[python][c++] Correctly manage lifetime of SOMAVFS with SOMAVFSFilebuf (#3976)

* [python] Correctly manage lifetime of `SOMAVFS` with `SOMAVFSFilebuf`

* Update apis/python/tests/test_basic_anndata_io.py

Co-authored-by: John Kerl <kerl.john.r@gmail.com>

* collect vfs

---------

Co-authored-by: John Kerl <kerl.john.r@gmail.com>

Author: nguyenv

apis/python/src/tiledbsoma/soma_vfs.cc

@@ -33,12 +33,12 @@ class SOMAVFS : public tiledb::VFS {
 class SOMAVFSFilebuf : public tiledb::impl::VFSFilebuf {
    private:
     std::streamsize offset_ = 0;
-    SOMAVFS vfs_;
+    std::shared_ptr<SOMAVFS> vfs_;
     std::ios::openmode openmode_;
 
    public:
-    SOMAVFSFilebuf(const VFS& vfs)
-        : tiledb::impl::VFSFilebuf(vfs)
+    SOMAVFSFilebuf(std::shared_ptr<SOMAVFS> vfs)
+        : tiledb::impl::VFSFilebuf(*vfs)
         , vfs_(vfs){};
 
     SOMAVFSFilebuf* open(const std::string& uri, std::ios::openmode openmode) {
@@ -59,7 +59,7 @@ class SOMAVFSFilebuf : public tiledb::impl::VFSFilebuf {
         } else if (whence == 1) {
             offset_ += seekoff(offset, std::ios::cur, std::ios::in);
         } else if (whence == 2) {
-            offset_ = vfs_.file_size(get_uri()) -
+            offset_ = vfs_->file_size(get_uri()) -
                       seekoff(offset, std::ios::end, std::ios::in);
         } else {
             TPY_ERROR_LOC(
@@ -124,15 +124,15 @@ class SOMAVFSFilebuf : public tiledb::impl::VFSFilebuf {
 };
 
 void load_soma_vfs(py::module& m) {
-    py::class_<SOMAVFS>(m, "SOMAVFS")
+    py::class_<SOMAVFS, std::shared_ptr<SOMAVFS>>(m, "SOMAVFS")
         .def(
             py::init([](std::shared_ptr<SOMAContext> context) {
                 return SOMAVFS(*context->tiledb_ctx());
             }),
             "ctx"_a);
 
     py::class_<SOMAVFSFilebuf>(m, "SOMAVFSFilebuf")
-        .def(py::init<const SOMAVFS&>())
+        .def(py::init<const std::shared_ptr<SOMAVFS>&>())
         .def(
             "open",
             [](SOMAVFSFilebuf& buf, const std::string& uri) {

apis/python/tests/test_basic_anndata_io.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import gc
 import json
 import random
 import tempfile
@@ -1557,3 +1558,16 @@ def test_from_anndata_byteorder_63459(tmp_path, conftest_pbmc_small):
     with tiledbsoma.Experiment.open(exp_uri) as E:
         new_ad = tiledbsoma.io.to_anndata(E, "RNA", X_layer_name="data")
         assert ad.uns["X"] == new_ad.uns["X"]
+
+
+def test_vfs_lifetime_65831():
+    # https://app.shortcut.com/tiledb-inc/story/65831/python-c-segv-memory-issues-in-somavfs-somavfsfilebuf
+    context = tiledbsoma.SOMATileDBContext()
+    vfs = tiledbsoma.pytiledbsoma.SOMAVFS(context.native_context)
+    fb = tiledbsoma.pytiledbsoma.SOMAVFSFilebuf(vfs).open(
+        str(TESTDATA / "pbmc-small.h5ad")
+    )
+    del vfs
+    gc.collect()  # Make sure that vfs is freed
+    fb.read(100)  # Implicitly ensure that read does not segfault
+    fb.close()