| author |  Marc Tiehuis <marc@tiehu.is> 2024-02-25 19:15:22+13:00 |
| committer |  Andrew Kelley <andrew@ziglang.org> 2024-02-25 12:37:03-08:00 |
| log | ff3bf983459c5c58ec0374f6f52627902f13e721 |
| tree | 3bd084aa161643c696dc0441b0b7e04bec836459 |
| parent | 08e886b8fec4c51366517b1c1d52d588c24abb4f |
fix large f128 values being incorrectly parsed as inf
Found while fuzzing. Previously 1.1897314953572317650857593266280070162E4932
was parsed as +inf, which caused issues for round-trip serialization of
floats. Only f128 had issues, but have added other tests for all
floating point large normals.
The max_exponent for f128 was wrong, it is subtly different in the
decimal code-path as it is based on where the decimal digit should go.
This needs to be 2 greater than the max exponent (e.g. 308 or 4932) to
work correctly (greater by 1, then we use a >= comparision).
In addition, I've removed the redundant `optimize` constant which was only
use for testing the slow path locally.3 files changed, 22 insertions(+), 19 deletions(-)
lib- std
- fmt
- parse_float.zig
- parse_float
lib/std/fmt/parse_float.zig+7
| ... | @@ -78,6 +78,13 @@ test "fmt.parseFloat nan and inf" { | ... | @@ -78,6 +78,13 @@ test "fmt.parseFloat nan and inf" { |
| 78 | } | 78 | } |
| 79 | } | 79 | } |
| 80 | 80 | ||
| 81 | test "fmt.parseFloat largest normals" { | ||
| 82 | try expectEqual(@as(u16, @bitCast(try parseFloat(f16, "65504"))), 0x7bff); | ||
| 83 | try expectEqual(@as(u32, @bitCast(try parseFloat(f32, "3.4028234664E38"))), 0x7f7f_ffff); | ||
| 84 | try expectEqual(@as(u64, @bitCast(try parseFloat(f64, "1.7976931348623157E308"))), 0x7fef_ffff_ffff_ffff); | ||
| 85 | try expectEqual(@as(u128, @bitCast(try parseFloat(f128, "1.1897314953572317650857593266280070162E4932"))), 0x7ffe_ffff_ffff_ffff_ffff_ffff_ffff_ffff); | ||
| 86 | } | ||
| 87 | |||
| 81 | test "fmt.parseFloat #11169" { | 88 | test "fmt.parseFloat #11169" { |
| 82 | try expectEqual(try parseFloat(f128, "9007199254740993.0"), 9007199254740993.0); | 89 | try expectEqual(try parseFloat(f128, "9007199254740993.0"), 9007199254740993.0); |
| 83 | } | 90 | } |
lib/std/fmt/parse_float/decimal.zig+1-1
| ... | @@ -63,7 +63,7 @@ pub fn Decimal(comptime T: type) type { | ... | @@ -63,7 +63,7 @@ pub fn Decimal(comptime T: type) type { |
| 63 | pub const max_digits_without_overflow = if (MantissaT == u64) 19 else 38; | 63 | pub const max_digits_without_overflow = if (MantissaT == u64) 19 else 38; |
| 64 | pub const decimal_point_range = if (MantissaT == u64) 2047 else 32767; | 64 | pub const decimal_point_range = if (MantissaT == u64) 2047 else 32767; |
| 65 | pub const min_exponent = if (MantissaT == u64) -324 else -4966; | 65 | pub const min_exponent = if (MantissaT == u64) -324 else -4966; |
| 66 | pub const max_exponent = if (MantissaT == u64) 310 else 4933; | 66 | pub const max_exponent = if (MantissaT == u64) 310 else 4934; |
| 67 | pub const max_decimal_digits = if (MantissaT == u64) 18 else 37; | 67 | pub const max_decimal_digits = if (MantissaT == u64) 18 else 37; |
| 68 | 68 | ||
| 69 | /// The number of significant digits in the decimal. | 69 | /// The number of significant digits in the decimal. |
lib/std/fmt/parse_float/parse_float.zig+14-18
| ... | @@ -5,8 +5,6 @@ const convertEiselLemire = @import("convert_eisel_lemire.zig").convertEiselLemir | ... | @@ -5,8 +5,6 @@ const convertEiselLemire = @import("convert_eisel_lemire.zig").convertEiselLemir |
| 5 | const convertSlow = @import("convert_slow.zig").convertSlow; | 5 | const convertSlow = @import("convert_slow.zig").convertSlow; |
| 6 | const convertHex = @import("convert_hex.zig").convertHex; | 6 | const convertHex = @import("convert_hex.zig").convertHex; |
| 7 | 7 | ||
| 8 | const optimize = true; | ||
| 9 | |||
| 10 | pub const ParseFloatError = error{ | 8 | pub const ParseFloatError = error{ |
| 11 | InvalidCharacter, | 9 | InvalidCharacter, |
| 12 | }; | 10 | }; |
| ... | @@ -41,25 +39,23 @@ pub fn parseFloat(comptime T: type, s: []const u8) ParseFloatError!T { | ... | @@ -41,25 +39,23 @@ pub fn parseFloat(comptime T: type, s: []const u8) ParseFloatError!T { |
| 41 | return convertHex(T, n); | 39 | return convertHex(T, n); |
| 42 | } | 40 | } |
| 43 | 41 | ||
| 44 | if (optimize) { | 42 | if (convertFast(T, n)) |f| { |
| 45 | if (convertFast(T, n)) |f| { | 43 | return f; |
| 46 | return f; | 44 | } |
| 47 | } | ||
| 48 | 45 | ||
| 49 | if (T == f16 or T == f32 or T == f64) { | 46 | if (T == f16 or T == f32 or T == f64) { |
| 50 | // If significant digits were truncated, then we can have rounding error | 47 | // If significant digits were truncated, then we can have rounding error |
| 51 | // only if `mantissa + 1` produces a different result. We also avoid | 48 | // only if `mantissa + 1` produces a different result. We also avoid |
| 52 | // redundantly using the Eisel-Lemire algorithm if it was unable to | 49 | // redundantly using the Eisel-Lemire algorithm if it was unable to |
| 53 | // correctly round on the first pass. | 50 | // correctly round on the first pass. |
| 54 | if (convertEiselLemire(T, n.exponent, n.mantissa)) |bf| { | 51 | if (convertEiselLemire(T, n.exponent, n.mantissa)) |bf| { |
| 55 | if (!n.many_digits) { | 52 | if (!n.many_digits) { |
| 53 | return bf.toFloat(T, n.negative); | ||
| 54 | } | ||
| 55 | if (convertEiselLemire(T, n.exponent, n.mantissa + 1)) |bf2| { | ||
| 56 | if (bf.eql(bf2)) { | ||
| 56 | return bf.toFloat(T, n.negative); | 57 | return bf.toFloat(T, n.negative); |
| 57 | } | 58 | } |
| 58 | if (convertEiselLemire(T, n.exponent, n.mantissa + 1)) |bf2| { | ||
| 59 | if (bf.eql(bf2)) { | ||
| 60 | return bf.toFloat(T, n.negative); | ||
| 61 | } | ||
| 62 | } | ||
| 63 | } | 59 | } |
| 64 | } | 60 | } |
| 65 | } | 61 | } |