hlfuse: pass the libfuse options to mfusepy as keyword arguments, fix getattr with a file handle

[ThomasWaldmann]

hlfuse.py called hlfuse.FUSE(self, mountpoint, options, foreground=True, use_ino=True), but mfusepy's third positional parameter is raw_fi, not the libfuse option list. Two consequences on master with the mfusepy backend:

• The libfuse mount options were never applied. The mount is rw instead of ro, default_permissions is missing, and user-given -o options like allow_other are silently ignored (checked in /proc/mounts; another user gets "Permission denied" despite -o allow_other). fsname=borgfs only looked right because mfusepy defaults the fsname to the operations class name, which happens to be borgfs.
• Reading a file that takes longer than a second fails with EINVAL. raw_fi was accidentally true, which open/read/release rely on (they use the fuse_file_info struct), but getattr with a file handle received the struct too and used it as a dict key: TypeError: unhashable type → EINVAL. The kernel sends that fgetattr when a read reaches EOF after the attribute timeout (1 s by default), so e.g. cat of any file that takes more than a second to read ends with cat: ...: Invalid argument.

The fix converts the option list to mfusepy keyword arguments, requests raw_fi=True explicitly and unwraps the file handle in getattr.

The regression test mounts an archive, checks via statvfs that the mount is read-only (this fails on master: the options are dropped), reads 4 KiB of a file, sleeps 1.5 s and then does a raw os.read past EOF (this fails on master with OSError: [Errno 22] Invalid argument). It uses os.read deliberately: a buffered f.read() does an fstat first, which refreshes the attributes without a file handle and hides the bug.

Verified on Linux (Debian 13, libfuse 3.17.2, mfusepy 3.1.1): the new test fails on master for both symptoms and passes with the fix, mount_cmds_test.py passes with BORG_FUSE_IMPL=mfusepy, and after the fix /proc/mounts shows ro,...,default_permissions,allow_other for borg mount -o allow_other.

Found while benchmarking #5110 (the long sequential FUSE reads would not complete without this).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.86%. Comparing base (366db25) to head (9056ef4).
⚠️ Report is 29 commits behind head on master.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #10340      +/-   ##
==========================================
+ Coverage   87.80%   87.86%   +0.06%     
==========================================
  Files         103      103              
  Lines       18828    18876      +48     
  Branches     2906     2915       +9     
==========================================
+ Hits        16531    16586      +55     
+ Misses       1598     1589       -9     
- Partials      699      701       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.