diff --git a/.github/workflows/test_wheel.yaml b/.github/workflows/test_wheel.yaml
index ad74ba06..ff03aff0 100644
--- a/.github/workflows/test_wheel.yaml
+++ b/.github/workflows/test_wheel.yaml
@@ -38,7 +38,7 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
           python-version: "3.11"
 
@@ -50,7 +50,7 @@ jobs:
 
       - name: Add msbuild to PATH
         if: runner.os == 'Windows'
-        uses: microsoft/setup-msbuild@v1.3
+        uses: microsoft/setup-msbuild@v2
 
       - name: delvewheel install
         if: runner.os == 'Windows'
@@ -58,7 +58,7 @@ jobs:
           python -m pip install delvewheel cython
 
       - name: Build wheels
-        uses: joerick/cibuildwheel@v2.16.2
+        uses: joerick/cibuildwheel@v2.16.5
 
       - name: Install wheels
         shell: bash -l {0}
diff --git a/.github/workflows/wheel.yml b/.github/workflows/wheel.yml
index d1b97832..e3cbcb21 100644
--- a/.github/workflows/wheel.yml
+++ b/.github/workflows/wheel.yml
@@ -32,9 +32,9 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.12"
 
       - name: Set up QEMU
         if: runner.os == 'Linux'
@@ -44,7 +44,7 @@ jobs:
 
       - name: Add msbuild to PATH
         if: runner.os == 'Windows'
-        uses: microsoft/setup-msbuild@v1
+        uses: microsoft/setup-msbuild@v2
 
       - name: delvewheel install
         if: runner.os == 'Windows'
@@ -52,7 +52,7 @@ jobs:
           python -m pip install delvewheel cython
 
       - name: Build wheels
-        uses: joerick/cibuildwheel@v2.16.2
+        uses: joerick/cibuildwheel@v2.16.5
 
       - uses: actions/upload-artifact@v3
         with:
@@ -87,9 +87,9 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.12"
 
       - name: Set up QEMU
         if: runner.os == 'Linux'
@@ -99,7 +99,7 @@ jobs:
 
       - name: Add msbuild to PATH
         if: runner.os == 'Windows'
-        uses: microsoft/setup-msbuild@v1
+        uses: microsoft/setup-msbuild@v2
 
       - name: delvewheel install
         if: runner.os == 'Windows'
@@ -107,7 +107,7 @@ jobs:
           python -m pip install delvewheel cython
 
       - name: Build wheels
-        uses: joerick/cibuildwheel@v2.16.2
+        uses: joerick/cibuildwheel@v2.16.5
 
       - uses: actions/upload-artifact@v3
         with:
@@ -142,9 +142,9 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.12"
 
       - name: Set up QEMU
         if: runner.os == 'Linux'
@@ -154,7 +154,7 @@ jobs:
 
       - name: Add msbuild to PATH
         if: runner.os == 'Windows'
-        uses: microsoft/setup-msbuild@v1
+        uses: microsoft/setup-msbuild@v2
 
       - name: delvewheel install
         if: runner.os == 'Windows'
@@ -162,7 +162,7 @@ jobs:
           python -m pip install delvewheel cython
 
       - name: Build wheels
-        uses: joerick/cibuildwheel@v2.16.2
+        uses: joerick/cibuildwheel@v2.16.5
 
       - uses: actions/upload-artifact@v3
         with:
@@ -197,9 +197,9 @@ jobs:
             fetch-depth: 0
 
         - name: Setup Python
-          uses: actions/setup-python@v4
+          uses: actions/setup-python@v5
           with:
-            python-version: "3.11"
+            python-version: "3.12"
 
         - name: Set up QEMU
           if: runner.os == 'Linux'
@@ -209,7 +209,7 @@ jobs:
 
         - name: Add msbuild to PATH
           if: runner.os == 'Windows'
-          uses: microsoft/setup-msbuild@v1
+          uses: microsoft/setup-msbuild@v2
 
         - name: delvewheel install
           if: runner.os == 'Windows'
@@ -217,7 +217,7 @@ jobs:
             python -m pip install delvewheel cython
 
         - name: Build wheels
-          uses: joerick/cibuildwheel@v2.16.2
+          uses: joerick/cibuildwheel@v2.16.5
 
         - uses: actions/upload-artifact@v3
           with:
@@ -246,12 +246,12 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
-          python-version: "3.11"
+          python-version: "3.12"
 
       - name: Build wheels
-        uses: joerick/cibuildwheel@v2.16.2
+        uses: joerick/cibuildwheel@v2.16.5
 
       - uses: actions/upload-artifact@v3
         with:
diff --git a/fastparquet/__init__.py b/fastparquet/__init__.py
index 03ac88ae..17e7f4b6 100755
--- a/fastparquet/__init__.py
+++ b/fastparquet/__init__.py
@@ -1,8 +1,8 @@
 """parquet - read parquet files."""
 
-from ._version import __version__
-from .writer import write, update_file_custom_metadata
-from . import core, schema, converted_types, api
-from .api import ParquetFile
-from .util import ParquetException
+from fastparquet._version import __version__
+from fastparquet.writer import write, update_file_custom_metadata
+from fastparquet import core, schema, converted_types, api
+from fastparquet.api import ParquetFile
+from fastparquet.util import ParquetException
 
diff --git a/fastparquet/api.py b/fastparquet/api.py
index 53caba12..f6e55426 100644
--- a/fastparquet/api.py
+++ b/fastparquet/api.py
@@ -6,16 +6,15 @@
 
 import numpy as np
 import fsspec
-from fastparquet.util import join_path
 import pandas as pd
 
-from . import core, schema, converted_types, encoding, dataframe, writer
-from . import parquet_thrift
-from .cencoding import ThriftObject, from_buffer
-from .json import json_decoder
-from .util import (default_open, default_remove, ParquetException, val_to_num,
+from fastparquet import core, schema, converted_types, encoding, dataframe, writer
+from fastparquet import parquet_thrift
+from fastparquet.cencoding import ThriftObject, from_buffer
+from fastparquet.json import json_decoder
+from fastparquet.util import (default_open, default_remove, ParquetException, val_to_num,
                    ops, ensure_bytes, ensure_str, check_column_names, metadata_from_many,
-                   ex_from_sep, _strip_path_tail, get_fs, PANDAS_VERSION)
+                   ex_from_sep, _strip_path_tail, get_fs, PANDAS_VERSION, join_path)
 
 
 # Find in names of partition files the integer matching "**part.*.parquet",
diff --git a/fastparquet/compression.py b/fastparquet/compression.py
index f62fc72f..01188d70 100644
--- a/fastparquet/compression.py
+++ b/fastparquet/compression.py
@@ -1,7 +1,7 @@
 
 import cramjam
 import numpy as np
-from . import parquet_thrift
+from fastparquet import parquet_thrift
 
 # TODO: use stream/direct-to-buffer conversions instead of memcopy
 
diff --git a/fastparquet/converted_types.py b/fastparquet/converted_types.py
index cd8858e7..d21a5e37 100644
--- a/fastparquet/converted_types.py
+++ b/fastparquet/converted_types.py
@@ -10,9 +10,9 @@
 import numpy as np
 import pandas as pd
 
-from . import parquet_thrift
-from .cencoding import time_shift
-from .json import json_decoder
+from fastparquet import parquet_thrift
+from fastparquet.cencoding import time_shift
+from fastparquet.json import json_decoder
 
 logger = logging.getLogger('parquet')  # pylint: disable=invalid-name
 
diff --git a/fastparquet/core.py b/fastparquet/core.py
index 3facf893..79c17762 100644
--- a/fastparquet/core.py
+++ b/fastparquet/core.py
@@ -1,17 +1,16 @@
-import warnings
 import numpy as np
 import pandas as pd
 
-from . import encoding
-from . encoding import read_plain
+from fastparquet import encoding
+from fastparquet.encoding import read_plain
 import fastparquet.cencoding as encoding
-from .compression import decompress_data, rev_map, decom_into
-from .converted_types import convert, simple, converts_inplace
-from .schema import _is_list_like, _is_map_like
-from .speedups import unpack_byte_array
-from . import parquet_thrift
-from .cencoding import ThriftObject, read_thrift
-from .util import val_to_num, ex_from_sep
+from fastparquet.compression import decompress_data, rev_map, decom_into
+from fastparquet.converted_types import convert, simple, converts_inplace
+from fastparquet.schema import _is_list_like, _is_map_like
+from fastparquet.speedups import unpack_byte_array
+from fastparquet import parquet_thrift
+from fastparquet.cencoding import ThriftObject
+from fastparquet.util import val_to_num
 
 
 def _read_page(file_obj, page_header, column_metadata):
diff --git a/fastparquet/dataframe.py b/fastparquet/dataframe.py
index 51ba40c4..afa7d01b 100644
--- a/fastparquet/dataframe.py
+++ b/fastparquet/dataframe.py
@@ -11,7 +11,7 @@
 from pandas.core.arrays.masked import BaseMaskedDtype
 import warnings
 
-from .util import PANDAS_VERSION
+from fastparquet.util import PANDAS_VERSION
 
 
 class Dummy(object):
diff --git a/fastparquet/encoding.py b/fastparquet/encoding.py
index 3525c18f..8e69c344 100755
--- a/fastparquet/encoding.py
+++ b/fastparquet/encoding.py
@@ -1,8 +1,8 @@
 """encoding.py - methods for reading parquet encoded data blocks."""
 import numpy as np
-from .cencoding import read_bitpacked1, NumpyIO
-from .speedups import unpack_byte_array
-from . import parquet_thrift
+from fastparquet.cencoding import read_bitpacked1, NumpyIO
+from fastparquet.speedups import unpack_byte_array
+from fastparquet import parquet_thrift
 
 
 def read_plain_boolean(raw_bytes, count, out=None):
diff --git a/fastparquet/evolve.py b/fastparquet/evolve.py
new file mode 100644
index 00000000..e69de29b
diff --git a/fastparquet/parquet_thrift/__init__.py b/fastparquet/parquet_thrift/__init__.py
index c71820ef..85860548 100644
--- a/fastparquet/parquet_thrift/__init__.py
+++ b/fastparquet/parquet_thrift/__init__.py
@@ -4,7 +4,7 @@
 
 def __getattr__(name):
     # for compatability with coe that calls, e.g., parquet_thrift.RowGroup(...)
-    from ..cencoding import ThriftObject
+    from fastparquet.cencoding import ThriftObject
     if name[0].isupper():
         return partial(ThriftObject.from_fields, thrift_name=name)
     raise AttributeError(name)
diff --git a/fastparquet/schema.py b/fastparquet/schema.py
index 2a0e988e..83a43afd 100755
--- a/fastparquet/schema.py
+++ b/fastparquet/schema.py
@@ -1,7 +1,7 @@
 """Utils for working with the parquet thrift models."""
 from collections import OrderedDict
 
-from . import parquet_thrift
+from fastparquet import parquet_thrift
 
 
 def schema_tree(schema, i=0):
diff --git a/fastparquet/thrift_structures.py b/fastparquet/thrift_structures.py
index 75410a4a..b8ae51d2 100644
--- a/fastparquet/thrift_structures.py
+++ b/fastparquet/thrift_structures.py
@@ -1,2 +1,5 @@
-from . import parquet_thrift
-from .cencoding import ThriftObject
+from fastparquet import parquet_thrift
+from fastparquet.cencoding import ThriftObject
+
+
+__all__ = ["ThriftObject", "parquet_thrift"]
diff --git a/fastparquet/util.py b/fastparquet/util.py
index 4e1c3115..7e93f37b 100644
--- a/fastparquet/util.py
+++ b/fastparquet/util.py
@@ -14,8 +14,8 @@
 
 import fsspec
 
-from . import parquet_thrift
-from .cencoding import ThriftObject
+from fastparquet import parquet_thrift
+from fastparquet.cencoding import ThriftObject
 from fastparquet import __version__
 
 PANDAS_VERSION = Version(pd.__version__)
diff --git a/fastparquet/writer.py b/fastparquet/writer.py
index 873d6954..b023eb2e 100644
--- a/fastparquet/writer.py
+++ b/fastparquet/writer.py
@@ -11,19 +11,17 @@
 
 from fastparquet.util import join_path
 
-from . import parquet_thrift
-from .api import ParquetFile, partitions, part_ids
-from .compression import compress_data
-from .converted_types import tobson
-from .json import json_encoder
-from .util import (default_open, default_mkdirs, check_column_names,
+from fastparquet import parquet_thrift, __version__, cencoding
+from fastparquet.api import ParquetFile, partitions, part_ids
+from fastparquet.compression import compress_data
+from fastparquet.converted_types import tobson
+from fastparquet.json import json_encoder
+from fastparquet.util import (default_open, default_mkdirs, check_column_names,
                    created_by, get_column_metadata,
                    norm_col_name, path_string, reset_row_idx, get_fs,
                    update_custom_metadata)
-from . import __version__
-from .speedups import array_encode_utf8, pack_byte_array
-from . import cencoding
-from .cencoding import NumpyIO, ThriftObject, from_buffer
+from fastparquet.speedups import array_encode_utf8, pack_byte_array
+from fastparquet.cencoding import NumpyIO, ThriftObject, from_buffer
 from decimal import Decimal
 
 MARKER = b'PAR1'