xhtml_parser

Rust XHTML Tree Parser

This is a simple XML/XHTML parser that constructs a read-only tree structure similar to a DOM from an Vec<u8> XML/XHTML file representation. This is used by the author for EPub reader embedded applications.

Loosely based on the PUGIXML parsing method and structure that is described here, it is an in-place parser: all strings are kept in the received Vec<u8> for which the parser takes ownership. Its content is modified to expand entities to their UTF-8 representation (in attribute values and PCData). Position index of elements is preseved in the vector. Tree nodes are kept to their minimum size for low-memory-constrained environments. A single pre-allocated vector contains all the nodes of the tree. Its maximum size depends on the xxx_node_count feature selected (see below).

The parsing process is limited to normal tags, attributes, and PCData content. No processing instruction (<? .. ?>), comment (<!-- .. -->), CDATA (<![CDATA .. ]]>), DOCTYPE (<!DOCTYPE .. >), or DTD inside DOCTYPE ([ ... ]) is retrieved. Basic validation is done to the XHTML structure to ensure content coherence.

• No unsafe construct.
• XML content must be UTF-8.
• Namespace prefix are removed from tag and attribute names (namespace_removal feature).
• Standard XML entities (&, <, >, ', and "), Unicode numerical character references (&#xhhhh; and &#nnnn;), and XHTML-related entities (as described here) are translated to their UTF-8 representation (parse_escapes feature).
• Node and Attribute vector index sizes, as well as the maximum XML file size, are configurable via features. The associated features permit you to adjust the size of structs required for the DOM tree to optimize memory usage.

Basic performance comparison

For performance comparison, a series of 20 runs were done with both PUGIXML (GNU C++) and this crate, using -O3 optimization and parsing the same 5.5 MB XML file containing 25K nodes and 25K attributes. Used the last version of PUGIXML and this crate with the default options. The values shown are the average summation of the durations with their standard deviation. Results may vary depending on the computer performance and many other aspects (system load, operating system, compiler versions, enabled options/features, data caching, etc.).

Effects of some features on node structure element size

Here is a table showing the effect that some feature combinaisons may have on the DOM-like structure sizes. The nodes and attributes structure element sizes are shown (separated with a '/'), depending on the following features:

• none, use_cstr, forward_only, use_cstr and forward_only combined.
• xxxx_node_count, xxxx_attr_count, xxxx_xml_size.

Licensing

The parser is open-source and can be freely used and modified under the terms of the MIT license.

Cargo defined Features

• default: Enables the default features of the parser.
• namespace_removal: Enables removal of XML namespaces from tag names during parsing. Default is enabled.
• parse_escapes: Enables parsing of character escapes sequences (&..;) in PCData nodes. Default is enabled.
• keep_ws_only_pcdata: all PCData nodes that are composed of whitespace only will be kept. Default is disabled.
• trim_pcdata: trim whitespaces at beginning and end of PCData nodes. Default is disabled.
• small_node_count: Uses 16-bit indices for the nodes vector. Default is enabled.
• medium_node_count: Uses 32-bit indices for the nodes vector. Default is disabled.
• large_node_count: Uses 64-bit indices for the nodes vector. Default is disabled.
• small_attr_count: Uses 16-bit indices for the attributes vector. Default is enabled.
• medium_attr_count: Uses 32-bit indices for the attributes vector. Default is disabled.
• large_attr_count: Uses 64-bit indices for the attributes vector. Default is disabled.
• small_xml_size: Allow XML files up to 64KB in length. Default is disabled.
• medium_xml_size: Allow XML files up to 4GB in length. Default is enabled.
• large_xml_size: Allow XML files up to 16 HexaBytes in length. Default is disabled.
• use_cstr: Uses an index into a null-terminated [u8] slice (C-style string) instead of a Range to represent string locations in the XML content. Default is disabled.
• forward_only: Removes node information and methods that permit going backward in the node structure. Default is disabled.
• all_features to get all features enabled under a single one, but without the following: xxxx_node_count, xxxx_attr_count, and xxxx_xml_size.

ChangeLog

[0.2.10] - 2025-07-15

• Added byte slice retrieval methods for node names, attribute names and values, and PCData.
• Some performance optimization for the Document::check_closing_tag() method located in the parser.rs file.
• Table display adjustment in README.md (no style tag allowed in github).

[0.2.9] - 2025-07-14

• Added 75 negative tests for potentially malformed XML content. Some method adjustments in support of malformed content processing.
• Added CStr retrieval methods for node names, attribute names and values, and PCData when the use_cstr feature is enabled.
• Tables formatting adjusted in documentation and readme file.

[0.2.8] - 2025-07-11

• New forward_only feature: This feature removes node information and methods that permit going backward in the node structure. This is to diminish the amount of memory required to keep the nodes structure, useful for memory-constrained context when backward displacement is not used. See the section on size effects for more information, combined or not with the use_cstr feature.

• Some code refactoring.

[0.2.7] - 2025-07-01

• Clippy (and pedantic) related refactoring.
• State::ReadTagClose parsing update. parser::translate_sequence() method revisited.
• Methods' #[inline] adjustments for better performance.
• Adjusted performance results after testing.
• Added small_attr_count, medium_attr_count, and large_attr_count features to use 16, 32, or 64-bit indices for the attributes vector, respectively. small_attr_count is the default value.
• Added small_xml_size, medium_xml_size, and large_xml_size features to accept xml file with a maximum size of 64KB (16-bit indices), 4GB (32-bit indices), or 16 HexaBytes (64-bit indices) respectively. medium_xml_size is the default value.
• The Document creation method now checks if the received XML vector is not too large for the selected capacity of nodes, attributes and file size.

[0.2.6] - 2025-06-30

Performance comparison revisited

• Using memchr for char search instead of .iter().position(). Can easily be changed through the parser::seach_char!() macro.
• The memchr crate is used without the std option.
• Using the last PUGIXML version with default options to redo performance comparison.
• Performance table adjusted accordingly.

[0.2.5] - 2025-06-29

• Restrict visibility of some methods to the crate.
• Basic performance comparison with PUGIXML.
• Code refactoring.
• phf crate version is now 0.12 .

[0.2.4] - 2025-06-29

• New feature: use_cstr: By using indices into null-terminated [u8] slices instead of a range of indices (to keep the location of strings located in the XML document), this feature reduces the size of nodes to 20 bytes instead of 24 (17% gain in size for each node). For attributes, the size is reduced from 16 bytes to 8 bytes (50% gain in size for each attribute). This change optimizes the memory required to keep the XML DOM-like tree accessible, which is particularly beneficial for embedded applications where available memory is limited. Note that using this feature reduces the overall performance of the parser by approximately 5% to 10%.
• New all_features to get all features enabled under a single one, but without the following: small_node_count, medium_node_count, and large_node_count.
• A new test case for performance computation was added.
• The no_feature feature was removed.

[0.2.3] - 2025-06-23

• Attribute value normalization: Whitespace (space, tab, carriage-return, line-feed) at the beginning and end of attribute values are removed. All other whitespace character sequences are replaced with a single space. All known entities (&..;) are translated.
• In PCData, carriage-return characters alone are replaced with a line-feed character; carriage-return are removed when followed by a line-feed.
• All parser macros are replaced with their equivalent inline method. This to simplify debugging and for better readability.
• Correction: with the keep_ws_only_pcdata feature enabled, whitespace only nodes are created after a first element tag is encountered.
• The parsing process is now ending once the first element is completely parsed (its ending tag has been encountered). All remaining content in the XML file is ignored.
• Chartype enum cleanup.
• Added no_feature feature.
• Added small_node_count, medium_node_count, and large_node_count features to use 16, 32, or 64-bit indices for the nodes vector, respectively. small_node_count is the default value.

[0.2.2] - 2025-06-17

• Better Comment, <!DOCTYPE .. > and <![CDATA[ .. ]]> bypassing parser algorithm.
• Added DTD bypassing.
• Added tests for these.
• Corrected README.md.

[0.2.1] - 2025-06-15

• Date adjustment in changelog.
• Added the changelog to README.md.

[0.2.0] - 2025-06-15

• Going to version [0.2.0] is required as the way that space characters present at the beginning and end of PCData nodes are processed is different, whether or not the following added features are enabled or disabled.
• Added keep_ws_only_pcdata: all PCData nodes that are composed of whitespace only will be kept. Default is disabled.
• Added trim_ws_pcdata: trim whitespaces at beginning and end of PCData nodes. Default is disabled.
• Corrected the description of the parse_escapes feature to add attribute values that are parsed for escapes sequences when that feature is enabled.
• Added test case for each feature. Requires to adjust selected feature before launching the individual tests.

[0.1.2] - 2025-06-12

• The Document parser method is no longer public outside of this crate.
• Added Nodes iterator to access document nodes in the sequence of creation. Accessible through the Document::all_nodes(), Document::descendants() and Node::descendants() methods.
• Added blank lines in the doc examples for better readability.
• Adjusted all examples to diminish the required use declarations.

[0.1.1] - 2025-06-11

• Added pub fn is(&self, name: &str) -> bool method to Attribute and Node modules.

• Added pub use entries in lib.rs to simplify usage in calling applications. All examples and tests have been modified in accordance with this change.

• Added Display trait definition for the ParseXmlError enum in the defs module.

• Removed the position field of the node_info struct as the information is available through the range fields of the NodeType enum.

[0.1.0] - 2025-06-10

Initial release.