const std = @import("std"); const expectEqualStrings = std.testing.expectEqualStrings; pub fn snakeToCamel(allocator: std.mem.Allocator, name: []const u8) ![]u8 { var utf8_name = (std.unicode.Utf8View.init(name) catch unreachable).iterator(); var target_inx: usize = 0; var previous_ascii: u8 = 0; // A single word will take the entire length plus our sentinel const rc = try allocator.alloc(u8, name.len + 1); while (utf8_name.nextCodepoint()) |cp| { if (cp > 0xff) return error.UnicodeNotSupported; const ascii_char = @truncate(u8, cp); if (ascii_char != '_') { if (previous_ascii == '_' and ascii_char >= 'a' and ascii_char <= 'z') { const uppercase_char = ascii_char - ('a' - 'A'); rc[target_inx] = uppercase_char; } else { rc[target_inx] = ascii_char; } target_inx = target_inx + 1; } previous_ascii = ascii_char; } rc[target_inx] = 0; // add zero sentinel return rc[0..target_inx]; } pub fn snakeToPascal(allocator: std.mem.Allocator, name: []const u8) ![]u8 { const rc = try snakeToCamel(allocator, name); if (rc[0] >= 'a' and rc[0] <= 'z') { const uppercase_char = rc[0] - ('a' - 'A'); rc[0] = uppercase_char; } return rc; } test "converts from snake to camelCase" { const allocator = std.testing.allocator; const camel = try snakeToCamel(allocator, "access_key_id"); defer allocator.free(camel); try expectEqualStrings("accessKeyId", camel); } test "single word" { const allocator = std.testing.allocator; const camel = try snakeToCamel(allocator, "word"); defer allocator.free(camel); try expectEqualStrings("word", camel); }