Merge pull request #1089 from rd4398/date-range-filtering

feat(bootstrap): add --max-release-age for multiple-versions mode

Author: LalatenduMohanty

e2e/ci_bootstrap_suite.sh

@@ -26,6 +26,7 @@ run_test "bootstrap_prerelease"
 run_test "bootstrap_cache"
 run_test "bootstrap_sdist_only"
 run_test "bootstrap_multiple_versions"
+run_test "bootstrap_max_release_age"
 
 test_section "bootstrap test-mode tests"
 run_test "mode_resolution"

e2e/test_bootstrap_max_release_age.sh

@@ -0,0 +1,98 @@
+#!/bin/bash
+# -*- indent-tabs-mode: nil; tab-width: 2; sh-indentation: 2; -*-
+
+# Test bootstrap with --max-release-age flag
+# Tests that old versions are filtered out by the max release age window
+# and that the filter also applies to build dependencies
+
+SCRIPTDIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+source "$SCRIPTDIR/common.sh"
+
+# certifi PyPI upload timeline (actual upload_time from PyPI JSON API):
+#
+#   certifi 2025.11.12  2025-11-12  (should be filtered — too old)
+#   certifi 2026.1.4    2026-01-04  (should be filtered — too old)
+#   certifi 2026.2.25   2026-02-25  (should be included — recent enough)
+#   certifi 2026.4.22   2026-04-22  (should be included — recent enough)
+#
+# Compute --max-release-age so certifi 2026.2.25 is inside the window
+# but certifi 2026.1.4 is outside. We anchor on certifi 2026.2.25's
+# upload date and add a buffer.
+MAX_AGE=$(python3 -c "
+from datetime import date
+# Age of certifi 2026.2.25 (uploaded 2026-02-25) + 10 day buffer
+age = (date.today() - date(2026, 2, 25)).days + 10
+print(age)
+")
+
+echo "Using --max-release-age=$MAX_AGE"
+
+fromager \
+  --log-file="$OUTDIR/bootstrap.log" \
+  --error-log-file="$OUTDIR/fromager-errors.log" \
+  --sdists-repo="$OUTDIR/sdists-repo" \
+  --wheels-repo="$OUTDIR/wheels-repo" \
+  --work-dir="$OUTDIR/work-dir" \
+  bootstrap \
+  --multiple-versions \
+  --max-release-age="$MAX_AGE" \
+  'certifi>=2025.11,<=2026.5'
+
+# Verify that recent versions were built (within age window)
+echo ""
+echo "Checking for expected versions..."
+for version in 2026.2.25 2026.4.22; do
+  if find "$OUTDIR/wheels-repo/downloads/" -name "certifi-$version-*.whl" | grep -q .; then
+    echo "✓ Found wheel for certifi $version (within max-release-age window)"
+  else
+    echo "✗ Missing wheel for certifi $version"
+    echo "ERROR: certifi $version should be within the max-release-age window"
+    echo ""
+    echo "Found wheels:"
+    find "$OUTDIR/wheels-repo/downloads/" -name 'certifi-*.whl'
+    exit 1
+  fi
+done
+
+# Verify that old versions were filtered out
+echo ""
+echo "Checking that old versions were filtered..."
+UNEXPECTED=""
+for version in 2025.11.12 2026.1.4; do
+  if find "$OUTDIR/wheels-repo/downloads/" -name "certifi-$version-*.whl" | grep -q .; then
+    echo "✗ Found wheel for certifi $version — should have been filtered by max-release-age"
+    UNEXPECTED="$UNEXPECTED $version"
+  else
+    echo "✓ certifi $version correctly filtered out by max-release-age"
+  fi
+done
+
+if [ -n "$UNEXPECTED" ]; then
+  echo ""
+  echo "ERROR: --max-release-age should have excluded:$UNEXPECTED"
+  exit 1
+fi
+
+# Verify that max-release-age filtering was applied (check log)
+echo ""
+echo "Checking log for max-release-age filtering..."
+if grep -q "published within.*days" "$OUTDIR/bootstrap.log"; then
+  echo "✓ Log confirms max-release-age filtering was applied"
+else
+  echo "✗ No max-release-age filtering found in log"
+  exit 1
+fi
+
+# Verify that build dependencies were also resolved within the window
+# setuptools is the build dependency for certifi
+echo ""
+echo "Checking that build dependencies were resolved..."
+if find "$OUTDIR/wheels-repo/downloads/" -name "setuptools-*.whl" | grep -q .; then
+  echo "✓ setuptools was built (build dependency of certifi)"
+else
+  echo "✗ setuptools was not built — build dependency resolution may have failed"
+  exit 1
+fi
+
+echo ""
+echo "SUCCESS: --max-release-age correctly filtered old versions and resolved build dependencies"

e2e/test_bootstrap_multiple_versions.sh

@@ -16,6 +16,14 @@ cat > "$constraints_file" <<EOF
 flit-core>=3.9,<3.12
 EOF
 
+# Compute --max-release-age dynamically: days since tomli 2.0.0 was uploaded
+# to PyPI (2021-12-13) plus a buffer, so the oldest version is always included.
+MAX_AGE=$(python3 -c "
+from datetime import date
+age = (date.today() - date(2021, 12, 13)).days
+print(age + 30)
+")
+
 # Use tomli with a version range that matches exactly 3 versions (2.0.0, 2.0.1, 2.0.2)
 # tomli has no runtime dependencies, making it fast to bootstrap
 # It uses flit-core as build backend, and we allow multiple flit-core versions
@@ -31,6 +39,7 @@ fromager \
   --constraints-file="$constraints_file" \
   bootstrap \
   --multiple-versions \
+  --max-release-age="$MAX_AGE" \
   'tomli>=2.0,<=2.0.2'
 
 # Check that wheels were built

src/fromager/bootstrap_requirement_resolver.py

@@ -173,7 +173,10 @@ def _resolve(
                 sdist_server_url=resolver.PYPI_SERVER_URL,
                 req_type=req_type,
             )
-            return resolver.find_all_matching_from_provider(provider, req)
+            max_age_cutoff = resolver._compute_max_age_cutoff(self.ctx)
+            return resolver.find_all_matching_from_provider(
+                provider, req, max_age_cutoff=max_age_cutoff
+            )
 
     def get_cached_resolution(
         self,

src/fromager/commands/bootstrap.py

@@ -110,6 +110,13 @@ def _get_requirements_from_args(
     default=False,
     help="Bootstrap all matching versions instead of only the highest version",
 )
+@click.option(
+    "--max-release-age",
+    "max_release_age",
+    type=click.IntRange(min=1),
+    default=None,
+    help="Reject package versions published more than this many days ago.",
+)
 @click.argument("toplevel", nargs=-1)
 @click.pass_obj
 def bootstrap(
@@ -121,6 +128,7 @@ def bootstrap(
     skip_constraints: bool,
     test_mode: bool,
     multiple_versions: bool,
+    max_release_age: int | None,
     toplevel: list[str],
 ) -> None:
     """Compute and build the dependencies of a set of requirements recursively
@@ -168,6 +176,13 @@ def bootstrap(
             )
             skip_constraints = True
 
+    if max_release_age is not None:
+        wkctx.set_max_release_age(max_release_age)
+        logger.info(
+            "max release age: rejecting versions older than %d days",
+            max_release_age,
+        )
+
     pre_built = wkctx.settings.list_pre_built()
     if pre_built:
         logger.info("treating %s as pre-built wheels", sorted(pre_built))
@@ -492,6 +507,13 @@ def write_constraints_file(
     default=False,
     help="Bootstrap all matching versions instead of only the highest version",
 )
+@click.option(
+    "--max-release-age",
+    "max_release_age",
+    type=click.IntRange(min=1),
+    default=None,
+    help="Reject package versions published more than this many days ago.",
+)
 @click.argument("toplevel", nargs=-1)
 @click.pass_obj
 @click.pass_context
@@ -506,6 +528,7 @@ def bootstrap_parallel(
     force: bool,
     max_workers: int | None,
     multiple_versions: bool,
+    max_release_age: int | None,
     toplevel: list[str],
 ) -> None:
     """Bootstrap and build-parallel
@@ -533,6 +556,7 @@ def bootstrap_parallel(
         sdist_only=True,
         skip_constraints=skip_constraints,
         multiple_versions=multiple_versions,
+        max_release_age=max_release_age,
         toplevel=toplevel,
     )
 

src/fromager/context.py

@@ -63,6 +63,7 @@ def __init__(
         settings_dir: pathlib.Path | None = None,
         wheel_server_url: str = "",
         cooldown: Cooldown | None = None,
+        max_release_age: datetime.timedelta | None = None,
     ):
         if active_settings is None:
             active_settings = packagesettings.Settings(
@@ -113,6 +114,17 @@ def __init__(
         self._parallel_builds = False
 
         self.cooldown: Cooldown | None = cooldown
+        self._max_release_age: datetime.timedelta | None = max_release_age
+
+    @property
+    def max_release_age(self) -> datetime.timedelta | None:
+        return self._max_release_age
+
+    def set_max_release_age(self, days: int) -> None:
+        """Set the maximum release age in days."""
+        if days < 1:
+            raise ValueError(f"max_release_age must be positive, got {days}")
+        self._max_release_age = datetime.timedelta(days=days)
 
     def enable_parallel_builds(self) -> None:
         self._parallel_builds = True

src/fromager/resolver.py

@@ -105,7 +105,10 @@ def resolve(
         ignore_platform=ignore_platform,
     )
     provider.cooldown = resolve_package_cooldown(ctx, req)
-    results = find_all_matching_from_provider(provider, req)
+    max_age_cutoff = _compute_max_age_cutoff(ctx)
+    results = find_all_matching_from_provider(
+        provider, req, max_age_cutoff=max_age_cutoff
+    )
     return results[0]
 
 
@@ -167,6 +170,24 @@ def resolve_package_cooldown(
     )
 
 
+def _compute_max_age_cutoff(
+    ctx: context.WorkContext,
+) -> datetime.datetime | None:
+    """Compute the cutoff time for max release age filtering.
+
+    Returns the oldest acceptable upload time, or None if disabled.
+    Uses the cooldown's bootstrap_time for consistency across a single run.
+    """
+    if ctx.max_release_age is None:
+        return None
+    bootstrap_time = (
+        ctx.cooldown.bootstrap_time
+        if ctx.cooldown is not None
+        else datetime.datetime.now(datetime.UTC)
+    )
+    return bootstrap_time - ctx.max_release_age
+
+
 def extract_filename_from_url(url: str) -> str:
     """Extract filename from URL and decode it."""
     path = urlparse(url).path
@@ -203,13 +224,22 @@ def ending(self, state: typing.Any) -> None:
 
 
 def find_all_matching_from_provider(
-    provider: BaseProvider, req: Requirement
+    provider: BaseProvider,
+    req: Requirement,
+    max_age_cutoff: datetime.datetime | None = None,
 ) -> list[tuple[str, Version]]:
     """Find all matching candidates from provider without full dependency resolution.
 
     This function collects ALL candidates that match the requirement, rather than
     performing full dependency resolution to find a single best candidate.
 
+    Args:
+        provider: The provider to query for candidates.
+        req: The requirement to match.
+        max_age_cutoff: If set, reject candidates published before this time.
+            If all candidates are older than the cutoff, all are kept and
+            a warning is emitted to avoid empty resolution.
+
     Returns list of (url, version) tuples sorted by version (highest first).
 
     IMPORTANT: This bypasses resolvelib's full resolver to collect all matching
@@ -242,10 +272,47 @@ def find_all_matching_from_provider(
             f"Unable to resolve requirement specifier {req} with constraint {constraint} using {provider_desc}: {original_msg}"
         ) from err
 
+    # Materialize candidates so we can iterate more than once if filtering
+    candidates_list = list(candidates)
+
+    if max_age_cutoff is not None:
+        logger.info(
+            "%s: found %d candidate(s) matching %s",
+            req.name,
+            len(candidates_list),
+            req,
+        )
+        max_age_days = (datetime.datetime.now(datetime.UTC) - max_age_cutoff).days
+        filtered = [
+            c
+            for c in candidates_list
+            if c.upload_time is None or c.upload_time >= max_age_cutoff
+        ]
+        dropped = len(candidates_list) - len(filtered)
+        if dropped:
+            logger.info(
+                "%s: have %d candidate(s) of %s published within %d days",
+                req.name,
+                len(filtered),
+                req,
+                max_age_days,
+            )
+        if filtered:
+            candidates_list = filtered
+        else:
+            logger.warning(
+                "%s: all %d candidate(s) of %s are older than %d days, "
+                "keeping all to avoid empty resolution",
+                req.name,
+                len(candidates_list),
+                req,
+                max_age_days,
+            )
+
     # Convert candidates to list of (url, version) tuples
     # Candidates are sorted by version (highest first) by BaseProvider.find_matches()
     # which calls sorted(candidates, key=attrgetter("version", "build_tag"), reverse=True)
-    return [(candidate.url, candidate.version) for candidate in candidates]
+    return [(c.url, c.version) for c in candidates_list]
 
 
 def get_project_from_pypi(

src/fromager/sources.py

@@ -181,7 +181,10 @@ def resolve_source(
         )
 
         # Get all matching candidates from provider
-        results = resolver.find_all_matching_from_provider(provider, req)
+        max_age_cutoff = resolver._compute_max_age_cutoff(ctx)
+        results = resolver.find_all_matching_from_provider(
+            provider, req, max_age_cutoff=max_age_cutoff
+        )
 
         # Return highest version (first in sorted list)
         url, version = results[0]