Merge pull request #211 from freedomofpress/196-reproducible-wheels

Makes wheel builds reproducible, with tests

Author: kushaldas

.circleci/config.yml

@@ -281,14 +281,15 @@ common-steps:
 version: 2.1
 jobs:
   tests:
-    docker:
-      - image: circleci/python:3.7-buster
+    machine:
+      image: ubuntu-2004:202010-01
     steps:
       - checkout
       - run:
           name: install test requirements and run tests
           command: |
-            virtualenv .venv
+            make install-deps
+            virtualenv -p python3 .venv
             source .venv/bin/activate
             pip install -r test-requirements.txt
             make test

.gitignore

@@ -1,2 +1,3 @@
+tests/__pycache__/
 debhelper-build-stamp
 *.debhelper.log

Makefile

@@ -61,6 +61,10 @@ test: ## Run test suite
 clean: ## Removes all non-version controlled packaging artifacts
 	rm -rf localwheels/*
 
+.PHONY: reprotest
+reprotest: ## Reproducibility test, currently only for wheels
+	pytest -vvs tests/test_reproducible_wheels.py
+
 .PHONY: help
 help: ## Prints this message and exits
 	@printf "Makefile for building SecureDrop Workstation packages\n"

scripts/build-sync-wheels

@@ -2,29 +2,56 @@
 
 import os
 import sys
-import json
 import subprocess
 import tempfile
 import shutil
 import argparse
 
 
+# Set SOURCE_DATE_EPOCH to a predictable value. Using the first
+# commit to the SecureDrop project:
+#
+#   git show -s 62bbe590afd77a6af2dcaed46c93da6e0cf40951 --date=unix
+#
+# which yields 1309379017
+os.environ["SOURCE_DATE_EPOCH"] = "1309379017"
+
+# Force sane umask for reproducibility's sake.
+os.umask(0o022)
+
+# When building wheels, pip defaults to a safe dynamic tmpdir path.
+# Some shared objects include the path from build, so we must make
+# the path predictable across builds.
+WHEEL_BUILD_DIR = "/tmp/pip-wheel-build"
+
+
 def main():
     parser = argparse.ArgumentParser(
         description="Builds and caches sources and wheels"
     )
     parser.add_argument(
         "-p",
-        help="Points to the project dirctory",
+        help="Points to the project directory",
     )
     parser.add_argument(
         "--cache", default="./localwheels", help="Final cache dir"
     )
+    parser.add_argument(
+        "--clobber", action="store_true", default=False,
+        help="Whether to overwrite wheels and source tarballs",
+    )
     args = parser.parse_args()
 
-    if not os.path.exists(args.p):
-        print("Project directory missing {0}.".format(args.p))
-        sys.exit(1)
+    if args.p.startswith("https://"):
+        git_clone_directory = tempfile.mkdtemp(prefix=os.path.basename(args.p))
+        cmd = f"git clone {args.p} {git_clone_directory}".split()
+        subprocess.check_call(cmd)
+        args.p = git_clone_directory
+    else:
+        git_clone_directory = ""
+        if not os.path.exists(args.p):
+            print("Project directory missing {0}.".format(args.p))
+            sys.exit(1)
 
     # Try requirements.txt in the repo root, otherwise try requirements/requirements.txt
     req_path = os.path.join(args.p, "requirements.txt")
@@ -36,6 +63,11 @@ def main():
         print("requirements.txt missing at {0}.".format(req_path))
         sys.exit(3)
 
+    if os.path.exists(WHEEL_BUILD_DIR):
+        shutil.rmtree(WHEEL_BUILD_DIR)
+    else:
+        os.mkdir(WHEEL_BUILD_DIR)
+
     with tempfile.TemporaryDirectory() as tmpdir:
         # The --require-hashes option will be used by default if there are
         # hashes in the requirements.txt file. We specify it anyway to guard
@@ -46,9 +78,9 @@ def main():
             "--no-binary",
             ":all:",
             "--require-hashes",
-            "-d",
+            "--dest",
             tmpdir,
-            "-r",
+            "--requirement",
             req_path,
         ]
         subprocess.check_call(cmd)
@@ -58,11 +90,15 @@ def main():
             "wheel",
             "--no-binary",
             ":all:",
-            "-f",
+            "--find-links",
             tmpdir,
-            "-w",
+            "--progress-bar",
+            "pretty",
+            "--wheel-dir",
             tmpdir,
-            "-r",
+            "--build",
+            WHEEL_BUILD_DIR,
+            "--requirement",
             req_path,
         ]
         subprocess.check_call(cmd)
@@ -76,13 +112,17 @@ def main():
             if name == "requirements.txt":  # We don't need this in cache
                 continue
             if name in cachenames:  # Means all ready in our cache
-                continue
+                if not args.clobber:
+                    continue
 
             # Else copy to cache
             filepath = os.path.join(tmpdir, name)
             shutil.copy(filepath, args.cache, follow_symlinks=True)
             print("Copying {0} to cache {1}".format(name, args.cache))
 
+    if git_clone_directory:
+        shutil.rmtree(git_clone_directory)
+
 
 if __name__ == "__main__":
     main()

scripts/install-deps

@@ -2,6 +2,11 @@
 # Installs required dependencies for building SecureDrop Worsktation packages.
 # Assumes a Debian 10 machine, ideally a Qubes AppVM.
 
+# If running in CI, we need to add the Ubuntu Bionic repo to download dh-virtualenv
+if [[ -v CIRCLE_BUILD_URL ]]; then
+  echo "deb http://archive.ubuntu.com/ubuntu/ bionic universe" | sudo tee -a /etc/apt/sources.list
+fi
+
 sudo apt-get update
 sudo apt-get install  \
     build-essential \
@@ -15,7 +20,8 @@ sudo apt-get install  \
     libssl-dev \
     python3-all \
     python3-pip \
-    python3-setuptools
+    python3-setuptools \
+    reprotest
 
 
 # Inspect the wheel files present locally. If repo was cloned

tests/test_reproducible_wheels.py

@@ -0,0 +1,43 @@
+import pytest
+import subprocess
+
+
+# These are the SDW repositories that we build wheels for.
+REPOS_WITH_WHEELS = [
+    "securedrop-client",
+    "securedrop-log",
+    "securedrop-proxy",
+]
+
+
+def get_repo_root():
+    cmd = "git rev-parse --show-toplevel".split()
+    top_level = subprocess.check_output(cmd).decode("utf-8").rstrip()
+    return top_level
+
+
+@pytest.mark.parametrize("repo_name", REPOS_WITH_WHEELS)
+def test_wheel_builds_are_reproducible(repo_name):
+    """
+    Uses 'reprotest' to confirm that the wheel build process, per repo,
+    is deterministic, i.e. all .whl files are created with the same checksum
+    across multiple builds.
+
+    Explanations of the excluded reproducibility checks:
+
+      * time: breaks HTTPS, so pip calls fail
+      * locales: some locales fail, would be nice to fix, but low priority
+      * kernel: x86_64 is the supported architecure, we don't ship others
+    """
+    repo_url = f"https://github.com/freedomofpress/{repo_name}"
+    cmd = [
+        "reprotest",
+        "-c",
+        f"./scripts/build-sync-wheels -p {repo_url} --clobber",
+        "--vary",
+        "+all, -time, -locales, -kernel",
+        ".",
+        "localwheels/*.whl",
+    ]
+    repo_root = get_repo_root()
+    subprocess.check_call(cmd, cwd=repo_root)