nanovg_gl.zig

const std = @import("std");
const builtin = @import("builtin");
const Allocator = std.mem.Allocator;
const ArrayList = std.ArrayList;

const use_webgl = builtin.cpu.arch.isWasm();
pub const gl = if (use_webgl)
    @import("web/webgl.zig")
else
    @cImport({
        @cInclude("glad/glad.h");
    });

const logger = std.log.scoped(.nanovg_gl);

const nvg = @import("nanovg.zig");
const internal = @import("internal.zig");

pub const Options = struct {
    debug: bool = false,
};

pub fn init(allocator: Allocator, options: Options) !nvg {
    const gl_context = try GLContext.init(allocator, options);

    const params = internal.Params{
        .user_ptr = gl_context,
        .renderCreate = renderCreate,
        .renderCreateTexture = renderCreateTexture,
        .renderDeleteTexture = renderDeleteTexture,
        .renderUpdateTexture = renderUpdateTexture,
        .renderGetTextureSize = renderGetTextureSize,
        .renderViewport = renderViewport,
        .renderCancel = renderCancel,
        .renderFlush = renderFlush,
        .renderFill = renderFill,
        .renderStroke = renderStroke,
        .renderTriangles = renderTriangles,
        .renderDelete = renderDelete,
    };
    return nvg{
        .ctx = try internal.Context.init(allocator, params),
    };
}

const GLContext = struct {
    allocator: Allocator,
    options: Options,
    shader: Shader,
    view: [2]f32,
    textures: ArrayList(Texture),
    texture_id: i32 = 0,
    vert_buf: gl.GLuint = 0,
    calls: ArrayList(Call),
    paths: ArrayList(Path),
    verts: ArrayList(internal.Vertex),
    uniforms: ArrayList(FragUniforms),

    fn init(allocator: Allocator, options: Options) !*GLContext {
        const self = try allocator.create(GLContext);
        self.* = GLContext{
            .allocator = allocator,
            .options = options,
            .shader = undefined,
            .view = .{ 0, 0 },
            .textures = ArrayList(Texture).init(allocator),
            .calls = ArrayList(Call).init(allocator),
            .paths = ArrayList(Path).init(allocator),
            .verts = ArrayList(internal.Vertex).init(allocator),
            .uniforms = ArrayList(FragUniforms).init(allocator),
        };
        return self;
    }

    fn deinit(ctx: *GLContext) void {
        ctx.shader.delete();
        ctx.textures.deinit();
        ctx.calls.deinit();
        ctx.paths.deinit();
        ctx.verts.deinit();
        ctx.uniforms.deinit();
        ctx.allocator.destroy(ctx);
    }

    fn castPtr(ptr: *anyopaque) *GLContext {
        return @alignCast(@ptrCast(ptr));
    }

    fn checkError(ctx: GLContext, str: []const u8) void {
        if (!ctx.options.debug) return;
        const err = gl.glGetError();
        if (err != gl.GL_NO_ERROR) {
            logger.err("GLError {X:0>8} after {s}", .{ err, str });
        }
    }

    fn allocTexture(ctx: *GLContext) !*Texture {
        var found_tex: ?*Texture = null;
        for (ctx.textures.items) |*tex| {
            if (tex.id == 0) {
                found_tex = tex;
                break;
            }
        }
        if (found_tex == null) {
            found_tex = try ctx.textures.addOne();
        }
        const tex = found_tex.?;
        tex.* = std.mem.zeroes(Texture);
        ctx.texture_id += 1;
        tex.id = ctx.texture_id;

        return tex;
    }

    fn findTexture(ctx: *GLContext, id: i32) ?*Texture {
        for (ctx.textures.items) |*tex| {
            if (tex.id == id) return tex;
        }
        return null;
    }
};

const ShaderType = enum(u2) {
    fill_gradient,
    fill_image,
    simple,
    image,
};

const Shader = struct {
    prog: gl.GLuint,
    frag: gl.GLuint,
    vert: gl.GLuint,

    view_loc: gl.GLint,
    tex_loc: gl.GLint,
    colormap_loc: gl.GLint,
    frag_loc: gl.GLint,

    fn create(shader: *Shader, header: [:0]const u8, vertsrc: [:0]const u8, fragsrc: [:0]const u8) !void {
        var status: gl.GLint = undefined;
        var str: [2][*]const u8 = undefined;
        var len: [2]gl.GLint = undefined;
        str[0] = header.ptr;
        len[0] = @intCast(header.len);

        shader.* = std.mem.zeroes(Shader);

        const prog = gl.glCreateProgram();
        const vert = gl.glCreateShader(gl.GL_VERTEX_SHADER);
        const frag = gl.glCreateShader(gl.GL_FRAGMENT_SHADER);
        str[1] = vertsrc.ptr;
        len[1] = @intCast(vertsrc.len);
        gl.glShaderSource(vert, 2, &str[0], &len[0]);
        str[1] = fragsrc.ptr;
        len[1] = @intCast(fragsrc.len);
        gl.glShaderSource(frag, 2, &str[0], &len[0]);

        gl.glCompileShader(vert);
        gl.glGetShaderiv(vert, gl.GL_COMPILE_STATUS, &status);
        if (status != gl.GL_TRUE) {
            printShaderErrorLog(vert, "shader", "vert");
            return error.ShaderCompilationFailed;
        }

        gl.glCompileShader(frag);
        gl.glGetShaderiv(frag, gl.GL_COMPILE_STATUS, &status);
        if (status != gl.GL_TRUE) {
            printShaderErrorLog(frag, "shader", "frag");
            return error.ShaderCompilationFailed;
        }

        gl.glAttachShader(prog, vert);
        gl.glAttachShader(prog, frag);

        gl.glBindAttribLocation(prog, 0, "vertex");
        gl.glBindAttribLocation(prog, 1, "tcoord");

        gl.glLinkProgram(prog);
        gl.glGetProgramiv(prog, gl.GL_LINK_STATUS, &status);
        if (status != gl.GL_TRUE) {
            printProgramErrorLog(prog, "shader");
            return error.ProgramLinkingFailed;
        }

        shader.prog = prog;
        shader.vert = vert;
        shader.frag = frag;

        shader.getUniformLocations();
    }

    fn delete(shader: Shader) void {
        if (shader.prog != 0) gl.glDeleteProgram(shader.prog);
        if (shader.vert != 0) gl.glDeleteShader(shader.vert);
        if (shader.frag != 0) gl.glDeleteShader(shader.frag);
    }

    fn getUniformLocations(shader: *Shader) void {
        shader.view_loc = gl.glGetUniformLocation(shader.prog, "viewSize");
        shader.tex_loc = gl.glGetUniformLocation(shader.prog, "tex");
        shader.colormap_loc = gl.glGetUniformLocation(shader.prog, "colormap");
        shader.frag_loc = gl.glGetUniformLocation(shader.prog, "frag");
    }

    fn printShaderErrorLog(shader: gl.GLuint, name: []const u8, shader_type: []const u8) void {
        var buf: [512]gl.GLchar = undefined;
        var len: gl.GLsizei = 0;
        gl.glGetShaderInfoLog(shader, 512, &len, &buf[0]);
        if (len > 512) len = 512;
        const log = buf[0..@intCast(len)];
        logger.err("Shader {s}/{s} error:\n{s}", .{ name, shader_type, log });
    }

    fn printProgramErrorLog(program: gl.GLuint, name: []const u8) void {
        var buf: [512]gl.GLchar = undefined;
        var len: gl.GLsizei = 0;
        gl.glGetProgramInfoLog(program, 512, &len, &buf[0]);
        if (len > 512) len = 512;
        const log = buf[0..@intCast(len)];
        logger.err("Program {s} error:\n{s}", .{ name, log });
    }
};

pub const Framebuffer = struct {
    fbo: gl.GLuint,
    rbo: gl.GLuint,
    texture: gl.GLuint,
    image: nvg.Image,

    pub fn create(vg: nvg, w: u32, h: u32, flags: nvg.ImageFlags) Framebuffer {
        var defaultFBO: gl.GLint = undefined;
        var defaultRBO: gl.GLint = undefined;
        var fb: Framebuffer = undefined;

        gl.glGetIntegerv(gl.GL_FRAMEBUFFER_BINDING, &defaultFBO);
        gl.glGetIntegerv(gl.GL_RENDERBUFFER_BINDING, &defaultRBO);
        defer {
            gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, @intCast(defaultFBO));
            gl.glBindRenderbuffer(gl.GL_RENDERBUFFER, @intCast(defaultRBO));
        }

        var image_flags = flags;
        image_flags.flip_y = true;
        image_flags.premultiplied = true;
        fb.image = vg.createImageRGBA(w, h, image_flags, null);

        const gl_ctx: *GLContext = @alignCast(@ptrCast(vg.ctx.params.user_ptr));
        fb.texture = gl_ctx.findTexture(fb.image.handle).?.tex;

        // frame buffer object
        gl.glGenFramebuffers(1, &fb.fbo);
        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, fb.fbo);

        // render buffer object
        gl.glGenRenderbuffers(1, &fb.rbo);
        gl.glBindRenderbuffer(gl.GL_RENDERBUFFER, fb.rbo);
        gl.glRenderbufferStorage(gl.GL_RENDERBUFFER, gl.GL_STENCIL_INDEX8, @intCast(w), @intCast(h));

        // combine all
        gl.glFramebufferTexture2D(gl.GL_FRAMEBUFFER, gl.GL_COLOR_ATTACHMENT0, gl.GL_TEXTURE_2D, fb.texture, 0);
        gl.glFramebufferRenderbuffer(gl.GL_FRAMEBUFFER, gl.GL_STENCIL_ATTACHMENT, gl.GL_RENDERBUFFER, fb.rbo);

        if (gl.glCheckFramebufferStatus(gl.GL_FRAMEBUFFER) != gl.GL_FRAMEBUFFER_COMPLETE) {
            logger.err("FBO incomplete", .{});
        }

        return fb;
    }

    pub fn bind(fb: Framebuffer) void {
        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, fb.fbo);
    }

    pub fn unbind() void {
        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, 0);
    }

    pub fn delete(fb: *Framebuffer, vg: nvg) void {
        if (fb.fbo != 0)
            gl.glDeleteFramebuffers(1, &fb.fbo);
        if (fb.rbo != 0)
            gl.glDeleteRenderbuffers(1, &fb.rbo);
        if (fb.image.handle >= 0)
            vg.deleteImage(fb.image);
        fb.fbo = 0;
        fb.rbo = 0;
        fb.texture = 0;
        fb.image.handle = -1;
    }
};

const Texture = struct {
    id: i32,
    tex: gl.GLuint,
    width: u32,
    height: u32,
    tex_type: internal.TextureType,
    flags: nvg.ImageFlags,
};

const Blend = struct {
    src_rgb: gl.GLenum,
    dst_rgb: gl.GLenum,
    src_alpha: gl.GLenum,
    dst_alpha: gl.GLenum,

    fn fromOperation(op: nvg.CompositeOperationState) Blend {
        return .{
            .src_rgb = convertBlendFuncFactor(op.src_rgb),
            .dst_rgb = convertBlendFuncFactor(op.dst_rgb),
            .src_alpha = convertBlendFuncFactor(op.src_alpha),
            .dst_alpha = convertBlendFuncFactor(op.dst_alpha),
        };
    }

    fn convertBlendFuncFactor(factor: nvg.BlendFactor) gl.GLenum {
        return switch (factor) {
            .zero => gl.GL_ZERO,
            .one => gl.GL_ONE,
            .src_color => gl.GL_SRC_COLOR,
            .one_minus_src_color => gl.GL_ONE_MINUS_SRC_COLOR,
            .dst_color => gl.GL_DST_COLOR,
            .one_minus_dst_color => gl.GL_ONE_MINUS_DST_COLOR,
            .src_alpha => gl.GL_SRC_ALPHA,
            .one_minus_src_alpha => gl.GL_ONE_MINUS_SRC_ALPHA,
            .dst_alpha => gl.GL_DST_ALPHA,
            .one_minus_dst_alpha => gl.GL_ONE_MINUS_DST_ALPHA,
            .src_alpha_saturate => gl.GL_SRC_ALPHA_SATURATE,
        };
    }
};

const CallType = enum {
    fill,
    fill_convex,
    stroke,
    triangles,
};

const Call = struct {
    call_type: CallType,
    image: i32,
    colormap: i32,
    clip_path_offset: u32,
    clip_path_count: u32,
    path_offset: u32,
    path_count: u32,
    triangle_offset: u32,
    triangle_count: u32,
    uniform_offset: u32,
    blend_func: Blend,

    // Stencils the clips paths into the most significant bit (0x80) of the stencil buffer
    fn stencilClipPaths(call: Call, ctx: *GLContext) void {
        const clip_paths = ctx.paths.items[call.clip_path_offset..][0..call.clip_path_count];

        setUniformsSimple(ctx);

        const convex = false;
        if (convex) {
            // Only write to the highest bit
            gl.glStencilMask(0x80);
            gl.glStencilFunc(gl.GL_ALWAYS, 0x80, 0xFF);
            gl.glStencilOp(gl.GL_KEEP, gl.GL_KEEP, gl.GL_REPLACE);

            for (clip_paths) |clip_path| {
                gl.glDrawArrays(gl.GL_TRIANGLE_FAN, @intCast(clip_path.fill_offset), @intCast(clip_path.fill_count));
            }
        } else {
            gl.glStencilMask(0x7F);
            gl.glStencilFunc(gl.GL_ALWAYS, 0x00, 0xFF);
            gl.glStencilOpSeparate(gl.GL_FRONT, gl.GL_KEEP, gl.GL_KEEP, gl.GL_INCR_WRAP);
            gl.glStencilOpSeparate(gl.GL_BACK, gl.GL_KEEP, gl.GL_KEEP, gl.GL_DECR_WRAP);
            gl.glDisable(gl.GL_CULL_FACE);
            for (clip_paths) |clip_path| {
                gl.glDrawArrays(gl.GL_TRIANGLE_FAN, @intCast(clip_path.fill_offset), @intCast(clip_path.fill_count));
            }
            gl.glEnable(gl.GL_CULL_FACE);

            // cover step
            gl.glStencilFunc(gl.GL_NOTEQUAL, 0x80, 0x7F);
            gl.glStencilMask(0xFF);
            gl.glStencilOp(gl.GL_ZERO, gl.GL_ZERO, gl.GL_REPLACE);
            gl.glDrawArrays(gl.GL_TRIANGLE_STRIP, @intCast(call.triangle_offset), @intCast(call.triangle_count));
        }
    }

    fn fill(call: Call, ctx: *GLContext) void {
        gl.glEnable(gl.GL_STENCIL_TEST);
        defer gl.glDisable(gl.GL_STENCIL_TEST);
        gl.glColorMask(gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE);

        if (call.clip_path_count > 0) {
            call.stencilClipPaths(ctx);

            gl.glStencilFunc(gl.GL_EQUAL, 0x80, 0x80);
            gl.glStencilMask(0x7F); // Don't affect clip bit
        } else {
            gl.glStencilFunc(gl.GL_ALWAYS, 0x00, 0xFF);
        }

        const paths = ctx.paths.items[call.path_offset..][0..call.path_count];

        // set bindpoint for solid loc
        setUniformsSimple(ctx);
        ctx.checkError("fill simple");

        gl.glStencilOpSeparate(gl.GL_FRONT, gl.GL_KEEP, gl.GL_KEEP, gl.GL_INCR_WRAP);
        gl.glStencilOpSeparate(gl.GL_BACK, gl.GL_KEEP, gl.GL_KEEP, gl.GL_DECR_WRAP);
        gl.glDisable(gl.GL_CULL_FACE);
        for (paths) |path| {
            gl.glDrawArrays(gl.GL_TRIANGLE_FAN, @intCast(path.fill_offset), @intCast(path.fill_count));
        }
        gl.glEnable(gl.GL_CULL_FACE);

        gl.glColorMask(gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE);

        setUniforms(ctx, call.uniform_offset, call.image, call.colormap);
        ctx.checkError("fill fill");

        // Draw fill
        gl.glStencilFunc(gl.GL_NOTEQUAL, 0x00, 0x7F);
        gl.glStencilMask(0xFF);
        gl.glStencilOp(gl.GL_ZERO, gl.GL_ZERO, gl.GL_ZERO);
        gl.glDrawArrays(gl.GL_TRIANGLE_STRIP, @intCast(call.triangle_offset), @intCast(call.triangle_count));
    }

    fn fillConvex(call: Call, ctx: *GLContext) void {
        defer if (call.clip_path_count > 0) gl.glDisable(gl.GL_STENCIL_TEST);
        if (call.clip_path_count > 0) {
            gl.glEnable(gl.GL_STENCIL_TEST);
            gl.glColorMask(gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE);
            defer gl.glColorMask(gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE);

            call.stencilClipPaths(ctx);

            gl.glStencilFunc(gl.GL_EQUAL, 0x80, 0xFF);
            gl.glStencilOp(gl.GL_ZERO, gl.GL_ZERO, gl.GL_ZERO);
        }

        const paths = ctx.paths.items[call.path_offset..][0..call.path_count];

        setUniforms(ctx, call.uniform_offset, call.image, call.colormap);
        ctx.checkError("fill convex");

        for (paths) |path| {
            gl.glDrawArrays(gl.GL_TRIANGLE_FAN, @intCast(path.fill_offset), @intCast(path.fill_count));
        }
    }

    fn stroke(call: Call, ctx: *GLContext) void {
        defer if (call.clip_path_count > 0) gl.glDisable(gl.GL_STENCIL_TEST);
        if (call.clip_path_count > 0) {
            gl.glEnable(gl.GL_STENCIL_TEST);
            gl.glColorMask(gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE, gl.GL_FALSE);
            defer gl.glColorMask(gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE);

            call.stencilClipPaths(ctx);

            gl.glStencilFunc(gl.GL_EQUAL, 0x80, 0xFF);
            gl.glStencilOp(gl.GL_ZERO, gl.GL_ZERO, gl.GL_ZERO);
        }

        const paths = ctx.paths.items[call.path_offset..][0..call.path_count];

        setUniforms(ctx, call.uniform_offset, call.image, call.colormap);
        // Draw Strokes
        for (paths) |path| {
            gl.glDrawArrays(gl.GL_TRIANGLE_STRIP, @intCast(path.stroke_offset), @intCast(path.stroke_count));
        }
    }

    fn triangles(call: Call, ctx: *GLContext) void {
        setUniforms(ctx, call.uniform_offset, call.image, call.colormap);
        ctx.checkError("triangles fill");
        gl.glDrawArrays(gl.GL_TRIANGLES, @intCast(call.triangle_offset), @intCast(call.triangle_count));
    }
};

const Path = struct {
    fill_offset: u32,
    fill_count: u32,
    stroke_offset: u32,
    stroke_count: u32,
};

fn maxVertCount(paths: []const internal.Path) usize {
    var count: usize = 0;
    for (paths) |path| {
        count += path.fill.len;
        count += path.stroke.len;
    }
    return count;
}

fn xformToMat3x4(m3: *[12]f32, t: *const [6]f32) void {
    m3[0] = t[0];
    m3[1] = t[1];
    m3[2] = 0;
    m3[3] = 0;
    m3[4] = t[2];
    m3[5] = t[3];
    m3[6] = 0;
    m3[7] = 0;
    m3[8] = t[4];
    m3[9] = t[5];
    m3[10] = 1;
    m3[11] = 0;
}

fn premulColor(c: nvg.Color) nvg.Color {
    return .{ .r = c.r * c.a, .g = c.g * c.a, .b = c.b * c.a, .a = c.a };
}

const FragUniforms = struct {
    scissor_mat: [12]f32, // matrices are actually 3 vec4s
    paint_mat: [12]f32,
    inner_color: nvg.Color,
    outer_color: nvg.Color,
    scissor_extent: [2]f32,
    scissor_scale: [2]f32,
    extent: [2]f32,
    radius: f32,
    feather: f32,
    tex_type: f32,
    shaderType: f32,

    fn fromPaint(frag: *FragUniforms, paint: *nvg.Paint, scissor: *internal.Scissor, ctx: *GLContext) i32 {
        var invxform: [6]f32 = undefined;

        frag.* = std.mem.zeroes(FragUniforms);

        frag.inner_color = premulColor(paint.inner_color);
        frag.outer_color = premulColor(paint.outer_color);

        if (scissor.extent[0] < -0.5 or scissor.extent[1] < -0.5) {
            @memset(&frag.scissor_mat, 0);
            frag.scissor_extent[0] = 1;
            frag.scissor_extent[1] = 1;
            frag.scissor_scale[0] = 1;
            frag.scissor_scale[1] = 1;
        } else {
            _ = nvg.transformInverse(&invxform, &scissor.xform);
            xformToMat3x4(&frag.scissor_mat, &invxform);
            frag.scissor_extent[0] = scissor.extent[0];
            frag.scissor_extent[1] = scissor.extent[1];
            frag.scissor_scale[0] = @sqrt(scissor.xform[0] * scissor.xform[0] + scissor.xform[2] * scissor.xform[2]);
            frag.scissor_scale[1] = @sqrt(scissor.xform[1] * scissor.xform[1] + scissor.xform[3] * scissor.xform[3]);
        }

        @memcpy(&frag.extent, &paint.extent);

        if (paint.image.handle != 0) {
            const tex = ctx.findTexture(paint.image.handle) orelse return 0;
            if (tex.flags.flip_y) {
                var m1: [6]f32 = undefined;
                var m2: [6]f32 = undefined;
                nvg.transformTranslate(&m1, 0, frag.extent[1] * 0.5);
                nvg.transformMultiply(&m1, &paint.xform);
                nvg.transformScale(&m2, 1, -1);
                nvg.transformMultiply(&m2, &m1);
                nvg.transformTranslate(&m1, 0, -frag.extent[1] * 0.5);
                nvg.transformMultiply(&m1, &m2);
                _ = nvg.transformInverse(&invxform, &m1);
            } else {
                _ = nvg.transformInverse(&invxform, &paint.xform);
            }
            frag.shaderType = @floatFromInt(@intFromEnum(ShaderType.fill_image));

            if (tex.tex_type == .rgba) {
                frag.tex_type = if (tex.flags.premultiplied) 0 else 1;
            } else if (paint.colormap.handle == 0) {
                frag.tex_type = 2;
            } else {
                frag.tex_type = 3;
            }
        } else {
            frag.shaderType = @floatFromInt(@intFromEnum(ShaderType.fill_gradient));
            frag.radius = paint.radius;
            frag.feather = paint.feather;
            _ = nvg.transformInverse(&invxform, &paint.xform);
        }

        xformToMat3x4(&frag.paint_mat, &invxform);

        return 1;
    }
};

fn setUniforms(ctx: *GLContext, uniform_offset: u32, image: i32, colormap: i32) void {
    const frag = &ctx.uniforms.items[uniform_offset];
    gl.glUniform4fv(ctx.shader.frag_loc, 11, @ptrCast(frag));

    if (colormap != 0) {
        if (ctx.findTexture(colormap)) |tex| {
            gl.glActiveTexture(gl.GL_TEXTURE0 + 1);
            gl.glBindTexture(gl.GL_TEXTURE_2D, tex.tex);
            gl.glActiveTexture(gl.GL_TEXTURE0 + 0);
        }
    }

    if (image != 0) {
        if (ctx.findTexture(image)) |tex| {
            gl.glBindTexture(gl.GL_TEXTURE_2D, tex.tex);
        }
    }
    // // If no image is set, use empty texture
    // if (tex == NULL) {
    // 	tex = glnvg__findTexture(gl->dummyTex);
    // }
    // glnvg__bindTexture(tex != NULL ? tex->tex : 0);
    ctx.checkError("tex paint tex");
}

fn setUniformsSimple(ctx: *GLContext) void {
    var frag = std.mem.zeroes(FragUniforms);
    frag.shaderType = @floatFromInt(@intFromEnum(ShaderType.simple));
    gl.glUniform4fv(ctx.shader.frag_loc, 11, @ptrCast(&frag));
}

fn renderCreate(uptr: *anyopaque) !void {
    const ctx = GLContext.castPtr(uptr);

    const vertSrc = @embedFile("glsl/fill.vert");
    const fragSrc = @embedFile("glsl/fill.frag");
    const fragHeader = "";
    try ctx.shader.create(fragHeader, vertSrc, fragSrc);

    gl.glGenBuffers(1, &ctx.vert_buf);

    // Some platforms does not allow to have samples to unset textures.
    // Create empty one which is bound when there's no texture specified.
    // ctx.dummyTex = glnvg__renderCreateTexture(NVG_TEXTURE_ALPHA, 1, 1, 0, NULL);
}

fn renderCreateTexture(uptr: *anyopaque, tex_type: internal.TextureType, w: u32, h: u32, flags: nvg.ImageFlags, data: ?[]const u8) !i32 {
    const ctx = GLContext.castPtr(uptr);
    var tex: *Texture = try ctx.allocTexture();

    gl.glGenTextures(1, &tex.tex);
    tex.width = w;
    tex.height = h;
    tex.tex_type = tex_type;
    tex.flags = flags;
    gl.glBindTexture(gl.GL_TEXTURE_2D, tex.tex);

    if (!use_webgl) {
        // GL 1.4 and later has support for generating mipmaps using a tex parameter.
        if (flags.generate_mipmaps) {
            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_GENERATE_MIPMAP, gl.GL_TRUE);
        }
    }

    const data_ptr = if (data) |d| d.ptr else null;
    switch (tex_type) {
        .none => {},
        .alpha => {
            gl.glPixelStorei(gl.GL_UNPACK_ALIGNMENT, 1);
            gl.glTexImage2D(gl.GL_TEXTURE_2D, 0, gl.GL_LUMINANCE, @intCast(w), @intCast(h), 0, gl.GL_LUMINANCE, gl.GL_UNSIGNED_BYTE, data_ptr);
            gl.glPixelStorei(gl.GL_UNPACK_ALIGNMENT, 4);
        },
        .rgba => gl.glTexImage2D(gl.GL_TEXTURE_2D, 0, gl.GL_RGBA, @intCast(w), @intCast(h), 0, gl.GL_RGBA, gl.GL_UNSIGNED_BYTE, data_ptr),
    }

    if (flags.generate_mipmaps) {
        const min_filter: gl.GLint = if (flags.nearest) gl.GL_NEAREST_MIPMAP_NEAREST else gl.GL_LINEAR_MIPMAP_LINEAR;
        gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MIN_FILTER, min_filter);
    } else {
        const min_filter: gl.GLint = if (flags.nearest) gl.GL_NEAREST else gl.GL_LINEAR;
        gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MIN_FILTER, min_filter);
    }
    const mag_filter: gl.GLint = if (flags.nearest) gl.GL_NEAREST else gl.GL_LINEAR;
    gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MAG_FILTER, mag_filter);

    const wrap_s: gl.GLint = if (flags.repeat_x) gl.GL_REPEAT else gl.GL_CLAMP_TO_EDGE;
    const wrap_t: gl.GLint = if (flags.repeat_y) gl.GL_REPEAT else gl.GL_CLAMP_TO_EDGE;
    gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_S, wrap_s);
    gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_T, wrap_t);

    if (use_webgl) {
        if (flags.generate_mipmaps) {
            gl.glGenerateMipmap(gl.GL_TEXTURE_2D);
        }
    }

    return tex.id;
}

fn renderDeleteTexture(uptr: *anyopaque, image: i32) void {
    const ctx = GLContext.castPtr(uptr);
    const tex = ctx.findTexture(image) orelse return;
    if (tex.tex != 0) gl.glDeleteTextures(1, &tex.tex);
    tex.* = std.mem.zeroes(Texture);
}

fn renderUpdateTexture(uptr: *anyopaque, image: i32, x_arg: u32, y: u32, w_arg: u32, h: u32, data_arg: ?[]const u8) i32 {
    _ = x_arg;
    _ = w_arg;
    const ctx = GLContext.castPtr(uptr);
    const tex = ctx.findTexture(image) orelse return 0;

    // No support for all of skip, need to update a whole row at a time.
    const color_size: u32 = if (tex.tex_type == .rgba) 4 else 1;
    const y0: u32 = y * tex.width;
    const data = &data_arg.?[y0 * color_size];
    const x = 0;
    const w = tex.width;

    gl.glBindTexture(gl.GL_TEXTURE_2D, tex.tex);
    switch (tex.tex_type) {
        .none => {},
        .alpha => {
            gl.glPixelStorei(gl.GL_UNPACK_ALIGNMENT, 1);
            gl.glTexSubImage2D(gl.GL_TEXTURE_2D, 0, x, @intCast(y), @intCast(w), @intCast(h), gl.GL_LUMINANCE, gl.GL_UNSIGNED_BYTE, data);
            gl.glPixelStorei(gl.GL_UNPACK_ALIGNMENT, 4);
        },
        .rgba => gl.glTexSubImage2D(gl.GL_TEXTURE_2D, 0, x, @intCast(y), @intCast(w), @intCast(h), gl.GL_RGBA, gl.GL_UNSIGNED_BYTE, data),
    }
    gl.glBindTexture(gl.GL_TEXTURE_2D, 0);

    return 1;
}

fn renderGetTextureSize(uptr: *anyopaque, image: i32, w: *u32, h: *u32) i32 {
    const ctx = GLContext.castPtr(uptr);
    const tex = ctx.findTexture(image) orelse return 0;
    w.* = tex.width;
    h.* = tex.height;
    return 1;
}

fn renderViewport(uptr: *anyopaque, width: f32, height: f32, devicePixelRatio: f32) void {
    const ctx = GLContext.castPtr(uptr);
    ctx.view[0] = width;
    ctx.view[1] = height;
    _ = devicePixelRatio;
}

fn renderCancel(uptr: *anyopaque) void {
    const ctx = GLContext.castPtr(uptr);
    ctx.verts.clearRetainingCapacity();
    ctx.paths.clearRetainingCapacity();
    ctx.calls.clearRetainingCapacity();
    ctx.uniforms.clearRetainingCapacity();
}

fn renderFlush(uptr: *anyopaque) void {
    const ctx = GLContext.castPtr(uptr);

    if (ctx.calls.items.len > 0) {
        // Setup required GL state.
        gl.glUseProgram(ctx.shader.prog);

        gl.glEnable(gl.GL_CULL_FACE);
        gl.glCullFace(gl.GL_BACK);
        gl.glFrontFace(gl.GL_CCW);
        gl.glEnable(gl.GL_BLEND);
        gl.glDisable(gl.GL_DEPTH_TEST);
        gl.glDisable(gl.GL_SCISSOR_TEST);
        gl.glColorMask(gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE, gl.GL_TRUE);
        gl.glStencilMask(0xffffffff);
        gl.glStencilOp(gl.GL_KEEP, gl.GL_KEEP, gl.GL_KEEP);
        gl.glStencilFunc(gl.GL_ALWAYS, 0, 0xffffffff);
        gl.glActiveTexture(gl.GL_TEXTURE0);
        gl.glBindTexture(gl.GL_TEXTURE_2D, 0);

        gl.glBindBuffer(gl.GL_ARRAY_BUFFER, ctx.vert_buf);
        gl.glBufferData(gl.GL_ARRAY_BUFFER, @intCast(ctx.verts.items.len * @sizeOf(internal.Vertex)), ctx.verts.items.ptr, gl.GL_STREAM_DRAW);
        gl.glEnableVertexAttribArray(0);
        gl.glEnableVertexAttribArray(1);
        gl.glVertexAttribPointer(0, 2, gl.GL_FLOAT, gl.GL_FALSE, @sizeOf(internal.Vertex), null);
        gl.glVertexAttribPointer(1, 2, gl.GL_FLOAT, gl.GL_FALSE, @sizeOf(internal.Vertex), @ptrFromInt(2 * @sizeOf(f32)));

        // Set view and texture just once per frame.
        gl.glUniform1i(ctx.shader.tex_loc, 0);
        gl.glUniform1i(ctx.shader.colormap_loc, 1);
        gl.glUniform2fv(ctx.shader.view_loc, 1, &ctx.view[0]);

        for (ctx.calls.items) |call| {
            gl.glBlendFuncSeparate(call.blend_func.src_rgb, call.blend_func.dst_rgb, call.blend_func.src_alpha, call.blend_func.dst_alpha);
            switch (call.call_type) {
                .fill => call.fill(ctx),
                .fill_convex => call.fillConvex(ctx),
                .stroke => call.stroke(ctx),
                .triangles => call.triangles(ctx),
            }
        }

        gl.glDisableVertexAttribArray(0);
        gl.glDisableVertexAttribArray(1);
        gl.glDisable(gl.GL_CULL_FACE);
        gl.glBindBuffer(gl.GL_ARRAY_BUFFER, 0);
        gl.glUseProgram(0);
        gl.glBindTexture(gl.GL_TEXTURE_2D, 0);
    }

    // Reset calls
    ctx.verts.clearRetainingCapacity();
    ctx.paths.clearRetainingCapacity();
    ctx.calls.clearRetainingCapacity();
    ctx.uniforms.clearRetainingCapacity();
}

fn renderFill(
    uptr: *anyopaque,
    paint: *nvg.Paint,
    composite_operation: nvg.CompositeOperationState,
    scissor: *internal.Scissor,
    bounds: [4]f32,
    clip_paths: []const internal.Path,
    paths: []const internal.Path,
) void {
    const ctx = GLContext.castPtr(uptr);

    const call = ctx.calls.addOne() catch return;
    call.* = std.mem.zeroes(Call);
    call.call_type = .fill;
    if (paths.len == 1 and paths[0].convex) {
        call.call_type = .fill_convex;
    }
    call.triangle_count = if (call.call_type == .fill or clip_paths.len > 0) 4 else 0;

    // Allocate vertices for all the paths.
    const maxverts = maxVertCount(clip_paths) + maxVertCount(paths) + call.triangle_count;
    ctx.verts.ensureUnusedCapacity(maxverts) catch return;

    if (call.triangle_count > 0) {
        // Quad
        call.triangle_offset = @intCast(ctx.verts.items.len);
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[2], .y = bounds[3], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[2], .y = bounds[1], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[0], .y = bounds[3], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[0], .y = bounds[1], .u = 0.5, .v = 1.0 });
    }

    if (clip_paths.len > 0) {
        // TODO: optimization for convex clip paths (clip_paths.len == 1 and clip_paths[0].convex)
        ctx.paths.ensureUnusedCapacity(clip_paths.len) catch return;
        call.clip_path_offset = @intCast(ctx.paths.items.len);
        call.clip_path_count = @intCast(clip_paths.len);

        for (clip_paths) |clip_path| {
            const copy = ctx.paths.addOneAssumeCapacity();
            copy.* = std.mem.zeroes(Path);
            if (clip_path.fill.len > 0) {
                copy.fill_offset = @intCast(ctx.verts.items.len);
                copy.fill_count = @intCast(clip_path.fill.len);
                ctx.verts.appendSliceAssumeCapacity(clip_path.fill);
            }
        }
    }

    ctx.paths.ensureUnusedCapacity(paths.len) catch return;
    call.path_offset = @intCast(ctx.paths.items.len);
    call.path_count = @intCast(paths.len);
    call.image = paint.image.handle;
    call.colormap = paint.colormap.handle;
    call.blend_func = Blend.fromOperation(composite_operation);

    for (paths) |path| {
        const copy = ctx.paths.addOneAssumeCapacity();
        copy.* = std.mem.zeroes(Path);
        if (path.fill.len > 0) {
            copy.fill_offset = @intCast(ctx.verts.items.len);
            copy.fill_count = @intCast(path.fill.len);
            ctx.verts.appendSliceAssumeCapacity(path.fill);
        }
        if (path.stroke.len > 0) {
            copy.stroke_offset = @intCast(ctx.verts.items.len);
            copy.stroke_count = @intCast(path.stroke.len);
            ctx.verts.appendSliceAssumeCapacity(path.stroke);
        }
    }

    // Fill shader
    call.uniform_offset = @intCast(ctx.uniforms.items.len);
    ctx.uniforms.ensureUnusedCapacity(1) catch return;
    _ = ctx.uniforms.addOneAssumeCapacity().fromPaint(paint, scissor, ctx);
}

fn renderStroke(
    uptr: *anyopaque,
    paint: *nvg.Paint,
    composite_operation: nvg.CompositeOperationState,
    scissor: *internal.Scissor,
    bounds: [4]f32,
    clip_paths: []const internal.Path,
    paths: []const internal.Path,
) void {
    const ctx = GLContext.castPtr(uptr);

    const call = ctx.calls.addOne() catch return;
    call.* = std.mem.zeroes(Call);
    call.call_type = .stroke;
    call.triangle_count = if (clip_paths.len > 0) 4 else 0;

    // Allocate vertices for all the paths.
    const maxverts = maxVertCount(clip_paths) + maxVertCount(paths) + call.triangle_count;
    ctx.verts.ensureUnusedCapacity(maxverts) catch return;

    if (call.triangle_count > 0) {
        // Quad
        call.triangle_offset = @intCast(ctx.verts.items.len);
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[2], .y = bounds[3], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[2], .y = bounds[1], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[0], .y = bounds[3], .u = 0.5, .v = 1.0 });
        ctx.verts.appendAssumeCapacity(.{ .x = bounds[0], .y = bounds[1], .u = 0.5, .v = 1.0 });
    }

    if (clip_paths.len > 0) {
        // TODO: optimization for convex clip paths (clip_paths.len == 1 and clip_paths[0].convex)
        ctx.paths.ensureUnusedCapacity(clip_paths.len) catch return;
        call.clip_path_offset = @intCast(ctx.paths.items.len);
        call.clip_path_count = @intCast(clip_paths.len);

        for (clip_paths) |clip_path| {
            const copy = ctx.paths.addOneAssumeCapacity();
            copy.* = std.mem.zeroes(Path);
            if (clip_path.fill.len > 0) {
                copy.fill_offset = @intCast(ctx.verts.items.len);
                copy.fill_count = @intCast(clip_path.fill.len);
                ctx.verts.appendSliceAssumeCapacity(clip_path.fill);
            }
        }
    }

    ctx.paths.ensureUnusedCapacity(paths.len) catch return;
    call.path_offset = @intCast(ctx.paths.items.len);
    call.path_count = @intCast(paths.len);
    call.image = paint.image.handle;
    call.colormap = paint.colormap.handle;
    call.blend_func = Blend.fromOperation(composite_operation);

    for (paths) |path| {
        const copy = ctx.paths.addOneAssumeCapacity();
        copy.* = std.mem.zeroes(Path);
        if (path.stroke.len > 0) {
            copy.stroke_offset = @intCast(ctx.verts.items.len);
            copy.stroke_count = @intCast(path.stroke.len);
            ctx.verts.appendSliceAssumeCapacity(path.stroke);
        }
    }

    // Fill shader
    call.uniform_offset = @intCast(ctx.uniforms.items.len);
    _ = ctx.uniforms.ensureUnusedCapacity(1) catch return;
    _ = ctx.uniforms.addOneAssumeCapacity().fromPaint(paint, scissor, ctx);
}

fn renderTriangles(uptr: *anyopaque, paint: *nvg.Paint, comp_op: nvg.CompositeOperationState, scissor: *internal.Scissor, verts: []const internal.Vertex) void {
    const ctx = GLContext.castPtr(uptr);

    const call = ctx.calls.addOne() catch return;
    call.* = std.mem.zeroes(Call);

    call.call_type = .triangles;
    call.image = paint.image.handle;
    call.colormap = paint.colormap.handle;
    call.blend_func = Blend.fromOperation(comp_op);

    call.triangle_offset = @intCast(ctx.verts.items.len);
    call.triangle_count = @intCast(verts.len);
    ctx.verts.appendSlice(verts) catch return;

    call.uniform_offset = @intCast(ctx.uniforms.items.len);
    const frag = ctx.uniforms.addOne() catch return;
    _ = frag.fromPaint(paint, scissor, ctx);
    frag.shaderType = @floatFromInt(@intFromEnum(ShaderType.image));
}

fn renderDelete(uptr: *anyopaque) void {
    const ctx = GLContext.castPtr(uptr);
    ctx.deinit();
}