# Changelog This document contains details of the various releases and their release dates. Dates are in the format `yyyy-mm-dd`. ## 0.3.4 - 2015-04-19 XML and HTML entities are decoded in the SAX parser before data is passed to a custom handler class. See commit da62fcd75d0889e4539e7390777a906a914a78c0 for more information. ## 0.3.3 - 2015-04-18 ### Improved lexer support for script/style tags Commit 73fbbfbdbdecafcf5f873b8a27e81c19a2e2ed0c improved support for lexing HTML script and style tags, ensuring that HTML such as the following is processed correctly: ### Lexing of extra quotes The XML lexer can now handle stray quotes that reside in the open tag of an element, for example: While technically invalid HTML certain websites such as contain HTML like this. See commit 6b779d788384b89ba30ef60c17a156216ba5b333 for more information. ### Lexing of doctypes containing newlines The XML lexer is now capable of lexing doctypes that contain newlines such as: See commit 9a0e31d0ae9fc8bbf9fdacb13100a7327d09157a for more information. ## 0.3.2 - 2015-04-15 ### Support for unquoted HTML attribute values Oga can now lex/parse HTML attribute values that don't use quotes. For example, the following is valid HTML: Foo And so is this: Foo/bar See Github issue and the following commits for more information: * bc9b9bc9537d9dc614b47323e0a6727a4ec2dd04 * d892ce97874ed0f1382df993c40a452530025f02 * afbb5858122d5aece252b957b3988787ed76168f * 23a441933ac659933646418ed62ba188bb20ff65 ### Counting newlines in XML declarations The XML lexer has been adjusted so that it counts newlines when processing XML declarations. While these newlines are not exposed to the resulting `Oga::XML::*` instances they are used when reporting errors. Previously the lexer wouldn't count newlines in XML declarations, leading to error messages referring to incorrect line numbers. This was fixed in commit e942086f2df0204fc7756c3df260297f5cadc7c2. ### Better lexer support for CDATA, comments and processing instructions The XML lexer has been tweaked so it can handle multi-line CDATA tags, comments and processing instructions, both when using a String and IO (or similar) as input. See Github issue and the following commits for more information: * b2ea20ba615953254554565e0c8b11587ac4f59c * ea8b4aa92fe746a9da19e94c3edf68b41495d992 * 8acc7fc743c9492eed2d9c885c22c1b5bec06d0f ### Performance Improvements To improve performance of the XPath evaluator (as well as generic code using Oga) the following methods now cache their return values: * `Oga::XML::Element#available_namespaces` * `Oga::XML::Element#namespace` * `Oga::XML::Node#html?` These cache of these methods is flushed automatically when needed. For example, registering a new namespace will flush the cache for `Element#available_namespaces` and `Element#namespace`. The performance of `Oga::XML::Traversal#each_node` has also been optimized, cutting down the amount of object allocations significantly. Combined these improvements should make XPath evaluation roughly 4 times faster. See the following commits for more information: * 739e3b474cb562f774a0e80f5f33b3b18ec7d8c5 * b42f9aaf322c6bb67a3ddfd2b350d72a45c1fd8f * fa838154fc19c938355e1d96c5e2dd4d8c299ba3 * b0359b37e536aef172b95b54dea91198b9512e15 ## 0.3.1 - 2015-04-08 Oga no longer decodes any HTML entities that appear inside the body of a ` Would effectively be turned into: See commit 4bdc8a3fdcc3111c1e2f7de983faaaf5bb6fffb1 for more information. ## 0.3.0 - 2015-04-03 ### Lexing of carriage returns Oga can now lex and parse XML documents using carriage returns for newlines. This was added in commit 0800654c962c20fb139a389245359bca9952dcd1. ### Improved handling of HTML namespaces Oga now ignores any declared namespaces when parsing HTML documents as HTML5 does not allow one to register custom namespaces. See commit 31764593070b29fcd16040a6a0bd553e464324cd for more information. ### Improved handling of explicitly declared default XML namespaces In the past explicitly defining the default XML namespace in a document would lead to Oga's XPath evaluator not being able to match any nodes. This has been fixed in commit 5adeae18d0e53fda3bcfb883b414dee8e3a9d87d. ### Caching of XPath/CSS expressions The CSS and XPath parsers now cache the ASTs of an expression used when querying a document using CSS or XPath. This can give a pretty noticable speed improvement, especially when running the same expression in a loop (or just many different times). Parsed expressions are stored in an LRU to prevent memory from growing forever. Currently the capacity is set to 1024 values but this can be changed as following: Oga::XPath::Parser::CACHE.maximum = 2048 Oga::CSS::Parser::CACHE.maximum = 2048 The LRU synchronizes method calls to allow safe usage from multiple threads. See the following commits for more info: * 66fa9f62ef1f5e2e447cdc724b42f2e1d58b0753 * 12aa21fb502a044d660cc53557d0a1208eb8e61d * 2c4e490614528dc873f8275fe10c34ae489cfee5 * 67d7d9af88787a8a810273e3451b194a6284b1ef ### Windows support While Oga for the most part already supported Windows a few changes for the extension compilation process were required to allow users to install Oga on Windows. Tests are run on AppVeyor (a continuous integration service for Windows platforms). Oga requires devkit () to be installed on non Cygwin/MinGW environments. Cygwin/MinGW environments probably already work, although I do not run any tests on these environments. ### SAX parsing of XML attributes Parsing of XML attributes using the SAX API was overhauled quite a bit. As these changes are not backwards compatible it's likely that existing SAX parsers will break. See commit d8b9725b82f93d92b10170612446fbbef6190fda for more information. ### Parser callbacks for XML attributes The XML parser has an extra callback method called `on_attribute` which is used to create a new attribute. This callback can be used in custom SAX parsers just like the other callbacks. ### Parser rewritten using ruby-ll The XML, CSS and XPath parsers have been re-written using ruby-ll (). While Racc served its purpose (until now) it has three main problems: 1. Performance is not as good as it should be. 2. The codebase is dated and generally hard to deal with, as such it's quite difficult to optimize in reasonable time. 3. LALR parser errors can be incredibly painful to debug. For this reason I wrote ruby-ll and replaced Oga's Racc based parsers with ruby-ll parsers. These parsers are LL(1) parsers which makes them a lot easier to debug. Performance is currently a tiny bit faster than the old Racc parsers, but this will be improved in the coming releases of both Oga and ruby-ll. See pull request for more information. ### Lazy decoding of XML/HTML entities In the past XML/HTML entities were decoded in the lexer, adding overhead even when not needed. This has been changed so that the decoding of entities only occurs when calling `XML::Text#text`. With this particular change also comes support for HTML entities and codepoint based XML/HTML entities. See commit 2ec91f130fcdfee918578d045b07367aec434260 for more information. ## 0.2.3 - 2015-03-04 This release adds support for lexing HTML `