canvas_ity

This is a tiny, single-header C++ library for rasterizing immediate-mode 2D vector graphics, closely modeled on the basic W3C (not WHATWG) HTML5 2D canvas specification.

The priorities for this library are high-quality rendering, ease of use, and compact size. Speed is important too, but secondary to the other priorities. Notably, this library takes an opinionated approach and does not provide options for trading off quality for speed.

Despite its small size, it supports nearly everything listed in the W3C HTML5 2D canvas specification, except for hit regions and getting certain properties. The main differences lie in the surface-level API to make this easier for C++ use, while the underlying implementation is carefully based on the specification. In particular, stroke, fill, gradient, pattern, image, and font styles are specified slightly differently (avoiding strings and auxiliary classes). Nonetheless, the goal is that this library could produce a conforming HTML5 2D canvas implementation if wrapped in a thin layer of JavaScript bindings. See the accompanying C++ automated test suite and its HTML5 port for a mapping between the APIs and a comparison of this library's rendering output against browser canvas implementations.

📝 Example

The following complete example program writes out a TGA image file and demonstrates path building, fills, strokes, line dash patterns, line joins, line caps, linear gradients, drop shadows, and compositing operations. See the HTML5 equivalent of the example on the right (scroll the code horizontally if needed) and compare them line-by-line. Note that the minor differences in shading are due to the library's use of gamma-correct blending whereas browsers typically ignore this.

canvas_ity	HTML5
#include <algorithm> #include <fstream> // Include the library header and implementation. #define CANVAS_ITY_IMPLEMENTATION #include "canvas_ity.hpp" int main() { // Construct the canvas. static int const width = 256, height = 256; canvas_ity::canvas context( width, height ); // Build a star path. context.move_to( 128.0f, 28.0f ); context.line_to( 157.0f, 87.0f ); context.line_to( 223.0f, 97.0f ); context.line_to( 175.0f, 143.0f ); context.line_to( 186.0f, 208.0f ); context.line_to( 128.0f, 178.0f ); context.line_to( 69.0f, 208.0f ); context.line_to( 80.0f, 143.0f ); context.line_to( 32.0f, 97.0f ); context.line_to( 98.0f, 87.0f ); context.close_path(); // Set up the drop shadow. context.set_shadow_blur( 8.0f ); context.shadow_offset_y = 4.0f; context.set_shadow_color( 0.0f, 0.0f, 0.0f, 0.5f ); // Fill the star with yellow. context.set_color( canvas_ity::fill_style, 1.0f, 0.9f, 0.2f, 1.0f ); context.fill(); // Draw the star with a thick red stroke and rounded points. context.line_join = canvas_ity::rounded; context.set_line_width( 12.0f ); context.set_color( canvas_ity::stroke_style, 0.9f, 0.0f, 0.5f, 1.0f ); context.stroke(); // Draw the star again with a dashed thinner orange stroke. float segments[] = { 21.0f, 9.0f, 1.0f, 9.0f, 7.0f, 9.0f, 1.0f, 9.0f }; context.set_line_dash( segments, 8 ); context.line_dash_offset = 10.0f; context.line_cap = canvas_ity::circle; context.set_line_width( 6.0f ); context.set_color( canvas_ity::stroke_style, 0.95f, 0.65f, 0.15f, 1.0f ); context.stroke(); // Turn off the drop shadow. context.set_shadow_color( 0.0f, 0.0f, 0.0f, 0.0f ); // Add a shine layer over the star. context.set_linear_gradient( canvas_ity::fill_style, 64.0f, 0.0f, 192.0f, 256.0f ); context.add_color_stop( canvas_ity::fill_style, 0.30f, 1.0f, 1.0f, 1.0f, 0.0f ); context.add_color_stop( canvas_ity::fill_style, 0.35f, 1.0f, 1.0f, 1.0f, 0.8f ); context.add_color_stop( canvas_ity::fill_style, 0.45f, 1.0f, 1.0f, 1.0f, 0.8f ); context.add_color_stop( canvas_ity::fill_style, 0.50f, 1.0f, 1.0f, 1.0f, 0.0f ); context.global_composite_operation = canvas_ity::source_atop; context.fill_rectangle( 0.0f, 0.0f, 256.0f, 256.0f ); // Fetch the rendered RGBA pixels from the entire canvas. unsigned char *image = new unsigned char[ height * width * 4 ]; context.get_image_data( image, width, height, width * 4, 0, 0 ); // Write them out to a TGA image file (TGA uses BGRA order). unsigned char header[] = { 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, width & 255, width >> 8, height & 255, height >> 8, 32, 40 }; for ( int pixel = 0; pixel < height * width; ++pixel ) std::swap( image[ pixel * 4 + 0 ], image[ pixel * 4 + 2 ] ); std::ofstream stream( "example.tga", std::ios::binary ); stream.write( reinterpret_cast< char * >( header ), sizeof( header ) ); stream.write( reinterpret_cast< char * >( image ), height * width * 4 ); delete[] image; }	<!DOCTYPE html> <html> <head> <title>Example</title> </head> <body> <canvas id="example" width="256" height="256"></canvas> <script type="text/javascript"> const context = document.getElementById( "example" ).getContext( "2d" ); // Build a star path. context.moveTo( 128.0, 28.0 ); context.lineTo( 157.0, 87.0 ); context.lineTo( 223.0, 97.0 ); context.lineTo( 175.0, 143.0 ); context.lineTo( 186.0, 208.0 ); context.lineTo( 128.0, 178.0 ); context.lineTo( 69.0, 208.0 ); context.lineTo( 80.0, 143.0 ); context.lineTo( 32.0, 97.0 ); context.lineTo( 98.0, 87.0 ); context.closePath(); // Set up the drop shadow. context.shadowBlur = 8.0; context.shadowOffsetY = 4.0; context.shadowColor = "rgba(0,0,0,0.5)"; // Fill the star with yellow. context.fillStyle = "#ffe633"; context.fill(); // Draw the star with a thick red stroke and rounded points. context.lineJoin = "round"; context.lineWidth = 12.0; context.strokeStyle = "#e60080"; context.stroke(); // Draw the star again with a dashed thinner orange stroke. const segments = [ 21.0, 9.0, 1.0, 9.0, 7.0, 9.0, 1.0, 9.0 ]; context.setLineDash( segments ); context.lineDashOffset = 10.0; context.lineCap = "round"; context.lineWidth = 6.0; context.strokeStyle = "#f2a626"; context.stroke(); // Turn off the drop shadow. context.shadowColor = "rgba(0,0,0,0.0)"; // Add a shine layer over the star. let gradient = context.createLinearGradient( 64.0, 0.0, 192.0, 256.0 ); gradient.addColorStop( 0.30, "rgba(255,255,255,0.0)" ); gradient.addColorStop( 0.35, "rgba(255,255,255,0.8)" ); gradient.addColorStop( 0.45, "rgba(255,255,255,0.8)" ); gradient.addColorStop( 0.50, "rgba(255,255,255,0.0)" ); context.fillStyle = gradient; context.globalCompositeOperation = "source-atop"; context.fillRect( 0.0, 0.0, 256.0, 256.0 ); </script> </body> </html>

✨ Features

High-quality rendering

• Trapezoidal area antialiasing provides very smooth antialiasing, even when lines are nearly horizontal or vertical.
• Gamma-correct blending, interpolation, and resampling are used throughout. It linearizes all colors and premultiplies alpha on input and converts back to unpremultiplied sRGB on output. This reduces muddiness on many gradients (e.g., red to green), makes line thicknesses more perceptually uniform, and avoids dark fringes when interpolating opacity.
• Bicubic convolution resampling is used whenever it needs to resample a pattern or image. This smoothly interpolates with less blockiness when magnifying, and antialiases well when minifying. It can simultaneously magnify and minify along different axes.
• Ordered dithering is used on output. This reduces banding on subtle gradients while still being compression-friendly.
• High curvature is handled carefully in line joins. Thick lines are drawn correctly as though tracing with a wide pen nib, even where the lines curve sharply. (Simpler curve offsetting approaches tend to show bite-like artifacts in these cases.)

Ease of use

• Provided as a single-header library with no dependencies beside the standard C++ library. There is nothing to link, and it even includes built-in binary parsing for TrueType font (TTF) files. It is pure CPU code and does not require a GPU or GPU context.
• Has extensive Doxygen-style documentation comments for the public API.
• Compiles cleanly at moderately high warning levels on most compilers.
• Shares no internal pointers, nor holds any external pointers. Newcomers to C++ can have fun drawing with this library without worrying so much about resource lifetimes or mutability.
• Uses no static or global variables. Threads may safely work with different canvas instances concurrently without locking.
• Allocates no dynamic memory after reaching the high-water mark. Except for the pixel buffer, flat std::vector instances embedded in the canvas instance handle all dynamic memory. This reduces fragmentation and makes it easy to change the code to reserve memory up front or even to use statically allocated vectors.
• Works with exceptions and RTTI disabled.
• Intentionally uses a plain C++03 style to make it as widely portable as possible, easier to understand, and (with indexing preferred over pointer arithmetic) easier to port natively to other languages. The accompanying test suite may also help with porting.

Compact size

• The source code for the entire library consists of a bit over 2300 lines (not counting comments or blanks), each no longer than 78 columns. Alternately measured, it has fewer than 1300 semicolons.
• The object code for the library can be less than 36 KiB on x86-64 with appropriate compiler settings for size.
• Due to the library's small size, the accompanying automated test suite achieves 100% line coverage of the library in gcov and llvm-cov.

⚠️ Limitations

• Trapezoidal antialiasing overestimates coverage where paths self-intersect within a single pixel. Where inner joins are visible, this can lead to a "grittier" appearance due to the extra windings used.
• Clipping uses an antialiased sparse pixel mask rather than geometrically intersecting paths. Therefore, it is not subpixel-accurate.
• Text rendering is extremely basic and mainly for convenience. It only supports left-to-right text, and does not do any hinting, kerning, ligatures, text shaping, or text layout. If you require any of those, consider using another library to provide those and feed the results to this library as either placed glyphs or raw paths.
• TRUETYPE FONT PARSING IS NOT SECURE! It does some basic validity checking, but should only be used with known-good or sanitized fonts.
• Parameter checking does not test for non-finite floating-point values.
• Rendering is single-threaded, not explicitly vectorized, and not GPU- accelerated. It also copies data to avoid ownership issues. If you need the speed, you are better off using a more fully-featured library.
• The library does no input or output on its own. Instead, you must provide it with buffers to copy into or out of.

💻 Usage

This is a single-header library. You may freely include it in any of your source files to declare the canvas_ity namespace and its members. However, to get the implementation, you must

#define CANVAS_ITY_IMPLEMENTATION

in exactly one C++ file before including this header.

Then, construct an instance of the canvas_ity::canvas class with the pixel dimensions that you want and draw into it using any of the various drawing functions. You can then use the get_image_data() function to retrieve the currently drawn image at any time.

See each of the public member function and data member (i.e., method and field) declarations for the full API documentation. Also see the accompanying C++ automated test suite for examples of the usage of each public member, and the test suite's HTML5 port for how these map to the HTML5 canvas API.

To build the test program, either just compile the one source file directly to an executable with a C++ compiler, e.g.:

g++ -O3 -o test test.cpp

or else use the accompanying CMake file. The CMake file enables extensive warnings and also offers targets for static analysis, dynamic analysis, measuring code size, and measuring test coverage.

By default, the test harness simply runs each test once and reports the results. However, with command line arguments, it can write PNG images of the test results, run tests repeatedly to benchmark them, run just a subset of the test, or write out a new table of expected image hashes. Run the program with --help to see the usage guide for more on these.

©️ License

This software is distributed as open source under the terms of the permissive ISC license.

📫 Contributing

Please do not send pull requests! They will be politely declined at this time. This library is open source, but not currently open to outside code contributions. It is also considered largely feature-complete. (Moreover, this GitHub repository is only a mirror for publishing releases from the author's local Mercurial repository.)

Bug reports, discussions, kudos, and notices of nifty projects built using this library are most welcome, however.