zeroshade commented on code in PR #37788:
URL: https://github.com/apache/arrow/pull/37788#discussion_r1334793825
##########
dev/archery/archery/integration/tester_go.py:
##########
@@ -119,3 +131,124 @@ def flight_request(self, port, json_path=None,
scenario_name=None):
if self.debug:
log(' '.join(cmd))
run_cmd(cmd)
+
+ def make_c_data_exporter(self):
+ return GoCDataExporter(self.debug, self.args)
+
+ def make_c_data_importer(self):
+ return GoCDataImporter(self.debug, self.args)
+
+
+_go_c_data_entrypoints = """
+ const char* ArrowGo_ExportSchemaFromJson(const char* json_path,
+ uintptr_t out);
+ const char* ArrowGo_ImportSchemaAndCompareToJson(
+ const char* json_path, uintptr_t c_schema);
+
+ const char* ArrowGo_ExportBatchFromJson(const char* json_path,
+ int num_batch,
+ uintptr_t out);
+ const char* ArrowGo_ImportBatchAndCompareToJson(
+ const char* json_path, int num_batch, uintptr_t c_array);
+
+ int64_t ArrowGo_BytesAllocated();
+ void ArrowGo_RunGC();
+ void ArrowGo_FreeError(const char*);
+ """
+
+
[email protected]_cache
+def _load_ffi(ffi, lib_path=_INTEGRATION_DLL):
+ ffi.cdef(_go_c_data_entrypoints)
+ dll = ffi.dlopen(lib_path)
+ dll.ArrowGo_ExportSchemaFromJson
+ return dll
+
+
+class _CDataBase:
+
+ def __init__(self, debug, args):
+ self.debug = debug
+ self.args = args
+ self.ffi = cdata.ffi()
+ self.dll = _load_ffi(self.ffi)
+
+ def _pointer_to_int(self, c_ptr):
+ return self.ffi.cast('uintptr_t', c_ptr)
+
+ def _check_go_error(self, go_error):
+ """
+ Check a `const char*` error return from an integration entrypoint.
+
+ A null means success, a non-empty string is an error message.
+ The string is dynamically allocated on the Go side.
+ """
+ assert self.ffi.typeof(go_error) is self.ffi.typeof("const char*")
+ if go_error != self.ffi.NULL:
+ try:
+ error = self.ffi.string(go_error).decode('utf8',
+ errors='replace')
+ raise RuntimeError(
+ f"Go C Data Integration call failed: {error}")
+ finally:
+ self.dll.ArrowGo_FreeError(go_error)
+
+ def _run_gc(self):
+ self.dll.ArrowGo_RunGC()
+
+
+class GoCDataExporter(CDataExporter, _CDataBase):
+ # Note: the Arrow Go C Data export functions expect their output
+ # ArrowStream or ArrowArray argument to be zero-initialized.
+ # This is currently ensured through the use of `ffi.new`.
Review Comment:
See https://github.com/apache/arrow-adbc/issues/729 for the full discussion,
but essentially the callbacks that were populated into the C Data
`ArrowArrayStream` struct for getting the schema and getting the next record
batch had the same assumption that the output arguments were zero-initialized.
So @lidavidm created `trampoline.c` which essentially provided wrappers around
the Go exported `streamGetSchema` and `streamGetNext` function pointers, and we
use the trampoline methods as the function pointers we use (see `exportStream`
in `cdata/exports.go`)
But like i said, it looks like we only did that for the function pointers we
set for the callbacks in the `ArrowArrayStream` struct and not for the other
exported functions for regular import/export.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
