chore: pre-commit hook to reject oversized files, ignore Testing/

.github/workflows/ci.yml

@@ -102,6 +102,41 @@ jobs:
           cmake --build ../build --target tests -j $(nproc)
           ctest --test-dir ../build -j $(nproc) --output-on-failure
 
+  file-size:
+    name: File size
+    # Only meaningful for a pull request: github.event.pull_request is unset
+    # on a push, and a push's commits are only ever ones already checked here
+    # as part of the PR that introduced them.
+    if: github.event_name == 'pull_request'
+    runs-on: ubuntu-24.04
+    steps:
+      - name: Clone Boost.OpenMethod
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      # Every blob any commit in the PR introduces, not just the files that
+      # differ in the final base...head diff - a commit that adds an
+      # oversized file and a later commit in the same PR that shrinks or
+      # removes it still leaves that blob in history forever once merged.
+      # Same limit and message as the local dev/git-hooks/pre-commit hook
+      # (dev/check-blob-size), so a contributor who skipped installing it
+      # still can't land an oversized file.
+      - name: Check for oversized blobs
+        run: |
+          status=0
+
+          git rev-list "${{ github.event.pull_request.base.sha }}..${{ github.event.pull_request.head.sha }}" |
+          while read -r c; do
+              git diff-tree -r --no-commit-id --diff-filter=ACMR "$c"
+          done | awk '{print $4, $NF}' | sort -u > /tmp/blobs.txt
+
+          while read -r oid path; do
+              dev/check-blob-size "$oid" "$path" || status=1
+          done < /tmp/blobs.txt
+
+          exit "$status"
+
   flat-headers:
     name: Flat headers
     runs-on: ubuntu-24.04

.gitignore

@@ -49,3 +49,8 @@ flat/
 # untracked-but-addable - `git add test` would commit tens of thousands of
 # lines of objects.
 bin/
+
+# CTest output. `ctest` writes Testing/Temporary/{LastTest.log,CTestCostData.txt}
+# relative to the directory it runs in, so this turns up at the root whenever
+# ctest is run from there rather than from build/.
+Testing/

dev/check-blob-size

@@ -0,0 +1,26 @@
+#!/bin/bash
+# Copyright (c) 2017-2026 Jean-Louis Leroy
+# Distributed under the Boost Software License, Version 1.0.
+# See accompanying file LICENSE_1_0.txt
+# or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+# Check a single blob's size against the project's oversized-file limit.
+# Shared by dev/git-hooks/pre-commit (every file staged for a commit) and the
+# "File size" CI job (every blob any commit in a pull request introduces), so
+# the limit and the message live in one place instead of two.
+#
+# Usage: dev/check-blob-size <oid> <path>
+# Exit 0 if the blob is within the limit (or <oid> isn't a blob at all - not
+# this script's problem), 1 with a message on stderr if it's over.
+
+limit=$((256 * 1024))   # 256 KiB
+
+oid="$1"
+path="$2"
+
+size=$(git cat-file -s "$oid") || exit 0
+if [ "$size" -gt "$limit" ]; then
+    printf 'oversized file: %s is %s bytes, over the %s byte limit (blob %s).\n' \
+        "$path" "$size" "$limit" "$oid" >&2
+    exit 1
+fi

dev/git-hooks/pre-commit

@@ -0,0 +1,41 @@
+#!/bin/bash
+# Copyright (c) 2017-2026 Jean-Louis Leroy
+# Distributed under the Boost Software License, Version 1.0.
+# See accompanying file LICENSE_1_0.txt
+# or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+# Reject a commit that stages an oversized file. Bypass with `git commit
+# --no-verify` when a large file really is project content. The "File size"
+# CI job runs the same check (via dev/check-blob-size) server-side, over
+# every blob a pull request's commits introduce, so a large file can't reach
+# develop just because a contributor skipped installing this hook.
+#
+# Not installed automatically - run dev/install-git-hooks once per clone (or
+# per worktree that doesn't share this one's .git/hooks; linked worktrees do).
+#
+# Sizes come from the index (`git cat-file -s :path`, inside
+# dev/check-blob-size), not the working tree, so this is right for
+# `git add -p`, for `git commit <paths>` (git hands the hook a temporary
+# index), and on a repo with no commits yet. --diff-filter=ACMR skips
+# deletions, which have no index entry to size.
+#
+# bash, not sh: `read -d` is a bashism, and dash is /bin/sh here. The redirect
+# from a process substitution keeps the loop out of a subshell, so `status`
+# survives it and every offender gets reported, not just the first. Git runs
+# hooks with the working tree's top level as the current directory, so the
+# plain relative path below reaches the right worktree's own copy of the
+# script even when .git/hooks/pre-commit is a symlink shared from another one
+# (dev/install-git-hooks points it at the worktree it was run from).
+
+status=0
+
+while IFS= read -r -d '' f; do
+    oid=$(git rev-parse ":$f") || continue
+    dev/check-blob-size "$oid" "$f" || status=1
+done < <(git diff --cached --name-only --diff-filter=ACMR -z)
+
+if [ "$status" -ne 0 ]; then
+    echo "pre-commit: unstage it, or commit with --no-verify." >&2
+fi
+
+exit "$status"

dev/install-git-hooks

@@ -0,0 +1,23 @@
+#!/bin/bash
+# Copyright (c) 2017-2026 Jean-Louis Leroy
+# Distributed under the Boost Software License, Version 1.0.
+# See accompanying file LICENSE_1_0.txt
+# or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+# Symlink dev/git-hooks/* into the hooks directory git will actually run -
+# .git/hooks in a plain clone, or the common git dir for a linked worktree, so
+# one run covers every worktree of this clone.
+#
+# Usage: dev/install-git-hooks
+
+set -euo pipefail
+
+root_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+hooks_src="$root_dir/dev/git-hooks"
+hooks_dst="$(cd "$root_dir" && git rev-parse --git-path hooks)"
+
+for hook in "$hooks_src"/*; do
+    name="$(basename "$hook")"
+    ln -sf "$hook" "$hooks_dst/$name"
+    echo "installed $hooks_dst/$name -> $hook"
+done