zig

commit ff3bf983459c5c58ec0374f6f52627902f13e721 (tree)
parent 08e886b8fec4c51366517b1c1d52d588c24abb4f
Author: Marc Tiehuis <marc@tiehu.is>
Date:   Sun, 25 Feb 2024 19:15:22 +1300

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.

Diffstat:
M
lib/std/fmt/parse_float.zig
 | 
7
+++++++
M
lib/std/fmt/parse_float/decimal.zig
 | 
2
+-
M
lib/std/fmt/parse_float/parse_float.zig
 | 
32
++++++++++++++------------------

3 files changed, 22 insertions(+), 19 deletions(-)
diff --git a/lib/std/fmt/parse_float.zig b/lib/std/fmt/parse_float.zig
@@ -78,6 +78,13 @@ test "fmt.parseFloat nan and inf" {
     }
 }
 
+test "fmt.parseFloat largest normals" {
+    try expectEqual(@as(u16, @bitCast(try parseFloat(f16, "65504"))), 0x7bff);
+    try expectEqual(@as(u32, @bitCast(try parseFloat(f32, "3.4028234664E38"))), 0x7f7f_ffff);
+    try expectEqual(@as(u64, @bitCast(try parseFloat(f64, "1.7976931348623157E308"))), 0x7fef_ffff_ffff_ffff);
+    try expectEqual(@as(u128, @bitCast(try parseFloat(f128, "1.1897314953572317650857593266280070162E4932"))), 0x7ffe_ffff_ffff_ffff_ffff_ffff_ffff_ffff);
+}
+
 test "fmt.parseFloat #11169" {
     try expectEqual(try parseFloat(f128, "9007199254740993.0"), 9007199254740993.0);
 }
diff --git a/lib/std/fmt/parse_float/decimal.zig b/lib/std/fmt/parse_float/decimal.zig
@@ -63,7 +63,7 @@ pub fn Decimal(comptime T: type) type {
         pub const max_digits_without_overflow = if (MantissaT == u64) 19 else 38;
         pub const decimal_point_range = if (MantissaT == u64) 2047 else 32767;
         pub const min_exponent = if (MantissaT == u64) -324 else -4966;
-        pub const max_exponent = if (MantissaT == u64) 310 else 4933;
+        pub const max_exponent = if (MantissaT == u64) 310 else 4934;
         pub const max_decimal_digits = if (MantissaT == u64) 18 else 37;
 
         /// The number of significant digits in the decimal.
diff --git a/lib/std/fmt/parse_float/parse_float.zig b/lib/std/fmt/parse_float/parse_float.zig
@@ -5,8 +5,6 @@ const convertEiselLemire = @import("convert_eisel_lemire.zig").convertEiselLemir
 const convertSlow = @import("convert_slow.zig").convertSlow;
 const convertHex = @import("convert_hex.zig").convertHex;
 
-const optimize = true;
-
 pub const ParseFloatError = error{
     InvalidCharacter,
 };
@@ -41,25 +39,23 @@ pub fn parseFloat(comptime T: type, s: []const u8) ParseFloatError!T {
         return convertHex(T, n);
     }
 
-    if (optimize) {
-        if (convertFast(T, n)) |f| {
-            return f;
-        }
+    if (convertFast(T, n)) |f| {
+        return f;
+    }
 
-        if (T == f16 or T == f32 or T == f64) {
-            // If significant digits were truncated, then we can have rounding error
-            // only if `mantissa + 1` produces a different result. We also avoid
-            // redundantly using the Eisel-Lemire algorithm if it was unable to
-            // correctly round on the first pass.
-            if (convertEiselLemire(T, n.exponent, n.mantissa)) |bf| {
-                if (!n.many_digits) {
+    if (T == f16 or T == f32 or T == f64) {
+        // If significant digits were truncated, then we can have rounding error
+        // only if `mantissa + 1` produces a different result. We also avoid
+        // redundantly using the Eisel-Lemire algorithm if it was unable to
+        // correctly round on the first pass.
+        if (convertEiselLemire(T, n.exponent, n.mantissa)) |bf| {
+            if (!n.many_digits) {
+                return bf.toFloat(T, n.negative);
+            }
+            if (convertEiselLemire(T, n.exponent, n.mantissa + 1)) |bf2| {
+                if (bf.eql(bf2)) {
                     return bf.toFloat(T, n.negative);
                 }
-                if (convertEiselLemire(T, n.exponent, n.mantissa + 1)) |bf2| {
-                    if (bf.eql(bf2)) {
-                        return bf.toFloat(T, n.negative);
-                    }
-                }
             }
         }
     }