zig/src-self-hosted/ir.zig

const std = @import("std");
const mem = std.mem;
const Allocator = std.mem.Allocator;
const Value = @import("value.zig").Value;
const assert = std.debug.assert;

pub const Inst = struct {
    tag: Tag,

    pub const all_types = .{
        Constant,
        PtrToInt,
        FieldPtr,
        Deref,
        Assembly,
        Unreach,
    };

    pub const Tag = enum {
        constant,
        ptrtoint,
        fieldptr,
        deref,
        @"asm",
        unreach,
    };

    /// This struct owns the `Value` memory. When the struct is deallocated,
    /// so is the `Value`. The value of a constant must be copied into
    /// a memory location for the value to survive after a const instruction.
    pub const Constant = struct {
        base: Inst = Inst{ .tag = .constant },
        value: *Value,
    };

    pub const PtrToInt = struct {
        base: Inst = Inst{ .tag = .ptrtoint },
    };

    pub const FieldPtr = struct {
        base: Inst = Inst{ .tag = .fieldptr },
    };

    pub const Deref = struct {
        base: Inst = Inst{ .tag = .deref },
    };

    pub const Assembly = struct {
        base: Inst = Inst{ .tag = .@"asm" },
    };

    pub const Unreach = struct {
        base: Inst = Inst{ .tag = .unreach },
    };
};

pub const ErrorMsg = struct {
    byte_offset: usize,
    msg: []const u8,
};

pub const Tree = struct {
    decls: std.ArrayList(*Inst),
    errors: std.ArrayList(ErrorMsg),
};

const ParseContext = struct {
    allocator: *Allocator,
    i: usize,
    source: []const u8,
    errors: *std.ArrayList(ErrorMsg),
};

pub fn parse(allocator: *Allocator, source: []const u8) Allocator.Error!Tree {
    var tree: Tree = .{
        .decls = std.ArrayList(*Inst).init(allocator),
        .errors = std.ArrayList(ErrorMsg).init(allocator),
    };
    var ctx: ParseContext = .{
        .allocator = allocator,
        .i = 0,
        .source = source,
        .errors = &tree.errors,
    };
    parseRoot(&ctx, &tree) catch |err| switch (err) {
        error.ParseFailure => {
            assert(tree.errors.items.len != 0);
        },
        else => |e| return e,
    };
    return tree;
}

pub fn parseRoot(ctx: *ParseContext, tree: *Tree) !void {
    // The IR format is designed so that it can be tokenized and parsed at the same time.
    var global_name_map = std.StringHashMap(usize).init(ctx.allocator);
    while (ctx.i < ctx.source.len) : (ctx.i += 1) switch (ctx.source[ctx.i]) {
        ';' => _ = try skipToAndOver(ctx, '\n'),
        '@' => {
            const at_start = ctx.i;
            const ident = try skipToAndOver(ctx, ' ');
            var ty: ?*Value = null;
            if (eatByte(ctx, ':')) {
                ty = try parseType(ctx);
                skipSpace(ctx);
            }
            try requireEatBytes(ctx, "= ");
            const inst = try parseInstruction(ctx);
            const ident_index = tree.decls.items.len;
            if (try global_name_map.put(ident, ident_index)) |_| {
                return parseError(ctx, "redefinition of identifier '{}'", .{ident});
            }
            try tree.decls.append(inst);
            continue;
        },
        ' ', '\n' => continue,
        else => |byte| return parseError(ctx, "unexpected byte: '{c}'", .{byte}),
    };
}

fn eatByte(ctx: *ParseContext, byte: u8) bool {
    if (ctx.i >= ctx.source.len) return false;
    if (ctx.source[ctx.i] != byte) return false;
    ctx.i += 1;
    return true;
}

fn skipSpace(ctx: *ParseContext) void {
    while (ctx.i < ctx.source.len and ctx.source[ctx.i] == ' ') : (ctx.i += 1) {}
}

fn requireEatBytes(ctx: *ParseContext, bytes: []const u8) !void {
    if (ctx.i + bytes.len > ctx.source.len)
        return parseError(ctx, "unexpected EOF", .{});
    if (!mem.eql(u8, ctx.source[ctx.i..][0..bytes.len], bytes))
        return parseError(ctx, "expected '{}'", .{bytes});
    ctx.i += bytes.len;
}

fn skipToAndOver(ctx: *ParseContext, byte: u8) ![]const u8 {
    const start_i = ctx.i;
    while (ctx.i < ctx.source.len) : (ctx.i += 1) {
        if (ctx.source[ctx.i] == byte) {
            const result = ctx.source[start_i..ctx.i];
            ctx.i += 1;
            return result;
        }
    }
    return parseError(ctx, "unexpected EOF", .{});
}

fn parseError(ctx: *ParseContext, comptime format: []const u8, args: var) error{ ParseFailure, OutOfMemory } {
    const msg = try std.fmt.allocPrint(ctx.allocator, format, args);
    (try ctx.errors.addOne()).* = .{
        .byte_offset = ctx.i,
        .msg = msg,
    };
    return error.ParseFailure;
}

fn parseType(ctx: *ParseContext) !*Value {
    return parseError(ctx, "TODO parse type", .{});
}

fn parseInstruction(ctx: *ParseContext) !*Inst {
    switch (ctx.source[ctx.i]) {
        '"' => return parseStringLiteralConst(ctx),
        '0'...'9' => return parseIntegerLiteralConst(ctx),
        else => {},
    }
    const fn_name = skipToAndOver(ctx, '(');
    return parseError(ctx, "TODO parse instruction '{}'", .{fn_name});
}

fn parseStringLiteralConst(ctx: *ParseContext) !*Inst {
    const start = ctx.i;
    ctx.i += 1; // skip over '"'

    while (ctx.i < ctx.source.len) : (ctx.i += 1) switch (ctx.source[ctx.i]) {
        '"' => {
            ctx.i += 1;
            const span = ctx.source[start..ctx.i];
            var bad_index: usize = undefined;
            const parsed = std.zig.parseStringLiteral(ctx.allocator, span, &bad_index) catch |err| switch (err) {
                error.InvalidCharacter => {
                    ctx.i = start + bad_index;
                    const bad_byte = ctx.source[ctx.i];
                    return parseError(ctx, "invalid string literal character: '{c}'\n", .{bad_byte});
                },
                else => |e| return e,
            };
            const bytes_val = try ctx.allocator.create(Value.Bytes);
            bytes_val.* = .{ .data = parsed };
            const const_inst = try ctx.allocator.create(Inst.Constant);
            const_inst.* = .{ .value = &bytes_val.base };
            return &const_inst.base;
        },
        '\\' => {
            ctx.i += 1;
            if (ctx.i >= ctx.source.len) break;
            continue;
        },
        else => continue,
    };
    return parseError(ctx, "unexpected EOF in string literal", .{});
}

fn parseIntegerLiteralConst(ctx: *ParseContext) !*Inst {
    return parseError(ctx, "TODO parse integer literal", .{});
}

pub fn main() anyerror!void {
    var arena = std.heap.ArenaAllocator.init(std.heap.page_allocator);
    defer arena.deinit();
    const allocator = &arena.allocator;

    const args = try std.process.argsAlloc(allocator);

    const src_path = args[1];
    const debug_error_trace = true;

    const source = try std.fs.cwd().readFileAlloc(allocator, src_path, std.math.maxInt(u32));

    const tree = try parse(allocator, source);
    if (tree.errors.items.len != 0) {
        for (tree.errors.items) |err_msg| {
            const loc = findLineColumn(source, err_msg.byte_offset);
            std.debug.warn("{}:{}:{}: error: {}\n", .{ src_path, loc.line + 1, loc.column + 1, err_msg.msg });
        }
        if (debug_error_trace) return error.ParseFailure;
        std.process.exit(1);
    }
}

fn findLineColumn(source: []const u8, byte_offset: usize) struct { line: usize, column: usize } {
    var line: usize = 0;
    var column: usize = 0;
    for (source[0..byte_offset]) |byte| {
        switch (byte) {
            '\n' => {
                line += 1;
                column = 0;
            },
            else => {
                column += 1;
            },
        }
    }
    return .{ .line = line, .column = column };
}