Merge branch 'main' of https://github.com/fastfloat/fast_float

Author: IRainman

.github/workflows/on-release.yml

@@ -17,7 +17,7 @@ jobs:
 
     steps:
       - uses: actions/checkout@v7
-      
+
       - name: Amalgamate fast_float.h
         run: |
           mkdir build
@@ -29,13 +29,13 @@ jobs:
           cp tests/string_test.cpp build/
           cd build
           g++ string_test.cpp
+          ./a.out
 
       - name: Upload Release Asset
-        uses: actions/upload-release-asset@v1
+        # --clobber replaces an asset that was already attached by hand, which
+        # would otherwise make the upload fail.
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        with:
-          upload_url: ${{ github.event.release.upload_url }}
-          asset_path: build/fast_float/fast_float.h
-          asset_name: fast_float.h
-          asset_content_type: text/plain
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release upload "${{ github.event.release.tag_name }}" \
+            build/fast_float/fast_float.h --clobber

.github/workflows/release.yml

@@ -0,0 +1,104 @@
+name: Release
+
+# Manually triggered. Bumps the version (patch, minor or major), regenerates
+# the amalgamated fast_float.h, runs the tests, commits the version bump to
+# main, tags it and publishes a GitHub release with fast_float.h attached.
+#
+# Releases created with GITHUB_TOKEN do not trigger other workflows, so
+# on-release.yml does not run for releases made here.
+
+'on':
+  workflow_dispatch:
+    inputs:
+      bump:
+        description: 'Which part of the version to bump'
+        required: true
+        default: 'patch'
+        type: choice
+        options:
+          - patch
+          - minor
+          - major
+
+permissions:
+  contents: write
+
+concurrency:
+  group: release
+  cancel-in-progress: false
+
+jobs:
+  release:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          ref: main
+          fetch-depth: 0
+
+      - name: Compute new version
+        id: version
+        run: |
+          current=$(git describe --abbrev=0 --tags)
+          current=${current#v}
+          IFS=. read -r major minor patch <<< "$current"
+          case "${{ inputs.bump }}" in
+            major) major=$((major + 1)); minor=0; patch=0 ;;
+            minor) minor=$((minor + 1)); patch=0 ;;
+            patch) patch=$((patch + 1)) ;;
+            *) echo "unknown bump type"; exit 1 ;;
+          esac
+          new="$major.$minor.$patch"
+          echo "current=$current" >> "$GITHUB_OUTPUT"
+          echo "new=$new" >> "$GITHUB_OUTPUT"
+          echo "Releasing $current -> $new (${{ inputs.bump }})"
+
+      - name: Bump version files and amalgamate
+        run: |
+          new="${{ steps.version.outputs.new }}"
+          IFS=. read -r major minor patch <<< "$new"
+          python3 script/release.py "$new"
+          rm -f CMakeLists.txt.bak README.md.bak MODULE.bazel.bak \
+                include/fast_float/float_common.h.bak
+          grep -q "project(fast_float VERSION $new LANGUAGES CXX)" CMakeLists.txt
+          grep -q "#define FASTFLOAT_VERSION_MAJOR $major$" include/fast_float/float_common.h
+          grep -q "#define FASTFLOAT_VERSION_MINOR $minor$" include/fast_float/float_common.h
+          grep -q "#define FASTFLOAT_VERSION_PATCH $patch$" include/fast_float/float_common.h
+          grep -q "version = \"$new\"," MODULE.bazel
+          grep -q "releases/download/v$new/fast_float.h" README.md
+          grep -q "#define FASTFLOAT_VERSION_PATCH $patch$" fast_float.h
+
+      - name: Test amalgamation
+        run: |
+          mkdir -p amalgamation/fast_float
+          cp fast_float.h amalgamation/fast_float/
+          g++ -O2 -std=c++17 -Iamalgamation -o amalgamation/string_test tests/string_test.cpp
+          ./amalgamation/string_test
+
+      - name: Build and test
+        run: |
+          cmake -B build -DFASTFLOAT_TEST=ON -DCMAKE_CXX_FLAGS='-Werror -Wundef'
+          cmake --build build --parallel
+          ctest --test-dir build --output-on-failure
+
+      - name: Commit, tag and push
+        run: |
+          new="${{ steps.version.outputs.new }}"
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add CMakeLists.txt README.md MODULE.bazel include/fast_float/float_common.h
+          git commit -m "Version $new"
+          git tag -a "v$new" -m "version $new"
+          git push origin main
+          git push origin "v$new"
+
+      - name: Create GitHub release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          new="${{ steps.version.outputs.new }}"
+          gh release create "v$new" \
+            --title "Version $new" \
+            --generate-notes \
+            --notes-start-tag "v${{ steps.version.outputs.current }}" \
+            fast_float.h

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.14)
 
 
-project(fast_float VERSION 8.3.0 LANGUAGES CXX)
+project(fast_float VERSION 8.3.2 LANGUAGES CXX)
 set(FASTFLOAT_CXX_STANDARD 11 CACHE STRING "the C++ standard to use for fastfloat")
 set(CMAKE_CXX_STANDARD ${FASTFLOAT_CXX_STANDARD})
 option(FASTFLOAT_TEST "Enable tests" OFF)

MODULE.bazel

@@ -2,7 +2,7 @@
 
 module(
     name = "fast_float",
-    version = "8.2.4",
+    version = "8.3.2",
     compatibility_level = 8,
 )
 

README.md

@@ -702,7 +702,7 @@ sufficiently recent version of CMake (3.11 or better at least):
 FetchContent_Declare(
   fast_float
   GIT_REPOSITORY https://github.com/irainman/fast_float.git
-  GIT_TAG tags/v8.3.0
+  GIT_TAG tags/v8.3.2
   GIT_SHALLOW TRUE)
 
 FetchContent_MakeAvailable(fast_float)
@@ -718,7 +718,7 @@ You may also use [CPM](https://github.com/cpm-cmake/CPM.cmake), like so:
 CPMAddPackage(
   NAME fast_float
   GITHUB_REPOSITORY "irainman/fast_float"
-  GIT_TAG v8.3.0)
+  GIT_TAG v8.3.2)
 ```
 
 ## Using as single header
@@ -730,7 +730,7 @@ if desired as described in the command line help.
 
 You may directly download automatically generated single-header files:
 
-<https://github.com/irainman/fast_float/releases/download/v8.3.0/fast_float.h>
+<https://github.com/irainman/fast_float/releases/download/v8.3.2/fast_float.h>
 
 ## Benchmarking
 

include/fast_float/ascii_number.h

@@ -30,10 +30,14 @@ fastfloat_inline FASTFLOAT_CONSTEVAL bool has_simd_opt() noexcept {
 // Next function can be micro-optimized, but compilers are entirely
 // able to optimize it well.
 template <typename UC>
-fastfloat_inline FASTFLOAT_CONSTEXPR14 bool is_integer(UC c) noexcept {
-  const auto d = c - UC('0');
-  // UC may be signed.
-  return d >= 0 && d <= 9;
+fastfloat_inline constexpr bool is_integer(UC c) noexcept {
+  // UC can be signed (wchar_t is a signed int on Linux and macOS), and the
+  // subtraction is then promoted to int and overflows for code units near the
+  // bottom of the range. Subtracting in the unsigned type wraps instead, which
+  // selects the same code units, as ch_to_digit already does.
+  using UnsignedUC = typename std::make_unsigned<UC>::type;
+  return static_cast<UnsignedUC>(static_cast<UnsignedUC>(c) -
+                                 static_cast<UnsignedUC>(UC('0'))) <= 9u;
 }
 
 #if FASTFLOAT_IS_BIG_ENDIAN
@@ -470,9 +474,12 @@ enum class parse_error : uint_fast8_t {
 
 template <typename UC> struct parsed_number_string_t {
   FASTFLOAT_NO_UNIQUE_ADDRESS am_mant_t mantissa;
-  FASTFLOAT_NO_UNIQUE_ADDRESS UC const *lastmatch;
   FASTFLOAT_NO_UNIQUE_ADDRESS am_pow_t exponent;
+  FASTFLOAT_NO_UNIQUE_ADDRESS UC const *lastmatch;
 
+  // The field order matters: placing 'error' next to the booleans avoids
+  // padding, keeping the struct at 64 bytes on 64-bit systems instead of 72.
+  // See https://github.com/fastfloat/fast_float/issues/418
   FASTFLOAT_NO_UNIQUE_ADDRESS parse_error error;
 #ifndef FASTFLOAT_ONLY_POSITIVE_C_NUMBER_WO_INF_NAN
   FASTFLOAT_NO_UNIQUE_ADDRESS bool negative;

include/fast_float/float_common.h

@@ -18,7 +18,7 @@
 
 #define FASTFLOAT_VERSION_MAJOR 8
 #define FASTFLOAT_VERSION_MINOR 3
-#define FASTFLOAT_VERSION_PATCH 0
+#define FASTFLOAT_VERSION_PATCH 2
 
 #define FASTFLOAT_STRINGIZE_IMPL(x) #x
 #define FASTFLOAT_STRINGIZE(x) FASTFLOAT_STRINGIZE_IMPL(x)

include/fast_float/parse_number.h

@@ -588,7 +588,9 @@ FASTFLOAT_CONSTEXPR20
                         int const decimal_exponent) noexcept {
 #ifndef FASTFLOAT_ONLY_POSITIVE_C_NUMBER_WO_INF_NAN
   const auto is_negative = mantissa < 0;
-  const auto m = static_cast<am_mant_t>(is_negative ? -mantissa : mantissa);
+  // Negate as unsigned: -mantissa overflows for the most negative int64_t.
+  const auto m = is_negative ? am_mant_t(0) - static_cast<am_mant_t>(mantissa)
+                                 : static_cast<am_mant_t>(mantissa);
 #else
   FASTFLOAT_ASSUME(mantissa >= 0);
   const auto m = static_cast<am_mant_t>(mantissa);

script/release.py

@@ -145,6 +145,18 @@ def toversionstring(major, minor, rev):
 
 print(f"{versionfile} modified")
 
+bazelfile = f"{maindir}{os.sep}MODULE.bazel"
+
+for line in fileinput.input(bazelfile, inplace=1, backup=".bak"):
+    line = re.sub(
+        r'^(\s*version = )"\d+\.\d+\.\d+",',
+        rf'\g<1>"{newversionstring}",',
+        line.rstrip(),
+    )
+    print(line)
+
+print(f"modified {bazelfile}, a backup was made")
+
 readmefile = f"{maindir}{os.sep}README.md"
 
 for line in fileinput.input(readmefile, inplace=1, backup=".bak"):
@@ -173,6 +185,7 @@ def toversionstring(major, minor, rev):
 
 if cp.returncode != 0:
     print("Failed to run amalgamate")
+    sys.exit(cp.returncode)
 else:
     print("amalgamate.py ran successfully")
     print(f"You should upload {maindir}{os.sep}fast_float.h")

tests/basictest.cpp

@@ -80,6 +80,12 @@ constexpr std::string_view const round_name(int d) {
   }
 }
 
+// Guard against padding regressions that hurt performance.
+// See https://github.com/fastfloat/fast_float/issues/418
+static_assert(sizeof(void *) != 8 ||
+                  sizeof(fast_float::parsed_number_string_t<char>) <= 64,
+              "parsed_number_string_t should fit in 64 bytes");
+
 #define FASTFLOAT_STR(x) #x
 #define SHOW_DEFINE(x) printf("%s='%s'\n", #x, FASTFLOAT_STR(x))
 
@@ -2475,6 +2481,10 @@ TEST_CASE("integer_times_pow10") {
     all::verify_integer_times_pow10(12345678901234567890ull, -42);
     all::verify_integer_times_pow10(std::numeric_limits<int64_t>::max(), 42);
     all::verify_integer_times_pow10(std::numeric_limits<int64_t>::max(), -42);
+    // the most negative int64_t has no positive counterpart to negate into
+    all::verify_integer_times_pow10(std::numeric_limits<int64_t>::min(), 0);
+    all::verify_integer_times_pow10(std::numeric_limits<int64_t>::min(), 42);
+    all::verify_integer_times_pow10(std::numeric_limits<int64_t>::min(), -42);
     all::verify_integer_times_pow10(std::numeric_limits<uint64_t>::max(), 42);
     all::verify_integer_times_pow10(std::numeric_limits<uint64_t>::max(), -42);
   }