add std.build.ConfigHeaderStep

This API converts a config.h.in file into config.h. This is useful when
introducing a build.zig file to an existing C/C++ project that is
configured with autotools or cmake.

The cmake syntax is not implemented yet.

lib/std/build.zig

@@ -21,6 +21,7 @@ const ThisModule = @This();
 
 pub const CheckFileStep = @import("build/CheckFileStep.zig");
 pub const CheckObjectStep = @import("build/CheckObjectStep.zig");
+pub const ConfigHeaderStep = @import("build/ConfigHeaderStep.zig");
 pub const EmulatableRunStep = @import("build/EmulatableRunStep.zig");
 pub const FmtStep = @import("build/FmtStep.zig");
 pub const InstallArtifactStep = @import("build/InstallArtifactStep.zig");
@@ -364,6 +365,17 @@ pub const Builder = struct {
         return run_step;
     }
 
+    pub fn addConfigHeader(
+        b: *Builder,
+        source: FileSource,
+        style: ConfigHeaderStep.Style,
+        values: anytype,
+    ) *ConfigHeaderStep {
+        const config_header_step = ConfigHeaderStep.create(b, source, style);
+        config_header_step.addValues(values);
+        return config_header_step;
+    }
+
     /// Allocator.dupe without the need to handle out of memory.
     pub fn dupe(self: *Builder, bytes: []const u8) []u8 {
         return self.allocator.dupe(u8, bytes) catch unreachable;
@@ -1427,6 +1439,7 @@ pub const Step = struct {
         emulatable_run,
         check_file,
         check_object,
+        config_header,
         install_raw,
         options,
         custom,

lib/std/build/ConfigHeaderStep.zig

@@ -0,0 +1,221 @@
+const std = @import("../std.zig");
+const ConfigHeaderStep = @This();
+const Step = std.build.Step;
+const Builder = std.build.Builder;
+
+pub const base_id: Step.Id = .config_header;
+
+pub const Style = enum {
+    /// The configure format supported by autotools. It uses `#undef foo` to
+    /// mark lines that can be substituted with different values.
+    autoconf,
+    /// The configure format supported by CMake. It uses `@@FOO@@` and
+    /// `#cmakedefine` for template substitution.
+    cmake,
+};
+
+pub const Value = union(enum) {
+    undef,
+    defined,
+    boolean: bool,
+    int: i64,
+    ident: []const u8,
+    string: []const u8,
+};
+
+step: Step,
+builder: *Builder,
+source: std.build.FileSource,
+style: Style,
+values: std.StringHashMap(Value),
+max_bytes: usize = 2 * 1024 * 1024,
+output_dir: []const u8,
+output_basename: []const u8,
+
+pub fn create(builder: *Builder, source: std.build.FileSource, style: Style) *ConfigHeaderStep {
+    const self = builder.allocator.create(ConfigHeaderStep) catch @panic("OOM");
+    const name = builder.fmt("configure header {s}", .{source.getDisplayName()});
+    self.* = .{
+        .builder = builder,
+        .step = Step.init(base_id, name, builder.allocator, make),
+        .source = source,
+        .style = style,
+        .values = std.StringHashMap(Value).init(builder.allocator),
+        .output_dir = undefined,
+        .output_basename = "config.h",
+    };
+    switch (source) {
+        .path => |p| {
+            const basename = std.fs.path.basename(p);
+            if (std.mem.endsWith(u8, basename, ".h.in")) {
+                self.output_basename = basename[0 .. basename.len - 3];
+            }
+        },
+        else => {},
+    }
+    return self;
+}
+
+pub fn addValues(self: *ConfigHeaderStep, values: anytype) void {
+    return addValuesInner(self, values) catch @panic("OOM");
+}
+
+fn addValuesInner(self: *ConfigHeaderStep, values: anytype) !void {
+    inline for (@typeInfo(@TypeOf(values)).Struct.fields) |field| {
+        switch (@typeInfo(field.type)) {
+            .Null => {
+                try self.values.put(field.name, .undef);
+            },
+            .Void => {
+                try self.values.put(field.name, .defined);
+            },
+            .Bool => {
+                try self.values.put(field.name, .{ .boolean = @field(values, field.name) });
+            },
+            .ComptimeInt => {
+                try self.values.put(field.name, .{ .int = @field(values, field.name) });
+            },
+            .EnumLiteral => {
+                try self.values.put(field.name, .{ .ident = @tagName(@field(values, field.name)) });
+            },
+            .Pointer => |ptr| {
+                switch (@typeInfo(ptr.child)) {
+                    .Array => |array| {
+                        if (ptr.size == .One and array.child == u8) {
+                            try self.values.put(field.name, .{ .string = @field(values, field.name) });
+                            continue;
+                        }
+                    },
+                    else => {},
+                }
+
+                @compileError("unsupported ConfigHeaderStep value type: " ++
+                    @typeName(field.type));
+            },
+            else => @compileError("unsupported ConfigHeaderStep value type: " ++
+                @typeName(field.type)),
+        }
+    }
+}
+
+fn make(step: *Step) !void {
+    const self = @fieldParentPtr(ConfigHeaderStep, "step", step);
+    const gpa = self.builder.allocator;
+    const src_path = self.source.getPath(self.builder);
+    const contents = try std.fs.cwd().readFileAlloc(gpa, src_path, self.max_bytes);
+
+    // The cache is used here not really as a way to speed things up - because writing
+    // the data to a file would probably be very fast - but as a way to find a canonical
+    // location to put build artifacts.
+
+    // If, for example, a hard-coded path was used as the location to put ConfigHeaderStep
+    // files, then two ConfigHeaderStep executing in parallel might clobber each other.
+
+    // TODO port the cache system from the compiler to zig std lib. Until then
+    // we construct the path directly, and no "cache hit" detection happens;
+    // the files are always written.
+    // Note there is very similar code over in WriteFileStep
+    const Hasher = std.crypto.auth.siphash.SipHash128(1, 3);
+    // Random bytes to make ConfigHeaderStep unique. Refresh this with new
+    // random bytes when ConfigHeaderStep implementation is modified in a
+    // non-backwards-compatible way.
+    var hash = Hasher.init("X1pQzdDt91Zlh7Eh");
+    hash.update(self.source.getDisplayName());
+    hash.update(contents);
+
+    var digest: [16]u8 = undefined;
+    hash.final(&digest);
+    var hash_basename: [digest.len * 2]u8 = undefined;
+    _ = std.fmt.bufPrint(
+        &hash_basename,
+        "{s}",
+        .{std.fmt.fmtSliceHexLower(&digest)},
+    ) catch unreachable;
+
+    self.output_dir = try std.fs.path.join(gpa, &[_][]const u8{
+        self.builder.cache_root, "o", &hash_basename,
+    });
+
+    var dir = std.fs.cwd().makeOpenPath(self.output_dir, .{}) catch |err| {
+        std.debug.print("unable to make path {s}: {s}\n", .{ self.output_dir, @errorName(err) });
+        return err;
+    };
+    defer dir.close();
+
+    var values_copy = try self.values.clone();
+    defer values_copy.deinit();
+
+    var output = std.ArrayList(u8).init(gpa);
+    defer output.deinit();
+    try output.ensureTotalCapacity(contents.len);
+
+    try output.appendSlice("/* This file was generated by ConfigHeaderStep using the Zig Build System. */\n");
+
+    var any_errors = false;
+    var line_index: u32 = 0;
+    var line_it = std.mem.split(u8, contents, "\n");
+    while (line_it.next()) |line| : (line_index += 1) {
+        if (!std.mem.startsWith(u8, line, "#")) {
+            try output.appendSlice(line);
+            try output.appendSlice("\n");
+            continue;
+        }
+        var it = std.mem.tokenize(u8, line[1..], " \t\r");
+        const undef = it.next().?;
+        if (!std.mem.eql(u8, undef, "undef")) {
+            try output.appendSlice(line);
+            try output.appendSlice("\n");
+            continue;
+        }
+        const name = it.rest();
+        const kv = values_copy.fetchRemove(name) orelse {
+            std.debug.print("{s}:{d}: error: unspecified config header value: '{s}'\n", .{
+                src_path, line_index + 1, name,
+            });
+            any_errors = true;
+            continue;
+        };
+        switch (kv.value) {
+            .undef => {
+                try output.appendSlice("/* #undef ");
+                try output.appendSlice(name);
+                try output.appendSlice(" */\n");
+            },
+            .defined => {
+                try output.appendSlice("#define ");
+                try output.appendSlice(name);
+                try output.appendSlice("\n");
+            },
+            .boolean => |b| {
+                try output.appendSlice("#define ");
+                try output.appendSlice(name);
+                try output.appendSlice(" ");
+                try output.appendSlice(if (b) "true\n" else "false\n");
+            },
+            .int => |i| {
+                try output.writer().print("#define {s} {d}\n", .{ name, i });
+            },
+            .ident => |ident| {
+                try output.writer().print("#define {s} {s}\n", .{ name, ident });
+            },
+            .string => |string| {
+                // TODO: use C-specific escaping instead of zig string literals
+                try output.writer().print("#define {s} \"{}\"\n", .{ name, std.zig.fmtEscapes(string) });
+            },
+        }
+    }
+
+    {
+        var it = values_copy.iterator();
+        while (it.next()) |entry| {
+            const name = entry.key_ptr.*;
+            std.debug.print("{s}: error: config header value unused: '{s}'\n", .{ src_path, name });
+        }
+    }
+
+    if (any_errors) {
+        return error.HeaderConfigFailed;
+    }
+
+    try dir.writeFile(self.output_basename, output.items);
+}

lib/std/build/LibExeObjStep.zig

@@ -28,6 +28,7 @@ const EmulatableRunStep = std.build.EmulatableRunStep;
 const CheckObjectStep = std.build.CheckObjectStep;
 const RunStep = std.build.RunStep;
 const OptionsStep = std.build.OptionsStep;
+const ConfigHeaderStep = std.build.ConfigHeaderStep;
 const LibExeObjStep = @This();
 
 pub const base_id = .lib_exe_obj;
@@ -266,6 +267,7 @@ pub const IncludeDir = union(enum) {
     raw_path: []const u8,
     raw_path_system: []const u8,
     other_step: *LibExeObjStep,
+    config_header_step: *ConfigHeaderStep,
 };
 
 pub const Kind = enum {
@@ -932,6 +934,11 @@ pub fn addIncludePath(self: *LibExeObjStep, path: []const u8) void {
     self.include_dirs.append(IncludeDir{ .raw_path = self.builder.dupe(path) }) catch unreachable;
 }
 
+pub fn addConfigHeader(self: *LibExeObjStep, config_header: *ConfigHeaderStep) void {
+    self.step.dependOn(&config_header.step);
+    self.include_dirs.append(.{ .config_header_step = config_header }) catch @panic("OOM");
+}
+
 pub fn addLibraryPath(self: *LibExeObjStep, path: []const u8) void {
     self.lib_paths.append(self.builder.dupe(path)) catch unreachable;
 }
@@ -1684,6 +1691,10 @@ fn make(step: *Step) !void {
                 try zig_args.append("-isystem");
                 try zig_args.append(fs.path.dirname(h_path).?);
             },
+            .config_header_step => |config_header| {
+                try zig_args.append("-I");
+                try zig_args.append(config_header.output_dir);
+            },
         }
     }
 

lib/std/build/RunStep.zig

@@ -17,7 +17,7 @@ const max_stdout_size = 1 * 1024 * 1024; // 1 MiB
 
 const RunStep = @This();
 
-pub const base_id = .run;
+pub const base_id: Step.Id = .run;
 
 step: Step,
 builder: *Builder,
@@ -59,7 +59,7 @@ pub fn create(builder: *Builder, name: []const u8) *RunStep {
     const self = builder.allocator.create(RunStep) catch unreachable;
     self.* = RunStep{
         .builder = builder,
-        .step = Step.init(.run, name, builder.allocator, make),
+        .step = Step.init(base_id, name, builder.allocator, make),
         .argv = ArrayList(Arg).init(builder.allocator),
         .cwd = null,
         .env_map = null,

lib/std/build/WriteFileStep.zig

@@ -63,14 +63,15 @@ fn make(step: *Step) !void {
     // files, then two WriteFileSteps executing in parallel might clobber each other.
 
     // TODO port the cache system from the compiler to zig std lib. Until then
-    // we use blake2b directly and construct the path, and no "cache hit"
-    // detection happens; the files are always written.
-    var hash = std.crypto.hash.blake2.Blake2b384.init(.{});
-
+    // we directly construct the path, and no "cache hit" detection happens;
+    // the files are always written.
+    // Note there is similar code over in ConfigHeaderStep.
+    const Hasher = std.crypto.auth.siphash.SipHash128(1, 3);
     // Random bytes to make WriteFileStep unique. Refresh this with
     // new random bytes when WriteFileStep implementation is modified
     // in a non-backwards-compatible way.
-    hash.update("eagVR1dYXoE7ARDP");
+    var hash = Hasher.init("eagVR1dYXoE7ARDP");
+
     {
         var it = self.files.first;
         while (it) |node| : (it = node.next) {
@@ -79,21 +80,23 @@ fn make(step: *Step) !void {
             hash.update("|");
         }
     }
-    var digest: [48]u8 = undefined;
+    var digest: [16]u8 = undefined;
     hash.final(&digest);
     var hash_basename: [64]u8 = undefined;
-    _ = fs.base64_encoder.encode(&hash_basename, &digest);
+    _ = std.fmt.bufPrint(
+        &hash_basename,
+        "{s}",
+        .{std.fmt.fmtSliceHexLower(&digest)},
+    ) catch unreachable;
     self.output_dir = try fs.path.join(self.builder.allocator, &[_][]const u8{
         self.builder.cache_root,
         "o",
         &hash_basename,
     });
-    // TODO replace with something like fs.makePathAndOpenDir
-    fs.cwd().makePath(self.output_dir) catch |err| {
+    var dir = fs.cwd().makeOpenPath(self.output_dir, .{}) catch |err| {
         std.debug.print("unable to make path {s}: {s}\n", .{ self.output_dir, @errorName(err) });
         return err;
     };
-    var dir = try fs.cwd().openDir(self.output_dir, .{});
     defer dir.close();
     {
         var it = self.files.first;