Upload h5py/H5PY_USER_GUIDE.txt with huggingface_hub

h5py/H5PY_USER_GUIDE.txt

@@ -0,0 +1,73 @@
+================================================================================
+  H5PY - USER GUIDE (Android Python STB) - Generated by RIMI
+================================================================================
+
+h5py 3.16.0 for Android (CPython 3.12) — read and write HDF5 files from Python.
+Built from the official PyPI sdist against HDF5 1.14.6 compiled from source
+for Android API 24. Both the low-level API (h5py.h5f, h5py.h5d, ...) and the
+high-level API (File, Group, Dataset, attrs) are included, plus the bundled
+test suite (h5py.tests).
+
+WHEELS (install exactly one, matching your device ABI)
+  h5py-3.16.0-cp312-cp312-android_24_x86_64.whl    (emulators)
+  h5py-3.16.0-cp312-cp312-android_24_arm64_v8a.whl (real phones)
+Requires: numpy>=1.21.2 (use the matching NATIVE numpy 2.5.2 wheel).
+
+WHAT IS VENDORED
+  h5py/libhdf5.so + h5py/libhdf5_hl.so — the HDF5 1.14.6 runtime built from
+  source with the NDK (16 KB pages, no RELRO, stripped). h5py/__init__.py
+  preloads both by absolute path on Android before any extension module
+  loads (same pattern as the pyarrow package), so `import h5py` just works.
+  No other native dependency: zlib comes from the Android system (libz.so).
+
+QUICK START
+  import h5py
+  import numpy as np
+  print(h5py.version.version, h5py.version.hdf5_version)  # 3.16.0 1.14.6
+
+  with h5py.File("demo.h5", "w") as f:
+      g = f.create_group("sensors")
+      d = g.create_dataset("temp", data=np.arange(100, dtype="f4"))
+      g.attrs["unit"] = "celsius"
+
+  with h5py.File("demo.h5", "r") as f:
+      print(f["sensors/temp"][...].sum(), f["sensors"].attrs["unit"])
+
+  # no storage permission? use a pure in-memory file:
+  with h5py.File("m.h5", "w", driver="core", backing_store=False) as f:
+      f.create_dataset("d", data=np.arange(10))
+
+COMPRESSION / FILTERS
+  gzip (deflate) is fully supported (system zlib):
+      f.create_dataset("z", data=data, compression="gzip", compression_opts=6)
+  shuffle + fletcher32 also work. SZIP and ROS3 (S3) drivers are NOT built
+  into this package (kept dependency-free); requesting them raises a clean
+  error, everything else is unaffected.
+
+ANDROID NOTES
+  - First import preloads ~4 MB of native HDF5; on low-RAM devices allow a
+    moment on first `import h5py`.
+  - Store files under the app's own directories (e.g. the Scripts folder or
+    getFilesDir()); shared storage may need extra permissions.
+  - Threading: the high-level API serialises HDF5 calls with its own lock
+    (OBJECTS_USE_LOCKING); the C library itself is built without the
+    experimental HDF5 threadsafe mode, same as recommended upstream.
+  - h5py 3.16 needs numpy>=1.21.2 at runtime; _npystrings needs numpy>=2
+    (numpy 2.5.2 is what these wheels were built and tested against).
+
+TESTED FEATURES (host-validated 16/16 against HDF5 1.14.6; device script
+Test_H5py.py repeats the same flow on device)
+  import + version wiring, groups, int/float/compound/vlen-string datasets,
+  file + group attributes, gzip write + read-back, chunked resizable
+  datasets, core-driver in-memory files, temp-file roundtrip.
+
+TROUBLESHOOTING
+  "numpy import" failure  -> install the NATIVE numpy wheel first.
+  "ImportError: libhdf5.so" -> reinstall this wheel (the two vendored .so
+    must sit next to __init__.py inside h5py/); do not rename them.
+  Version-mismatch warning -> should never appear (runtime == built-against
+    1.14.6); if it does, a foreign libhdf5 is shadowing the vendored one.
+
+================================================================================
+  Generated by RIMI
+================================================================================