feat: support raw symlinks from declare_symlink() (#930)

Using the `ctx.actions.declare_symlink()` API, rules can create explicit
symlinks to arbitrary
paths, which Bazel will keep as symlinks. The packaging logic treats
these
as regular files, so tries to read through them and store the content
they point to. This
is incorrect for two reasons: they are supposed to be left as symlinks
and may be dangling
symlinks from the perspective of packaging.

To fix, introduce a new type of file entry: `raw_symlink`. These entries
mean the `src` is
a symlink whose stored path should be stored into the destination
verbatim, also as a symlink.
Such artifacts are identified using `File.is_symlink`, a Bazel 8+ API.

Work towards #929

Author: tonyaiuto

pkg/private/manifest.py

@@ -19,6 +19,7 @@
 
 
 # These must be kept in sync with the declarations in private/pkg_files.bzl
+ENTRY_IS_RAW_LINK = "raw_symlink"  # Entry is a file: take content from <src>
 ENTRY_IS_FILE = "file"  # Entry is a file: take content from <src>
 ENTRY_IS_LINK = "symlink"  # Entry is a symlink: dest -> <src>
 ENTRY_IS_DIR = "dir"  # Entry is an empty dir

pkg/private/pkg_files.bzl

@@ -45,6 +45,7 @@ load(
     "get_repo_mapping_manifest",
 )
 
+ENTRY_IS_RAW_LINK = "raw_symlink"  # Entry is a symlink kept as-is
 ENTRY_IS_FILE = "file"  # Entry is a file: take content from <src>
 ENTRY_IS_LINK = "symlink"  # Entry is a symlink: dest -> <src>
 ENTRY_IS_DIR = "dir"  # Entry is an empty dir
@@ -444,9 +445,16 @@ def add_from_default_info(
                 d_path = mapping_context.path_mapper(base_path + "/" + rf.short_path)
                 fmode = "0755" if rf == the_executable else mapping_context.default_mode
                 _check_dest(mapping_context.content_map, d_path, rf, src.label, mapping_context.allow_duplicates_with_different_content)
+                if hasattr(rf, "is_symlink") and rf.is_symlink:  # File.is_symlink is Bazel 8+
+                    entry_type = ENTRY_IS_RAW_LINK
+                elif rf.is_directory:
+                    entry_type = ENTRY_IS_TREE
+                else:
+                    entry_type = ENTRY_IS_FILE
+
                 mapping_context.content_map[d_path] = _DestFile(
                     src = rf,
-                    entry_type = ENTRY_IS_TREE if rf.is_directory else ENTRY_IS_FILE,
+                    entry_type = entry_type,
                     origin = src.label,
                     mode = fmode,
                     user = mapping_context.default_user,
@@ -531,9 +539,15 @@ def add_single_file(mapping_context, dest_path, src, origin, mode = None, user =
     """
     dest = dest_path.strip("/")
     _check_dest(mapping_context.content_map, dest, src, origin, mapping_context.allow_duplicates_with_different_content)
+
+    if hasattr(src, "is_symlink") and src.is_symlink:  # File.is_symlink is Bazel 8+
+        entry_type = ENTRY_IS_RAW_LINK
+    else:
+        entry_type = ENTRY_IS_FILE
+
     mapping_context.content_map[dest] = _DestFile(
         src = src,
-        entry_type = ENTRY_IS_FILE,
+        entry_type = entry_type,
         origin = origin,
         mode = mode,
         user = user or mapping_context.default_user,

pkg/private/tar/build_tar.py

@@ -332,6 +332,8 @@ def add_manifest_entry(self, entry, file_attributes):
         attrs['ids'] = (entry.uid, entry.uid)
     if entry.type == manifest.ENTRY_IS_LINK:
       self.add_link(entry.dest, entry.src, **attrs)
+    elif entry.type == manifest.ENTRY_IS_RAW_LINK:
+      self.add_link(entry.dest, os.readlink(entry.src), **attrs)
     elif entry.type == manifest.ENTRY_IS_DIR:
       self.add_empty_dir(self.normalize_path(entry.dest), **attrs)
     elif entry.type == manifest.ENTRY_IS_TREE:

pkg/verify_archive_test_main.py.tpl

@@ -109,7 +109,7 @@ class VerifyArchiveTest(unittest.TestCase):
   def verify_links(self, verify_links):
     for link, target in verify_links.items():
       if link not in self.paths:
-        self.fail('Required link (%s) is not in the archive' % link)
+        self.fail('Required link (%s) is not in the archive, found %s' % (link, self.paths))
       if self.links[link] != target:
         self.fail('link (%s) points to the wrong place. Expected (%s) got (%s)' %
             (link, target, self.links[link]))

tests/tar/BUILD

@@ -24,6 +24,7 @@ load("//pkg:verify_archive.bzl", "verify_archive_test")
 load("//pkg/private/tar:tar.bzl", "SUPPORTED_TAR_COMPRESSIONS", "pkg_tar")
 load("//tests:my_package_name.bzl", "my_package_naming")
 load("//tests/util:defs.bzl", "directory", "fake_artifact", "link_tree")
+load(":defs.bzl", "raw_symlinks")
 
 package(
     default_applicable_licenses = ["//:license"],
@@ -629,6 +630,31 @@ pkg_tar(
 # Test with symlinks
 #
 
+raw_symlinks(name = "raw_symlinks")
+
+pkg_tar(
+    name = "raw_symlinks_tar",
+    srcs = [":raw_symlinks"],
+    include_runfiles = True,
+)
+
+verify_archive_test(
+    name = "raw_symlinks_test",
+    tags = ["no_windows"],
+    target = ":raw_symlinks_tar",
+    # This test checks using relative symlinks, but Bazel always writes absolute
+    # symlinks on windows.
+    # See https://github.com/bazelbuild/bazel/issues/14224
+    target_compatible_with = select({
+        "@platforms//os:windows": ["@platforms//:incompatible"],
+        "//conditions:default": [],
+    }),
+    verify_links = {
+        "raw_symlinks_link1": "link1",
+        "raw_symlinks_link1.runfiles/_main/tests/tar/raw_symlinks_runfile_link": "../runfile_link",
+    },
+)
+
 link_tree(
     name = "node_modules",
     links = {

tests/tar/defs.bzl

@@ -0,0 +1,17 @@
+"""Helpers for testing tar packaging."""
+
+def _raw_symlinks_impl(ctx):
+    link1 = ctx.actions.declare_symlink(ctx.label.name + "_link1")
+    ctx.actions.symlink(output = link1, target_path = "./link1")
+
+    runfile_link = ctx.actions.declare_symlink(ctx.label.name + "_runfile_link")
+    ctx.actions.symlink(output = runfile_link, target_path = "../runfile_link")
+
+    return [DefaultInfo(
+        files = depset([link1]),
+        runfiles = ctx.runfiles([runfile_link]),
+    )]
+
+raw_symlinks = rule(
+    implementation = _raw_symlinks_impl,
+)