Changelog

Version 2.9.7

Unreleased

  • tojson filter marks output as safe to match documented behavior. #718

  • Resolved a bug where getting debug locals for tracebacks could modify template context.

  • Fixed a bug where having many {% elif ... %} blocks resulted in a “too many levels of indentation” error. These blocks now compile to native elif ..: instead of else: if ..: #759

Version 2.9.6

Released 2017-04-03

  • Fixed custom context behavior in fast resolve mode #675

Version 2.9.5

Released 2017-01-28

  • Restored the original repr of the internal _GroupTuple because this caused issues with ansible and it was an unintended change. #654

  • Added back support for custom contexts that override the old resolve method since it was hard for people to spot that this could cause a regression.

  • Correctly use the buffer for the else block of for loops. This caused invalid syntax errors to be caused on 2.x and completely wrong behavior on Python 3 #669

  • Resolve an issue where the {% extends %} tag could not be used with async environments. #668

  • Reduce memory footprint slightly by reducing our unicode database dump we use for identifier matching on Python 3 #666

  • Fixed autoescaping not working for macros in async compilation mode. #671

Version 2.9.4

Released 2017-01-10

  • Solved some warnings for string literals. #646

  • Increment the bytecode cache version which was not done due to an oversight before.

  • Corrected bad code generation and scoping for filtered loops. #649

  • Resolved an issue where top-level output silencing after known extend blocks could generate invalid code when blocks where contained in if statements. #651

  • Made the truncate.leeway default configurable to improve compatibility with older templates.

Version 2.9.3

Released 2017-01-08

  • Restored the use of blocks in macros to the extend that was possible before. On Python 3 it would render a generator repr instead of the block contents. #645

  • Set a consistent behavior for assigning of variables in inner scopes when the variable is also read from an outer scope. This now sets the intended behavior in all situations however it does not restore the old behavior where limited assignments to outer scopes was possible. For more information and a discussion see #641

  • Resolved an issue where block scoped would not take advantage of the new scoping rules. In some more exotic cases a variable overriden in a local scope would not make it into a block.

  • Change the code generation of the with statement to be in line with the new scoping rules. This resolves some unlikely bugs in edge cases. This also introduces a new internal With node that can be used by extensions.

Version 2.9.2

Released 2017-01-08

  • Fixed a regression that caused for loops to not be able to use the same variable for the target as well as source iterator. #640

  • Add support for a previously unknown behavior of macros. It used to be possible in some circumstances to explicitly provide a caller argument to macros. While badly buggy and unintended it turns out that this is a common case that gets copy pasted around. To not completely break backwards compatibility with the most common cases it’s now possible to provide an explicit keyword argument for caller if it’s given an explicit default. #642

Version 2.9.1

Released 2017-01-07

  • Resolved a regression with call block scoping for macros. Nested caller blocks that used the same identifiers as outer macros could refer to the wrong variable incorrectly.

Version 2.9

Released 2017-01-07, codename Derivation

  • Change cache key definition in environment. This fixes a performance regression introduced in 2.8.

  • Added support for generator_stop on supported Python versions (Python 3.5 and later)

  • Corrected a long standing issue with operator precedence of math operations not being what was expected.

  • Added support for Python 3.6 async iterators through a new async mode.

  • Added policies for filter defaults and similar things.

  • Urlize now sets “rel noopener” by default.

  • Support attribute fallback for old-style classes in 2.x.

  • Support toplevel set statements in extend situations.

  • Restored behavior of Cycler for Python 3 users.

  • Subtraction now follows the same behavior as other operators on undefined values.

  • map and friends will now give better error messages if you forgot to quote the parameter.

  • Depend on MarkupSafe 0.23 or higher.

  • Improved the truncate filter to support better truncation in case the string is barely truncated at all.

  • Change the logic for macro autoescaping to be based on the runtime autoescaping information at call time instead of macro define time.

  • Ported a modified version of the tojson filter from Flask to Jinja2 and hooked it up with the new policy framework.

  • Block sets are now marked safe by default.

  • On Python 2 the asciification of ASCII strings can now be disabled with the compiler.ascii_str policy.

  • Tests now no longer accept an arbitrary expression as first argument but a restricted one. This means that you can now properly use multiple tests in one expression without extra parentheses. In particular you can now write foo is divisibleby 2 or foo is divisibleby 3 as you would expect.

  • Greatly changed the scoping system to be more consistent with what template designers and developers expect. There is now no more magic difference between the different include and import constructs. Context is now always propagated the same way. The only remaining differences is the defaults for with context and without context.

  • The with and autoescape tags are now built-in.

  • Added the new select_autoescape function which helps configuring better autoescaping easier.

  • Fixed a runtime error in the sandbox when attributes of async generators were accessed.

Version 2.8.1

Released 2016-12-29

  • Fixed the for_qs flag for urlencode.

  • Fixed regression when applying int to non-string values.

  • SECURITY: if the sandbox mode is used format expressions are now sandboxed with the same rules as in Jinja. This solves various information leakage problems that can occur with format strings.

Version 2.8

Released 2015-07-26, codename Replacement

  • Added target parameter to urlize function.

  • Added support for followsymlinks to the file system loader.

  • The truncate filter now counts the length.

  • Added equalto filter that helps with select filters.

  • Changed cache keys to use absolute file names if available instead of load names.

  • Fixed loop length calculation for some iterators.

  • Changed how Jinja2 enforces strings to be native strings in Python 2 to work when people break their default encoding.

  • Added make_logging_undefined() which returns an undefined object that logs failures into a logger.

  • If unmarshalling of cached data fails the template will be reloaded now.

  • Implemented a block set tag.

  • Default cache size was increased to 400 from a low 50.

  • Fixed is number test to accept long integers in all Python versions.

  • Changed is number to accept Decimal as a number.

  • Added a check for default arguments followed by non-default arguments. This change makes {% macro m(x, y=1, z) %} a syntax error. The previous behavior for this code was broken anyway (resulting in the default value being applied to y).

  • Add ability to use custom subclasses of jinja2.compiler.CodeGenerator and jinja2.runtime.Context by adding two new attributes to the environment (code_generator_class and context_class) (pull request :issue:`404`).

  • Added support for context/environment/evalctx decorator functions on the finalize callback of the environment.

  • Escape query strings for urlencode properly. Previously slashes were not escaped in that place.

  • Add ‘base’ parameter to ‘int’ filter.

Version 2.7.3

Released 2014-06-06

  • Security issue: Corrected the security fix for the cache folder. This fix was provided by RedHat.

Version 2.7.2

Released 2014-01-10

  • Prefix loader was not forwarding the locals properly to inner loaders. This is now fixed.

  • Security issue: Changed the default folder for the filesystem cache to be user specific and read and write protected on UNIX systems. See Debian bug 734747 for more information.

Version 2.7.1

Released 2013-08-07

  • Fixed a bug with call_filter not working properly on environment and context filters.

  • Fixed lack of Python 3 support for bytecode caches.

  • Reverted support for defining blocks in included templates as this broke existing templates for users.

  • Fixed some warnings with hashing of undefineds and nodes if Python is run with warnings for Python 3.

  • Added support for properly hashing undefined objects.

  • Fixed a bug with the title filter not working on already uppercase strings.

Version 2.7

Released 2013-05-20, codename Translation

  • Choice and prefix loaders now dispatch source and template lookup separately in order to work in combination with module loaders as advertised.

  • Fixed filesizeformat.

  • Added a non-silent option for babel extraction.

  • Added urlencode filter that automatically quotes values for URL safe usage with utf-8 as only supported encoding. If applications want to change this encoding they can override the filter.

  • Added keep-trailing-newline configuration to environments and templates to optionally preserve the final trailing newline.

  • Accessing last on the loop context no longer causes the iterator to be consumed into a list.

  • Python requirement changed: 2.6, 2.7 or >= 3.3 are required now, supported by same source code, using the “six” compatibility library.

  • Allow contextfunction and other decorators to be applied to __call__.

  • Added support for changing from newline to different signs in the wordwrap filter.

  • Added support for ignoring memcache errors silently.

  • Added support for keeping the trailing newline in templates.

  • Added finer grained support for stripping whitespace on the left side of blocks.

  • Added map, select, reject, selectattr and rejectattr filters.

  • Added support for loop.depth to figure out how deep inside a recursive loop the code is.

  • Disabled py_compile for pypy and python 3.

Version 2.6

Released 2011-07-24, codename Convolution

  • Internal attributes now raise an internal attribute error now instead of returning an undefined. This fixes problems when passing undefined objects to Python semantics expecting APIs.

  • Traceback support now works properly for PyPy. (Tested with 1.4)

  • Implemented operator intercepting for sandboxed environments. This allows application developers to disable builtin operators for better security. (For instance limit the mathematical operators to actual integers instead of longs)

  • Groupby filter now supports dotted notation for grouping by attributes of attributes.

  • Scoped blocks now properly treat toplevel assignments and imports. Previously an import suddenly “disappeared” in a scoped block.

  • Automatically detect newer Python interpreter versions before loading code from bytecode caches to prevent segfaults on invalid opcodes. The segfault in earlier Jinja2 versions here was not a Jinja2 bug but a limitation in the underlying Python interpreter. If you notice Jinja2 segfaulting in earlier versions after an upgrade of the Python interpreter you don’t have to upgrade, it’s enough to flush the bytecode cache. This just no longer makes this necessary, Jinja2 will automatically detect these cases now.

  • The sum filter can now sum up values by attribute. This is a backwards incompatible change. The argument to the filter previously was the optional starting index which defaults to zero. This now became the second argument to the function because it’s rarely used.

  • Like sum, sort now also makes it possible to order items by attribute.

  • Like sum and sort, join now also is able to join attributes of objects as string.

  • The internal eval context now has a reference to the environment.

  • Added a mapping test to see if an object is a dict or an object with a similar interface.

Version 2.5.5

Released 2010-10-18

  • Built documentation is no longer part of release.

Version 2.5.4

Released 2010-10-17

  • Fixed extensions not loading properly with overlays.

  • Work around a bug in cpython for the debugger that causes segfaults on 64bit big-endian architectures.

Version 2.5.3

Released 2010-10-17

  • Fixed an operator precedence error introduced in 2.5.2. Statements like “-foo.bar” had their implicit parentheses applied around the first part of the expression (“(-foo).bar”) instead of the more correct “-(foo.bar)”.

Version 2.5.2

Released 2010-08-18

  • Improved setup.py script to better work with assumptions people might still have from it (--with-speedups).

  • Fixed a packaging error that excluded the new debug support.

Version 2.5.1

Released 2010-08-17

  • StopIteration exceptions raised by functions called from templates are now intercepted and converted to undefineds. This solves a lot of debugging grief. (StopIteration is used internally to abort template execution)

  • Improved performance of macro calls slightly.

  • Babel extraction can now properly extract newstyle gettext calls.

  • Using the variable num in newstyle gettext for something else than the pluralize count will no longer raise a KeyError.

  • Removed builtin markup class and switched to markupsafe. For backwards compatibility the pure Python implementation still exists but is pulled from markupsafe by the Jinja2 developers. The debug support went into a separate feature called “debugsupport” and is disabled by default because it is only relevant for Python 2.4

  • Fixed an issue with unary operators having the wrong precedence.

Version 2.5

Released 2010-05-29, codename Incoherence

  • Improved the sort filter (should have worked like this for a long time) by adding support for case insensitive searches.

  • Fixed a bug for getattribute constant folding.

  • Support for newstyle gettext translations which result in a nicer in-template user interface and more consistent catalogs. (Newstyle Gettext)

  • It’s now possible to register extensions after an environment was created.

Version 2.4.1

Released 2010-04-20

  • Fixed an error reporting bug for undefined.

Version 2.4

Released 2010-04-13, codename Correlation

  • The environment template loading functions now transparently pass through a template object if it was passed to it. This makes it possible to import or extend from a template object that was passed to the template.

  • Added a ModuleLoader that can load templates from precompiled sources. The environment now features a method to compile the templates from a configured loader into a zip file or folder.

  • The _speedups C extension now supports Python 3.

  • Added support for autoescaping toggling sections and support for evaluation contexts (Evaluation Context).

  • Extensions have a priority now.

Version 2.3.1

Released 2010-02-19

  • Fixed an error reporting bug on all python versions

  • Fixed an error reporting bug on Python 2.4

Version 2.3

Released 2010-02-10, codename 3000 Pythons

  • Fixes issue with code generator that causes unbound variables to be generated if set was used in if-blocks and other small identifier problems.

  • Include tags are now able to select between multiple templates and take the first that exists, if a list of templates is given.

  • Fixed a problem with having call blocks in outer scopes that have an argument that is also used as local variable in an inner frame #360.

  • Greatly improved error message reporting #339

  • Implicit tuple expressions can no longer be totally empty. This change makes {% if %} a syntax error now. #364

  • Added support for translator comments if extracted via babel.

  • Added with-statement extension.

  • Experimental Python 3 support.

Version 2.2.1

Released 2009-09-14

  • Fixes some smaller problems for Jinja2 on Jython.

Version 2.2

Released 2009-09-13, codename Kong

  • Include statements can now be marked with ignore missing to skip non existing templates.

  • Priority of not raised. It’s now possible to write not foo in bar as an alias to foo not in bar like in python. Previously the grammar required parentheses (not (foo in bar)) which was odd.

  • Fixed a bug that caused syntax errors when defining macros or using the {% call %} tag inside loops.

  • Fixed a bug in the parser that made {{ foo[1, 2] }} impossible.

  • Made it possible to refer to names from outer scopes in included templates that were unused in the callers frame #327

  • Fixed a bug that caused internal errors if names where used as iteration variable and regular variable after the loop if that variable was unused before the loop. #331

  • Added support for optional scoped modifier to blocks.

  • Added support for line-comments.

  • Added the meta module.

  • Renamed (undocumented) attribute “overlay” to “overlayed” on the environment because it was clashing with a method of the same name.

  • Speedup extension is now disabled by default.

Version 2.1.1

Released 2008-12-25

  • Fixed a translation error caused by looping over empty recursive loops.

Version 2.1

Released 2008-11-23, codename Yasuzō

  • Fixed a bug with nested loops and the special loop variable. Before the change an inner loop overwrote the loop variable from the outer one after iteration.

  • Fixed a bug with the i18n extension that caused the explicit pluralization block to look up the wrong variable.

  • Fixed a limitation in the lexer that made {{ foo.0.0 }} impossible.

  • Index based subscribing of variables with a constant value returns an undefined object now instead of raising an index error. This was a bug caused by eager optimizing.

  • The i18n extension looks up foo.ugettext now followed by foo.gettext if an translations object is installed. This makes dealing with custom translations classes easier.

  • Fixed a confusing behavior with conditional extending. loops were partially executed under some conditions even though they were not part of a visible area.

  • Added sort filter that works like dictsort but for arbitrary sequences.

  • Fixed a bug with empty statements in macros.

  • Implemented a bytecode cache system. (Bytecode Cache)

  • The template context is now weakref-able

  • Inclusions and imports “with context” forward all variables now, not only the initial context.

  • Added a cycle helper called cycler.

  • Added a joining helper called joiner.

  • Added a compile_expression method to the environment that allows compiling of Jinja expressions into callable Python objects.

  • Fixed an escaping bug in urlize

Version 2.0

Released 2008-07-17, codename Jinjavitus

  • The subscribing of objects (looking up attributes and items) changed from slightly. It’s now possible to give attributes or items a higher priority by either using dot-notation lookup or the bracket syntax. This also changed the AST slightly. Subscript is gone and was replaced with Getitem and Getattr. For more information see the implementation details.

  • Added support for preprocessing and token stream filtering for extensions. This would allow extensions to allow simplified gettext calls in template data and something similar.

  • Added jinja2.environment.TemplateStream.dump().

  • Added missing support for implicit string literal concatenation. {{ "foo" "bar" }} is equivalent to {{ "foobar" }}

  • else is optional for conditional expressions. If not given it evaluates to false.

  • Improved error reporting for undefined values by providing a position.

  • filesizeformat filter uses decimal prefixes now per default and can be set to binary mode with the second parameter.

  • Fixed bug in finalizer

Version 2.0rc1

Released 2008-06-09

  • First release of Jinja2