Take into account GDAL >= 3.7 GDT_Int8 signed int8 datatype (fixes #2635) (#2656)

rasterio/_base.pyx

@@ -160,8 +160,10 @@ cdef _band_dtype(GDALRasterBandH band):
     """Resolve dtype of a given band, deals with signed/unsigned byte ambiguity"""
     cdef const char * ptype
     cdef int gdal_dtype = GDALGetRasterDataType(band)
-    if gdal_dtype == GDT_Byte:
-        # Can be uint8 or int8, need to check PIXELTYPE property
+    if gdal_dtype == GDT_Byte and dtypes.dtype_rev["int8"] == 1:
+        # Before GDAL 3.7, int8 was dealt by GDAL as a GDT_Byte (1)
+        # with PIXELTYPE=SIGNEDBYTE metadata item in IMAGE_STRUCTURE
+        # metadata domain.
         ptype = GDALGetMetadataItem(band, 'PIXELTYPE', 'IMAGE_STRUCTURE')
         if ptype and strncmp(ptype, 'SIGNEDBYTE', 10) == 0:
             return 'int8'

rasterio/_io.pyx

@@ -1437,7 +1437,9 @@ cdef class DatasetWriterBase(DatasetReaderBase):
             # if we can crosswalk those.
             gdal_dtype = _get_gdal_dtype(self._init_dtype)
 
-            if _getnpdtype(self._init_dtype) == _getnpdtype('int8'):
+            # Before GDAL 3.7, int8 was dealt by GDAL as a GDT_Byte (1)
+            # with PIXELTYPE=SIGNEDBYTE creation option.
+            if _getnpdtype(self._init_dtype) == _getnpdtype('int8') and gdal_dtype == 1:
                 options = CSLSetNameValue(options, 'PIXELTYPE', 'SIGNEDBYTE')
 
             # Create a GDAL dataset handle.

rasterio/dtypes.py

@@ -10,6 +10,7 @@ import numpy
 from rasterio.env import GDALVersion
 
 _GDAL_AT_LEAST_35 = GDALVersion.runtime().at_least("3.5")
+_GDAL_AT_LEAST_37 = GDALVersion.runtime().at_least("3.7")
 
 bool_ = 'bool'
 ubyte = uint8 = 'uint8'
@@ -47,10 +48,15 @@ if _GDAL_AT_LEAST_35:
     dtype_fwd[12] = int64 # GDT_Int64
     dtype_fwd[13] = uint64 # GDT_UInt64
 
+if _GDAL_AT_LEAST_37:
+    dtype_fwd[14] = sbyte # GDT_Int8
+
 dtype_rev = dict((v, k) for k, v in dtype_fwd.items())
 
 dtype_rev["uint8"] = 1
-dtype_rev["int8"] = 1
+if not _GDAL_AT_LEAST_37:
+    dtype_rev["int8"] = 1
+
 dtype_rev["complex"] = 11
 dtype_rev["complex_int16"] = 8
 
@@ -82,6 +88,9 @@ if _GDAL_AT_LEAST_35:
     typename_fwd[12] = 'Int64'
     typename_fwd[13] = 'UInt64'
 
+if _GDAL_AT_LEAST_37:
+    typename_fwd[14] = 'Int8'
+
 typename_rev = dict((v, k) for k, v in typename_fwd.items())
 
 dtype_ranges = {

rasterio/gdal.pxi

@@ -221,6 +221,7 @@ cdef extern from "gdal.h" nogil:
     ctypedef enum GDALDataType:
         GDT_Unknown
         GDT_Byte
+        GDT_Int8
         GDT_UInt16
         GDT_Int16
         GDT_UInt32

tests/test_int8.py

@@ -3,9 +3,12 @@ import numpy
 import pytest
 
 import rasterio
+from rasterio.env import GDALVersion
+
+gdal_version = GDALVersion.runtime()
 
 
-@pytest.mark.xfail(reason="Likely upstream bug")
+@pytest.mark.xfail(gdal_version < GDALVersion(3, 7), reason="GDAL < 3.7 doesn't handle signed int8 correctly")
 @pytest.mark.parametrize("nodata", [-1, -128])
 def test_write_int8_mem(nodata):
     profile = {

tests/test_write.py

@@ -12,10 +12,9 @@ from .conftest import requires_gdal35, gdal_version
 import rasterio
 from rasterio.drivers import blacklist
 from rasterio.enums import MaskFlags, Resampling
-from rasterio.env import Env
+from rasterio.env import Env, GDALVersion
 from rasterio.errors import RasterioIOError
 
-
 def test_validate_dtype_None(tmpdir):
     """Raise TypeError if there is no dtype"""
     name = str(tmpdir.join("lol.tif"))
@@ -118,7 +117,10 @@ def test_write_sbyte(tmpdir):
 
     info = subprocess.check_output(["gdalinfo", "-stats", name]).decode('utf-8')
     assert "Minimum=-33.000, Maximum=-33.000, Mean=-33.000, StdDev=0.000" in info
-    assert 'SIGNEDBYTE' in info
+    if GDALVersion.runtime() < GDALVersion(3, 7):
+        assert 'SIGNEDBYTE' in info
+    else:
+        assert 'Int8' in info
 
 
 @pytest.mark.gdalbin