numpy arrays endianess is incorrect

[ilyasdc]

Code for reproduction

from pycromanager import Studio
import numpy as np

h,w = 512,512
img = np.ones((h,w),dtype=np.int16)
mm = Studio(convert_camel_case=False)
mmimg = mm.data().createImage(img,
                  h,
                  w,
                  2,
                  1,
                  mm.data().coordsBuilder().build(),
                  mm.data().metadataBuilder().build())
print(mmimg.getIntensityAt(0,0),img[0,0])

Expected outcome
1 1

Actual outcome
256 1

Fix
The proposed fix is trivial but I don't know how to make it safe for any architecture/OS:

• I don't think Java endianess changes between architectures ?
• if it does, how does one check that the correct endianess is sent ?

diff --git a/bridge.py b/bridge_fix.py
index 8efbddd..9a86c4d 100644
--- a/bridge.py
+++ b/bridge_fix.py
@@ -804,11 +804,11 @@ def _serialize_arg(arg):
     if type(arg) in [bool, str, int, float]:
         return arg  # json handles serialization
     elif type(arg) == np.ndarray:
-        return arg.tobytes()
+        return arg.byteswap().tobytes()
     elif isinstance(arg, _JavaObjectShadow):
         return {"hash-code": arg._hash_code}
     else:
-        raise Exception("Unknown argumetn type")
+        raise Exception("Unknown argument type")
 
 
 def _check_single_method_spec(method_spec, fn_args):

PS: the iterate flag that I proposed and you merged while ago (cf #372) is not included in the signature of JavaObject

pycro-manager/pycromanager/zmq_bridge/wrappers.py

Lines 39 to 56 in 93e08b1

	class JavaObject(_JavaObjectShadow):
	"""
	Instance of a an object on the Java side. Returns a Python "Shadow" of the object, which behaves
	just like the object on the Java side (i.e. same methods, fields). Methods of the object can be inferred at
	runtime using iPython autocomplete
	"""
	def __new__(
	cls,
	classpath,
	args: list = None,
	port=_Bridge.DEFAULT_PORT,
	timeout=_Bridge.DEFAULT_TIMEOUT,
	new_socket=False,
	convert_camel_case=True,
	debug=False,
	):
	"""

so you can't create Studio object with this argument, unless you create the bridge manually (which is not anymore the advised way if I'm correct)

[henrypinkard]

• I don't think Java endianess changes between architectures ?

I'm not sure.

• if it does, how does one check that the correct endianess is sent ?

Seems like the safest way would be to check the endianness on the java side, and pass it along with binary data so it can be directly decoded on the python side

PS: the iterate flag that I proposed and you merged while ago (cf #372) is not included in the signature of JavaObject

I'm happy to merge a PR adding this to the signature (and/or anywhere else you think it is needed)

[ilyasdc]

Ok thanks for your input, I will make a proper PR !