[bug] `PermissionError` during `conan cache restore` when extracting read-only files on Windows

[danimtb]

Describe the bug

Conan Version: 2.31.1
Operating System: Windows 11
Python Version: Python 3.12.0

---

Description

When running conan cache restore on Windows, the extraction fails with PermissionError: [Errno 13] Permission denied.

This occurs when an archive contains read-only files—such as MSYS2 system files (etc/pki/ca-trust/...), Git pack files (.git/objects/pack/*.idx), or pre-existing cache files marked as read-only. Python’s tarfile.extractall() is unable to open these existing files in write mode (wb) on Windows, causing the restore process to crash.

---

Logs / Traceback

λ conan cache restore ros-cache.tgz
ERROR: Traceback (most recent call last):
  File "C:\python\Lib\site-packages\conan\cli\cli.py", line 297, in main
    cli.run(args)
  File "C:\python\Lib\site-packages\conan\cli\cli.py", line 194, in run
    command.run(self._conan_api, args[0][1:])
  File "C:\python\Lib\site-packages\conan\cli\command.py", line 213, in run
    sub.run(conan_api, parser, *args)
  File "C:\python\Lib\site-packages\conan\cli\command.py", line 231, in run
    info = self._method(conan_api, parent_parser, self._parser, *args)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\python\Lib\site-packages\conan\cli\commands\cache.py", line 335, in cache_restore
    package_list = conan_api.cache.restore(path)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\python\Lib\site-packages\conan\api\subapi\cache.py", line 399, in restore
    the_tar.extractall(path=cache_folder)
  File "C:\python\Lib\tarfile.py", line 2268, in extractall
    self._extract_one(tarinfo, path, set_attrs=not tarinfo.isdir(),
  File "C:\python\Lib\tarfile.py", line 2335, in _extract_one
    self._handle_fatal_error(e)
  File "C:\python\Lib\tarfile.py", line 2331, in _extract_one
    self._extract_member(tarinfo, os.path.join(path, tarinfo.name),
  File "C:\python\Lib\tarfile.py", line 2414, in _extract_member
    self.makefile(tarinfo, targetpath)
  File "C:\python\Lib\tarfile.py", line 2459, in makefile
    with bltn_open(targetpath, "wb") as target:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
PermissionError: [Errno 13] Permission denied: 'C:\\.c2-reloc\\p\\msys2f91ae1bf3386a\\p\\bin\\msys64\\etc\\pki\\ca-trust\\extracted\\java\\cacerts'

ERROR: [Errno 13] Permission denied: 'C:\\.c2-reloc\\p\\msys2f91ae1bf3386a\\p\\bin\\msys64\\etc\\pki\\ca-trust\\extracted\\java\\cacerts'

---

Expected Behavior

conan cache restore should handle pre-existing read-only attributes on Windows and extract files successfully without raising a PermissionError.

---

Possible Fix

In conan/api/subapi/cache.py, adjust destination file permissions before extraction to ensure they are writable:

import os
import stat

# ... inside cache restore logic ...
for member in the_tar.getmembers():
    target_path = os.path.join(cache_folder, member.name)
    if os.path.exists(target_path):
        os.chmod(target_path, stat.S_IWRITE | stat.S_IREAD)

the_tar.extractall(path=cache_folder)

How to reproduce it

1. Create a Conan package that includes read-only files (e.g., an MSYS2 environment or a Git repository source directory).
2. Save the cache using conan cache save ... -f ros-cache.tgz.
3. Restore the archive on a Windows system using conan cache restore ros-cache.tgz.

[danimtb]

As per @memsharded comment at #20254 (comment)

Now the conan cache restore extracts all the archives in the tar file, regardless of the content in the compressed file and the contents in the current cache.

Beyond the permissions issue, there are other things that could be improved:

• There is no crash protection: If a restore command dies mid-extraction while overwriting an existing package, it will leave a half-written folder that the DB still considers valid. We should consider marking folders as dirty if they are interrupted and extraction can be discarded later.

• The paths in the tar file are different from the destination cache, so restore unzips the package into the wrong folder and then moves it to the right one. This behavior lead into different issues

  • The empty build folders like empty b/<hash>/d in the cache forever, unknown to the DB and to every conan cache command.
  • The pkglist.json is extracted into the root of the store and never removed.
  • If a package already occupies the destination, it must be deleted first to free the spot. That rmdir() is the second PermissionError in this issue. Unzipping straight into the final folder would mean nothing to delete.
  • The delete happens before the move, so a crash in between leaves the DB claiming a package that no longer exists on disk. A crash after unzipping but before moving strands the files in b/<hash>/ where nothing will ever find them.

So maybe we should resolve the DB layout before extracting, skip revisions already present (refreshing timestamp/LRU), skip per top-level subfolder (e, es, s, p) so partial archives still add missing sources, extract straight into the final layout under a dirty marker, and keep destructive overwrite only for metadata.

[danimtb]

I have opened a new PR proposing a refactor that addresses the points in my previous comment #20270