text/richtext processor

Utility for processing formatted text content in text/richtext format, as defined in RFC 1341, section 7.1.3. The processor supports loading, normalizing and converting the input into multiple output formats, and is extensible.

This project is on a hiatus because I'm currently involved in ML & LLM research, which I find more challenging and therefore engaging.

Differences from the format specification

This implementation, while compatible with the specification, is slightly more lenient as for what kind of input it tolerates (as recommended by the specification itself), but also supports a more strict behavior.

Additional features

The processor supports the UTF8 encoding, and uses UTF8 internally and in its output. The default input encoding can be set to either US-ASCII or UTF8. Sections marked by the ISO-8859-X and US-ASCII commands are fully supported, including arbitrary nesting, and will have their contents converted to UTF8 during tokenization.

While the RFC states that 'As of the publication of this document, the legitimate values for "X" [in "ISO-8859-X"] are the digits 1 through 9.', all ISO-8859 parts are supported, that is including 10, 11, 13, 14, 15 and 16 (The work in making ISO-8859-12 for Devanagari was officially abandoned in 1997).
The processor is extensible through custom command processing interpreters, enabling easier implementation of extensions to the standard set of text/richtext commands.

Supported (optional) extensions to text/richtext syntax

Command names may be of any length and contain any characters except for < and >, while the RFC states that a command name must be up to 40 characters, all of which must be 'in US-ASCII, restricted to the alphanumeric and hyphen ("-") characters' (these constraints can be enforced by providing a custom command interpreter if desired).
Command names can be processed in a case-sensitive way if desired. The specification-recommended (implied, actually) behavior is to treat command names as case-insensitive, and this implementation supports this. When in case-sensitive mode, the standard command names are recognized in capitalization used in their definitions on pages 23 and 24 of RFC 1341.

Please note that the C implementation is capable of interpreting only the A-Z letters of english alphabet in case-insensitive mode (which is compliant with the RFC 1341) due to complexity of turning unicode text to upper/lower case.

Behavior based on vagueness or ambiguity in the specification

Comment content, if containing commands, must be well-formed. The specification does not state how should the commands insides comments be treated and validated.
Single line of text may combine variously flushed text (e.g. a few words aligned left and few words aligned right). Note that not all output formats may support this feature.
Indentation level may be changed within a single line of text, and nested formatting commands (e.g. <Bold>) can be used to increase their effect (e.g. even bolder text). Note that no all output formats may support this.
While the specification states that 'Those implementations that can do so are encouraged to deal reasonably with improperly nested richtext', this implementation rejects such content with an error. The specification does not state how should such situation be resolved (the two obvious choices are terminate all unterminated commands and ignoring extra terminations, and preserving the behavior of the command not terminated before terminating its ancestor until the command at hand is terminated; and various commands may be preferred to be treated differently). Because of this, this implementation does not attempt to resolve it and rejects the content. If there is a need for accepting unbalanced and improperly nested commands, the recommendation is to implement the desired balancing and re-nesting and pass the repaired content to the processor.
The <SamePage> command acts as layout block separator and is processed into a pair of blocks marking the start and end of the same-page content. The content is therefore placed inside a non-explicit paragraph. This still allows the renderer to process it as inline content if desired, even though that makes the layout computation lot more complex (requiring possible bail-out and backtracking) instead of than treating such content as stand-alone content. The renderers are therefore not required to render same-page content as continuation of the current line because of this.

Internal process details

The processor handles the provided text/richtext input using the following pipeline:

input text/richtext document
             ||
             ||
             \/
      .-------------.
      |  Tokenizer  |  Performs tokenization on input and converts tokens to
      '-------------'  UTF-8
             ||
             ||  tokens
             \/
        .----------.
        |  Parser  |  Parses the tokens into an abstract syntax tree (AST)
        '----------'
             ||
             ||  abstract syntax tree
             \/
   .-------------------.
   |  Layout resolver  |  Creates normalized abstract layout of input document
   '-------------------'
             ||
             ||  input document's abstract layout
             \/
.-------------------------.
|  Layout post-processor  |  Modifies the pre-processed document in any desired
'-------------------------'  way
             ||
             ||  post-processed abstract layout
             \/
   .-------------------.
   |  Output renderer  |  Formats the abstract layout into the output format.
   '-------------------'
             ||
             ||
             \/
the input document in output format

While the Tokenizer, Parser and Layout resolver are fixed, the Layout post-processor and Output renderer are chosen by the caller of the processor. The Layout resolver, and some Layout post-processors and Output renderers, support further behavior customization through callback function(s) that enable more complex processing and easier implementation of extensions to the basic text/richtext feature set.

Simplified example usage

#include <stdlib.h>
#include <stdio.h>
#include "bool.h"
#include "processor.h"
#include "string.h"
#include "output/simple_plaintext.h"

string *input = string_from("<bold>This</bold> is a simple demo");
OutputRenderer outputRenderer = simplePlaintextOutputRenderer;
SimplePlaintextOutputRendererConfiguration config;
config.maxLineLength = 80;

ProcessorResult *result =
    process(input, true, true, NULL, NULL, outputRenderer, &config);
string *output = result->result.output;

/* Convert output to a C-style string */
char *output_c = calloc((output->length + 1) * sizeof(char));
memcpy(output_c, output->content, output->length);
print(output_c); /* Prints "This is a simple demo" */

Code portability, compatibility and style

This project aims to maximize portability and compatibility with various compilers. To accomplish this, it is written in ANSI C (C89) with the -Wtraditional gcc flag set. While this does not guarantee compatibility with every C compiler out there (the original K&R C compiler, for example), if you are using an open source, freeware, or free for non-commercial use C compiler that cannot compile this project, please let us know, we might be able to fix this.

The project is verified to be compilable in the following compilers:

gcc 9.4 or higher

The code itself uses no dependencies apart from the standard lib C, using only functions from ANSI C era.

The code style is borrowed from Linux kernel and is enforced using indent.

Use of assistive AI technologies in this project

This project has been developed without the use of assistive AI technologies, coding or other. This is primarily due to unresolved (at the time of writing this) legal concerns regarding any content generated by an AI/ML system.

This can and will be reconsidered as the situation further unfolds.

seznam/richtext-processor