Conda keeps trying to downgrade arcpy 2.9 and dependencies to 2.8 version?

[Joshua-Young]

After doing a new install of ArcGIS Pro 2.9 I cloned the default Python environment and ran conda update --all. I get the following warning:

The environment is inconsistent, please check the package plan carefully
The following packages are causing the inconsistency:

- esri/win-64::arcgis==1.9.1=py37_2327
- esri/win-64::arcpy==2.9=py37_arcgispro_32704
- esri/win-64::cftime==1.5.0=py37_1
- esri/win-64::gdal==3.3.0=arcgispro_py37_16778
- esri/win-64::h5py==2.10.0=py37_arcgispro_12
- esri/noarch::lerc==2.2=py_0
- esri/win-64::matplotlib==3.4.2=py37_arcgispro_1
- defaults/win-64::mkl-service==2.3.0=py37h196d8e1_0
- defaults/win-64::mkl_fft==1.3.0=py37h46781fe_0
- esri/win-64::mkl_random==1.2.0=py37_0
- esri/win-64::netcdf4==1.5.7=py37_arcgispro_2
- defaults/win-64::numexpr==2.7.3=py37hcbcaa1e_0
- esri/win-64::numpy==1.20.1=py37_0
- defaults/win-64::pandas==1.2.3=py37hf11a4ad_0
- esri/noarch::saspy==3.7.3=py_0
- defaults/win-64::scipy==1.6.2=py37h14eb087_0
- esri/win-64::swat==1.9.3=py37_0
- defaults/noarch::xarray==0.17.0=pyhd3eb1b0_0

I figured out that it is esri/win-64::arcpy==2.9=py37_arcgispro_32704 that is causing the issue. Conda keeps trying to downgrade the following:

The following packages will be DOWNGRADED:

appdirs 1.4.4-py_0 --> 1.4.4-pyhd3eb1b0_0
arcgispro 2.9-0 --> 2.8-0
arcpy 2.9-py37_arcgispro_32704 --> 2.8-py37_arcgispro_29734
gdal 3.3.0-arcgispro_py37_16778 --> 2.3.3-arcgispro_py37_16747
h5py 2.10.0-py37_arcgispro_12 --> 2.10.0-py37_arcgispro_10
html5lib 1.1-py_0 --> 1.1-pyhd3eb1b0_0
jdcal 1.4.1-py_0 --> 1.4.1-pyhd3eb1b0_0
jsonschema 3.2.0-py_2 --> 3.2.0-pyhd3eb1b0_2
netcdf4 1.5.7-py37_arcgispro_2 --> 1.5.4-py37_arcgispro_6
pro_notebook_inte~ 2.9-py37_9 --> 2.8-py37_0
protobuf 3.13.0.1-py37ha925a31_1 --> 3.6.1-py37h33f27b4_0
requests-toolbelt 0.9.1-py_0 --> 0.9.1-pyhd3eb1b0_0
traitlets 4.3.3-py37_0 --> 4.3.3-py37haa95532_0
wcwidth 0.2.5-py_0 --> 0.2.5-pyhd3eb1b0_0
zlib 1.2.11-h62dcd97_4 --> 1.2.11-h8cc25b3_4

I have been able to duplicate this on every computer we have put ArcGIS Pro 2.9 on. Is this a bug in the arcpy Python library? It is also keeping pandas locked at version 1.2.4 for some reason. I was using the latest version of pandas with arcpy 2.8.

[DanPatterson · Accepted solution]

pandas 1.3.4 py37h6214cd6_0 is what I have in my base environment.

I don't "upgrade".  I did a complete removal of 2.8 and then installed pro 2.9

don't do an update all, these are pinned

arcgis 1.9.1.*
arcgispro 2.9.*
arcpy 2.9.*
freetype 2.10.*
gdal 3.3.0 arcgispro_py37_16778
h5py 2.10.*
matplotlib 3.4.2*
mkl 2020.0 arcgispro_167
nbconvert 5.6.*
netcdf4 1.5.7*
notebook 5.7.*
numpy 1.20.*
protobuf 3.13.0.1*
pro_notebook_integration 2.9.*
python 3.7.*
xlrd 1.2.*

If you want to upgrade pandas try --no-pin

It wants to remove/install a bunch of packages so assess what impact it may have on your clone.

>conda update pandas --dry-run

or

>conda update pandas --no-pin --dry-run

numpy, in particular, will cause you issues, which will affect pandas etc.

As a final test (in your clone), let it downgrade to 2.8, then update arcpy to 2.9 (it has been suggested by others and seems to have worked for them)

[Joshua-Young]

@DanPattersonthank you for the suggestions. The last suggestion worked best for me. I cloned a new Python environment, let it downgrade arcpy to 2.8 when I ran the update --all command, and then forced an install of arcpy 2.9. The odd thing was I ran the update command on arcpy 2.8 and conda said it was up-to-date. Either way, my other Python libraries are up-to-date and arcpy is back at 2.9, so again thank you for the tips. I still get the inconsistent packages warning when I install new packages but everything seems to be working for now.

[DanPatterson]

Hopefully things will improve when python moves to 3.8/9/10, being stuck at 3.7 is beginning to cause issues for me because numpy's newest version dumped 3.7 and other packages are beginning to bark as well, but arcpy/numpy are linked and something is holding python upgrades back.  Maybe arcpy 3.0