Remove packages based on dependencies

conda_subchannel/cli.py

@@ -112,10 +112,25 @@ def configure_parser(parser: argparse.ArgumentParser):
         action="append",
         help="Remove packages matching this spec. Can be used several times.",
     )
+    parser.add_argument(
+        "--remove-depends",
+        action="append",
+        help="Remove packages that have a dependency on this pattern. Can be used several times.",
+    )
 
 
 def execute(args: argparse.Namespace) -> int:
-    if not any([args.after, args.before, args.keep, args.remove, args.keep_tree, args.prune]):
+    if not any(
+        [
+            args.after,
+            args.before,
+            args.keep,
+            args.remove,
+            args.keep_tree,
+            args.prune,
+            args.remove_depends,
+        ]
+    ):
         raise ArgumentError("Please provide at least one filter.")
 
     with Spinner("Syncing source channel"):
@@ -130,6 +145,7 @@ def execute(args: argparse.Namespace) -> int:
         "subdir_datas": subdir_datas,
         "specs_to_keep": args.keep,
         "specs_to_remove": args.remove,
+        "specs_depends_to_remove": args.remove_depends,
         "specs_to_prune": args.prune,
         "trees_to_keep": args.keep_tree,
         "after": args.after,

conda_subchannel/core.py

@@ -5,6 +5,9 @@
 import hashlib
 import json
 import logging
+import fnmatch as _fnmatch
+import re
+from functools import lru_cache
 from datetime import datetime, timezone
 from pathlib import Path
 from typing import TYPE_CHECKING
@@ -32,6 +35,47 @@
 log = logging.getLogger(f"conda.{__name__}")
 
 
+# Copied from https://github.com/conda-forge/conda-forge-repodata-patches-feedstock/blob/2daa1f620d750a16849d9bfe340885ac1ad39dc5/recipe/patch_yaml_utils.py#L47C1-L78C35
+@lru_cache(maxsize=32768)
+def fnmatch(name, pat):
+    """Test whether FILENAME matches PATTERN with custom
+    allowed optional space via '?( *)'.
+
+    This is useful to match single names with or without a version
+    but not other packages.
+
+    Here are various cases to illustrate how this works:
+
+      - 'numpy*' will match 'numpy', 'numpy >=1', and 'numpy-blah >=10'.
+      - 'numpy?( *)' will match only 'numpy', 'numpy >=1'.
+      - 'numpy' only matches 'numpy'
+
+    **doc string below is from python stdlib**
+
+    Patterns are Unix shell style:
+
+    *       matches everything
+    ?       matches any single character
+    [seq]   matches any character in seq
+    [!seq]  matches any char not in seq
+
+    An initial period in FILENAME is not special.
+    Both FILENAME and PATTERN are first case-normalized
+    if the operating system requires it.
+    If you don't want this, use fnmatchcase(FILENAME, PATTERN).
+    """
+    name = os.path.normcase(name)
+    pat = os.path.normcase(pat)
+    match = _fnmatch_build_re(pat)
+    return match(name) is not None
+
+
+@lru_cache(maxsize=32768)
+def _fnmatch_build_re(pat):
+    repat = "(?s:\\ .*)?".join([_fnmatch.translate(p)[:-2] for p in pat.split("?( *)")]) + "\\Z"
+    return re.compile(repat).match
+
+
 def _fetch_channel(channel, subdirs=None, repodata_fn=REPODATA_FN):
     def fetch(url):
         subdir_data = SubdirData(Channel(url), repodata_fn=repodata_fn)
@@ -74,13 +118,15 @@ def _reduce_index(
     subdir_datas: Iterable[SubdirData],
     specs_to_keep: Iterable[str | MatchSpec] | None = None,
     specs_to_remove: Iterable[str | MatchSpec] | None = None,
+    specs_depends_to_remove: Iterable[str] | None = None,
     specs_to_prune: Iterable[str | MatchSpec] | None = None,
     trees_to_keep: Iterable[str | MatchSpec] | None = None,
     after: int | None = None,
     before: int | None = None,
 ) -> dict[tuple[str, str], PackageRecord]:
     specs_to_keep = [MatchSpec(spec) for spec in (specs_to_keep or ())]
     specs_to_remove = [MatchSpec(spec) for spec in (specs_to_remove or ())]
+    specs_depends_to_remove = specs_depends_to_remove or ()
     specs_to_prune = [MatchSpec(spec) for spec in (specs_to_prune or ())]
     trees_to_keep = [MatchSpec(spec) for spec in (trees_to_keep or ())]
     if trees_to_keep or specs_to_keep or after is not None or before is not None:
@@ -153,7 +199,14 @@ def _reduce_index(
         for key, record in records.items():
             if spec.match(record):
                 to_remove.add(key)
-
+
+    for spec in specs_depends_to_remove:
+        for key, record in records.items():
+            for dep in record["depends"]:
+                if fnmatch(dep, spec):
+                    to_remove.add(key)
+                    break
+
     for key in to_remove:
         records.pop(key)
 
@@ -206,7 +259,12 @@ def _checksum(path, algorithm, buffersize=65536):
     return hash_impl.hexdigest()
 
 
-def _write_channel_index_html(source_channel: Channel, channel_path: Path, cli_flags: dict[str, Any], served_at: str | None = None):
+def _write_channel_index_html(
+    source_channel: Channel,
+    channel_path: Path,
+    cli_flags: dict[str, Any],
+    served_at: str | None = None,
+):
     templates_dir = Path(__file__).parent / "templates"
     environment = jinja2.Environment(loader=jinja2.FileSystemLoader(templates_dir))
     template = environment.get_template("channel.j2.html")
@@ -313,4 +371,4 @@ def _sortkey_package_filenames(fn: str):
             if field.isdigit():
                 build_number = field
                 break
-    return name, VersionOrder(version), build_number, ext
+    return name, VersionOrder(version), build_number, ext

docs/usage.md

@@ -4,7 +4,7 @@
 
 ```
 $ conda subchannel --help
-usage: conda subchannel -c CHANNEL [--repodata-fn REPODATA_FN] [--base-url BASE_URL] [--output PATH] [--subdir PLATFORM] [--after TIME] [--before TIME] [--keep-tree SPEC] [--keep SPEC] [--remove SPEC] [-h]
+usage: conda subchannel -c CHANNEL [--repodata-fn REPODATA_FN] [--base-url BASE_URL] [--output PATH] [--served-at URL] [--subdir PLATFORM] [--after TIME] [--before TIME] [--keep-tree SPEC] [--keep SPEC] [--prune SPEC] [--remove SPEC] [--remove-depends REMOVE_DEPENDS] [-h]
 
 Create subsets of conda channels thanks to CEP-15 metadata
 
@@ -17,18 +17,20 @@ options:
                         '--channel'. Only needed if the user wants to mirror the required packages
                         separately.
   --output PATH         Directory where the subchannel repodata.json artifacts will be written to.
+  --served-at URL       URL or location where the subchannel files will be eventually served. Used for the HTML output.
   --subdir PLATFORM, --platform PLATFORM
                         Process records for this platform. Defaults to osx-arm64. noarch is always included. Can be used several times.
   --after TIME          Timestamp as ts:<float> or date as YYYY-[MM[-DD[-HH[-MM[-SS]]]]]
   --before TIME         Timestamp as ts:<float> or date as YYYY-[MM[-DD[-HH[-MM[-SS]]]]]
   --keep-tree SPEC      Keep packages matching this spec and their dependencies. Can be used
                         several times.
   --keep SPEC           Keep packages matching this spec only. Can be used several times.
-  --prune SPEC          Remove the distributions of this package name that do not match the
-                        given constraints.
+  --prune SPEC          Remove the distributions of this package name that do not match this spec.
   --remove SPEC         Remove packages matching this spec. Can be used several times.
+  --remove-depends REMOVE_DEPENDS
+                        Remove packages that have a dependency on this pattern. Can be used several times.
   -h, --help            Show this help message and exit.
-  ```
+```
 
 
 ## Filtering algorithm
@@ -40,7 +42,8 @@ In the first phase, we _select_ which records are going to be kept. Everything e
 1. A selection list is built. Records in this list are added if:
   - They match specs in `--keep-tree`, or any of the dependencies in their tree (assessed recursively).
   - They match any of the specs in `--keep`.
-  - Their timestamp is within the limits marked by `--before` and `--after`, when applicable. 
+  - Their timestamp is within the limits marked by `--before` and `--after`, when applicable.
 2. At this point, records that didn't make it to the selection list are removed.
 3. The specs defined `--prune` are processed. Records that have the same name but don't match the spec are removed. Everything else is ignored.
 4. Records matching any of the specs in `--remove` are filtered out.
+5. Records that have dependencies matching any of the patterns in ``--remove-depends`` are filtered out.

tests/test_subchannel.py

@@ -8,6 +8,7 @@
 from conda.models.channel import Channel
 from conda.models.match_spec import MatchSpec
 from conda.exceptions import ArgumentError, DryRunExit, PackagesNotFoundError
+from conda_libmamba_solver.exceptions import LibMambaUnsatisfiableError
 from conda.testing import conda_cli  # noqa
 
 
@@ -322,4 +323,49 @@ def test_pruned_python(conda_cli, tmp_path):
             "--channel",
             channel_path,
             "nodejs",
-        )
+        )
+
+
+def test_remove_depends(conda_cli, tmp_path):
+    spec = "python_abi 3.9.*"
+    channel_path = tmp_path / "channel"
+    out, err, rc = conda_cli(
+        "subchannel",
+        "-c",
+        "conda-forge",
+        "--remove-depends",
+        spec,
+        "--output",
+        channel_path,
+    )
+    print(out)
+    print(err, file=sys.stderr)
+    assert rc == 0
+
+    # This should be solvable, we didn't remove anything for Python 3.10
+    with pytest.raises(DryRunExit):
+        conda_cli(
+            "create",
+            "--dry-run",
+            "-n",
+            "unused",
+            "--override-channels",
+            "--channel",
+            channel_path,
+            "python=3.10",
+            "pyarrow",
+        )
+
+    # This fails as we removed all numpy builds for 3.9
+    with pytest.raises(LibMambaUnsatisfiableError):
+        conda_cli(
+            "create",
+            "--dry-run",
+            "-n",
+            "unused",
+            "--override-channels",
+            "--channel",
+            channel_path,
+            "python=3.9",
+            "pyarrow",
+        )