remain experimental until this item is closed out. The database schema required to support LTI has changed slightly as a result.

Changed from Django templates to use Jinja2 (this requires almost no changes to the actual sample code templates and makes the intention of the samples much clearer). Thanks to Christopher Lee for recommending this change.

Possible breaking change to wsgi module to refactor authority setting to “canonical_root”, modified WSGIContext object to accept an optional canonical_root argument and removed the optional authority argument from get_app_root and get_url. The authority setting was previously a misnomer and the wsgi sammples were not working properly with localhost.

Changed wsgi module to use the OSFilePath wrapper for file paths for better compatibility with Posix file systems that use binary strings for file paths. This module was causing test failures due to some use of os.path module with mixed string types.

#29 https connections fail on POST after remote server hangup

The currently implemented solution is to allow an open ssl socket to be idle in the ‘blocked’ state for a maximum of 2s before sending a new request. After that time we tear down the socket and build a new one. This may now be a bit aggressive given the newer SSL behaviour (which differentiates issues in the underlying socket with different SSL exceptions).

#30 Provide http connection clean-up thread

The implementation is not as intelligent as I’d like it to be. The protocol version that a server last used is stored on the connection object and is lost when we clean up idle connections. Although it is likely that a new connection will speak the same protocol as the previous one there is little harm in going in to protocol detection mode again (we declare ourselves HTTP/1.1) other than the problem of using transfer encodings on an initial POST. In particular, we only drop out of keep-alive mode when the server has actually responded with an HTTP/1.0 response.

#38 Make Pyslet run under Python 3

See above for details.

#43 Fixes for Python running on Windows

This issue came back again, both unicode file name problems and further problems due to timing in unittests. Fixed this time by mocking and monkey-patching the time.time function in the QTI tests.

#47 Improve CharClass-derived doc strings

Fixed - no functional changes.

#49 Typo in pyslet/odata2/csdl.py

Fixed OData serialisation of LongDescription element - thanks to @thomaseitler

#51 Fix processing of dates in JSON format OData communication by the #server

We now accept ISO string formatted dates for both DateTime and DateTimeOffset. Note that providing a timezone other than Z (+00:00) when setting a DateTime will cause the time to be zone-shifted to UTC before the value is set. Thanks to @ianwj5int.

#53 Use datetime.date to create datetime object

You can now set DateTimeValue using a standard python datetime.date, the value is extended to be 00:00:00 on that date. Thanks to @nmichaud

#54 Fix use of super to remove self

Fixed Atom Date handling bug, thanks to @nmichaud

#55 Replace print_exception with logging (this includes the traceback)

Thanks to @ianwj5int for reporting.

#56 Garbage received when server delays response

This was caused by a bug when handling 401 responses in HTTP client

The issue affected any response that was received as a result of a resend (after a redirect or 401 response). The stream used to receive the data in the follow-up request was not being reset correctly and this resulted in a chunk of 0x00 bytes being written before the actual content.

This bug was discovered following changes in the 20160209 build when StringIO was replaced with BytesIO for Python 3 compatibility. StringIO.truncate moves the stream pointer, BytesIO.truncate does not. As a result all resends where the 3xx or 401 response had a non-zero length body were being affected. Previously the bug only affected the rarer use case of resends of streamed downloads to real files, i.e., requests created by passing an open file in the res_body argument of ClientRequest.

With thanks to @karurosu for reporting.

#58 OData default values (PUT/PATCH/MERGE)

Warning: if you use Pyslet for an OData server please check that PUTs are still working as required.

Changed the SQL data stores to use DEFAULT values from the metadata file as part of the CREATE TABLE queries. Modified update_entity in memds, and SQL storage layers to use MERGE semantics by default, added option to enable replace (PUT) semantics using column defaults. This differs from the previous (incorrect behaviour) where unselected properties were set to NULL.

Updated OData server to support MERGE and ensured that PUT now uses the correct semantics (set to default instead of NULL) for values missing from the incoming request.

Improved error handling to reduce log noise in SQL layer.

#60 authentication example in docs

Added a first cut at a documentation page for HTTP auth.

#61 Add support for NTLM

Experimental support for NTLM authentication now available using the python-ntlm3 module from pip/GitHub which must be installed before you can use NTLM. The module is in pyslet.ntlmauth and it can be used in a similar way to Basic auth (see set_ntlm_credentials for details.)

Improved handling of error responses in all HTTP requests (includes a Python 3 bug fix) to enable the connection to be kept open more easily during pipelined requests that are terminated early by a final response from the server. This allows a large POST that generates a 401 response to abort sending of chunked bodies and retry without opening a new connection - vital for NTLM which is connection based.

Added automated resend after 417 Expectation failed responses as per latest HTTP guidance. (Even for POST requests!)

#64 Add a LICENSE file

Added to distribution

#65 syntax error in sqlds.SQLCollectionBase.sql_expression_substring

Also added an override for SQLite given the lack of support for the standard substring syntax.

#70 Fix for grouped unary expressions

The bug is best illustrated by attempting to parse OData expressions containing “(not false)”. Thanks to @torokokill for spotting the issue.

#71 $filter fails when querying fieldnames matching OData literal types

The names that introduce typed literals such as time, datetime, guid, binary, X, etc. can now be used in URL expressions without raising parser errors. The reserved names null, true and false continue to be interpreted as literals so properties with any of those names cannot be referred to in expressions. Thanks to @soundstripe for reporting this.

#72 Travis CI tests failing in Python 3.5

Resolved but Travis no longer builds Python 2.6, see above for details.

#74 New release with bugfixes?

Resolved with the release of 0.7

1.2.3. Untracked Fixes

HTTP related:

Fixed an issue with HTTP resends (e.g., when following redirects) that meant that the retry algorithm was causing the client to back off when more than 1 resend was required.

Added compatibility in HTTP client for parsing dates from headers where the server uses the zone designator “UTC” instead of the required “GMT”.

Fixed a bug where the HTTP client would fail if it received multiple WWW-Authenticate headers in the same response (parser bug).

Better handling of non-blocking io in HTTP client fixing issues when a message body is being received to a local stream that is itself blocked. Includes a new wrapper for RawIOBase in Python 2.6 (with a fix for blocking stream bug)

Fixed bug in HTTP client when following relative path redirects

XML/HTML Parser:

Deprecated XML Element construction with name override to improve handling of super.

Fixed a bug in the parsing of HTML content where unexpected elements that belong in the <head> were causing any preceding <body> content to be ignored. Added the get_or_add_child method to XML Elements to deal with cases where add_child’s ‘reset’ of the element’s children is undesired.

Fixed a bug in the XML parser where the parsed DTD was not being set in the Document instance.

CDATA sections were not being generated properly by the (old) function pyslet.xml.structures.EscapeCDSect(), causing the HTML style and script tags to have their content rendered incorrectly. These tags are not part of the QTI content model so this bug is unlikely to have had an impact on real data.

XMLEntity class is now a context manager to help ensure that files are closed before garbage collection. Unittests were triggering resource leak warnings in Python 3.

Fixed a bug in the XML tests that shows up on Windows if the xml test files are checked out with auto-translation of line ends.


Fixed a bug in the detect_encoding function in unicode5 module (most likely benign).

Added support for expanded dates to iso8601 module (merged from OData v4 branch).

Refactoring of second truncation in iso8601 to use Python decimals.

Fix for comparison of midnight TimePoints not in canonical form

vfs: VirtualFilePath objects are now sortable.

Use of nested generators was triggering future warnings in Python 3, refactored to catch StopIteration as per: https://www.python.org/dev/peps/pep-0479/

Added SortableMixin to emulate Python 3 TypeErrors in comparisons and to simplify implementation of comparison/hash operators in custom classes. As a result, some Time/TimePoint comparisons which used to raise ValueError (e.g., due to incompatible precision) now return False for == and != operators and raise TypeError for inequalities (<, >, etc). OData is unaffected as OData time values of the same EDM type are always comparable.

Re-factored previously undocumented stream classes into their own module, in particular the Pipe implementation used for inter-thread communication. Adding documentation for them.

Re-factored the WSGI InputWrapper from rfc5023 into the http modules.

Sample code:

The sample code has also been updated to work in Python 3, including the weather OData service using MySQL but this now connects through PyMySQL as MySQLdb is not supported in Python 3.

scihub.esa.int has been renamed to scihub.copernicus.eu and the sample code has been updated accordingly with the latest metadata-fixes and tested using Python 3.

1.3. Version 0.6.20160201

Summary of New Features:
LTI module rewritten, now suitable for real applications! WSGI-based web-app framework built using Pyslet’s DAL MySQL Database connector for Pyslet’s DAL SSL, Certificates and HTTP Basic Authentication HTTP Cookies URNs

#3 PEP-8 driven refactoring (ongoing)

Added new method decorators to make supporting renamed and redirected methods easier. Added checks for ambiguous names in classes likely to have been sub-classed by third-party code.

#8 Support for SSL Certificates in HTTP Clients

Fixed certificate support in OData and Atom clients. See blog post for further information on how to use certificates: http://swl10.blogspot.co.uk/2014/11/basic-authentication-ssl-and-pyslets.html

#9 HTTP client retry strategy

Improved HTTP retries with simple Fibonacci-based back-off. Also fixed a bug where, if the first request after a server timed out an idle connection is a POST, the request would fail.

#12 bug when using numeric or named parameters in DB API

The basic bug is fixed and I’ve also added support for paramstyle ‘format’.

#14 content element missing in media-link entries

Fixed. Affected atom xml formatted entities only.

#15 MySQL implementation of Pyslet’s DAL (ongoing)

Changes to the core DAL to deal to better support other DB modules. These included added support for LIMIT clauses to speed up paged access to large entity sets. Implementation of a retry strategy when database commands return OperationalError (e.g., MySQL idle timeouts). An updated connection pool manager and an optional pool cleaner method to clean up idle database connections.

#18 Possible bug in parsing AssociationSet names

Added a compatibility mode to odata2.csdl to enable the metadata model to optionally accept hyphen or dash characters in simple identifiers using:

import pyslet.odata2.csdl as edm

#19 OData Function parameter handling

Enabled function parameter passing in OData service operations. Only primitive types are supported but they are now parsed correctly from the query string and coerced to the declared parameter type. Bound functions now receive them as a dictionary of SimpleValue instances.

#20 HTTP Basic Authentication

Fixed an issue with the OData basic authentication support, in some cases the HTTP client was waiting for a 401 when it could have offered the credentials preemptively. See also the following blog article: http://swl10.blogspot.co.uk/2014/11/basic-authentication-ssl-and-pyslets.html

#22 Support for navigation properties in OData expressions

Although the code always contained support in general, the mapping to SQL did not previously support the use of table joins in SQL expressions. This release adds support for joins (but not for nested joins).

#23 A Framework for WSGI-based LTI Applications

Added a new module to make it easier to write WSGI-based applications. Re-factored the existing Basic LTI module to use the new oauthlib and Pyslet’s own OData-inspired data access layer.

#24 ESA Sentinel mission compatibility

Added the capability to override the metadata used by an OData server to deal with validation issues in some services. Clients can now also be created from an offline copy of the service root document.

#26 HTTP client eats memory when downloading large unchunked files

Fixed the download buffer which was failing to write out data until an entire chunk (or the entire download) was complete.

#29 https connections fail on POST after remote server hangup

Partial mitigation with an agressive 2s window in which to start sending a follow-up request when pipelining through https. This is a crude solution and the bug remains open for a more robust solution based around use of the Expect header in HTTP/1.1.

#30 HTTP client cleanup thread

Added an optional parameter to the HTTP client constructor that creates a cleanup thread to close down idle connections periodically.

#31 Removed reliance on Host header in wsgi app class

There are a number of ways an application can be attacked using a forged Host header, wsgi now ignores the Host header and uses a new setting for the preferred scheme//host:port.

#32 get_certificate_chain

Implemented a function to create a complete certificate chain. Implemented using pyOpenSSL with a lot of help from this article

#33 Fixed exception: ‘NoneType’ object has no attribute ‘current_thread’ on exit

Caused by an overly ambitious __del__ method in SQLEntityContainer.

#34 Fixed missing Edm prefix in OData sample code #35 Fixed missing import in rfc5023 (atom protocol) module #36 Fixed incorrect error messages in OData $filter queries #37 Extended comparison operators in OData to include DateTimeOffset values

All thanks to @ianwj5int for spotting

#38 Python 3 compatibility work

I have started revising modules to support Python 3. This is not yet production ready but it is a small impact on existing modules. I have done my best to maintain compatibility, in practice code should continue to work with no changes required.

The most likely failure mode is that you may find a unicode string in Python 2 where you expected a plain str. This can have a knock-on effect of promoting data to unicode, e.g., through formatting operations. In general the returned types of methods are just being clarified and unicode values are returned only where they may have been returned previously anyway. However, in the case of the URI attributes in the rfc2396 module the types have changed from str to unicode in this release.

This is work in progress but the impact is likely to be minimal at this stage.

#40 & #41 Composite keys and Slug headers

Key hints were not working properly between the OData client and server implementations, and were not working at all when the key was composite. It is now possible to pass the formatted entity key predicate (including the brackets) as a Slug to the OData server and it will attempt to parse it and use that key where allowed by the underlying data layer.

#43 Fixes for Python running on Windows

The only substantive changes required were to the way we check for io failures when IOError is raised and the way we handle URI containing non-ASCII characters. Some of the unit tests were also affected due to issues with timing, including the reduced precision of time.time() on Windows-based systems.

Untracked enhancements:

Added a new module to support HTTP cookies. The HTTP/OData client can now be configured to accept cookies. The default behaviour is to ignore them so this won’t affect existing applications.

Added a new module to support URN syntax to provide a better implementation of the IMS LTI vocabularies.

Added an optional params dictionary to the OData expression parser to make it much easier to parse parameterized OData queries.

Added new methods for creating and executing drop table statements in the DAL.

Reworked sample code for the weather data server, included example driver files for mod_wsgi

Other fixes:

Fixed an issue in the OData client that caused basic key lookup in filtered entity collections to use both a key predicate and a $filter query option. This was causing the filter to be ignored, now the key predicate will be added to the filter rather than the path segment.

Fixed the OData DateTime parser to accept (and discard) any time zone specifier given in the literal form as it is now allowed in the ABNF and may therefore be generated by OData servers.

Fixed a bug in the OData server which meant that requests for JSON format responses were not being limited by the builtin topmax and would therefore attempt to return all matching entities in a single response.

Fixed a bug in the OData server which meant that use of $count was causing the $filter to be ignored!

Fixed a bug in the OData URI parser that prevent compound keys from working properly when zealous escaping was used.

Fixed a bug in the OData server which meant that error messages that contained non-ASCII characters were causing a 500 error due to character encoding issues when outputting the expected OData error format.

Fixed a bug in the OData expression evaluator when evaluating expressions that traversed navigation properties over optional relations. If there was no associated entity an error was being raised.

Fixed a bug in the SQL DAL implementation which means that navigation properties that require joining across a composite key were generating syntax errors, e.g., in SQLite the message ‘near “=”: syntax error’ would be seen.

Fixed a bug in the SQLite DAL implementation which means that in-memory databases were not working correctly in multi-threaded environments.

Fixed XML parser bug, ID elements in namespaced documents were not being handled properly.

Fixed bug in the OData server when handling non-URI characters in entity keys

Fixed a bug with composite key handling in media streams when using the SQL layer

1.4. Version 0.5.20140801

Summary of New Features:

  • OData Media Resources
  • HTTP Package refactoring and retry handling
  • Python 2.6 Support

Tracked issues addressed in this release:

#1 added a Makefile to make it easier for others to build and develop the code

Added a tox.ini file to enable support for tox (a tool for running the unittests in multiple Python environments).

#3 PEP-8 driven refactoring (ongoing)

#2 Migrated the code from SVN to git: https://github.com/swl10/pyslet

#4 Added support for read-only properties and tests for auto generated primary and foreign key values

#6 added integration between git and travis ci (thanks @sassman for your help with this)

#10 restored support for Python 2.6

1.4.1. Other Fixes

OData URLs with reserved values in their keys were failing. For example Entity(‘why%3F’) was not being correctly percent-decoded by the URI parsing class ODataURI. Furthermore, the server implementation was fixed to deal with the fact that PATH_INFO in the WSGI environ dictionary follows the CGI convention of being URL-decoded.

1.5. Version 0.4 and earlier

These are obsolete, version 0.4 was developed on Google Code as an integral part of the QTI Migration tool.

1.6. PyAssess

A precursor to Pyslet. For more information see: https://code.google.com/p/qtimigration/wiki/PyAssess