On ArcGIS Pro 3.5, MGET installation instructions remove arcpy module #47
Description
Currently, Step 3 of the MGET 3.4.0 installation instructions say to issue this command:
micromamba install --channel conda-forge --yes mget3
But when you do that on a clone of the ArcGIS Pro 3.5 arcgispro-py3 conda environment, it will remove the arcpy module:
(arcgispro-py3-mget2) C:\Users\jjr8\AppData\Local\ESRI\conda\envs\arcgispro-py3-mget2>micromamba install --channel conda-forge mget3
conda-forge/noarch 21.2MB @ 21.0MB/s 1.0s
conda-forge/win-64 29.0MB @ 25.3MB/s 1.1s
Pinned packages:
<List of pinned packages deleted from this GitHub issue for brevity>
Transaction
Prefix: C:\Users\jjr8\AppData\Local\ESRI\conda\envs\arcgispro-py3-mget2
Updating specs:
- mget3
warning libmamba Extracted package cache 'C:\Users\jjr8\AppData\Roaming\mamba\pkgs\python_abi-3.11-2_cp311' has invalid url
Package Version Build Channel Size
--------------------------------------------------------------------------------------------
Install:
--------------------------------------------------------------------------------------------
+ _openmp_mutex 4.5 2_gnu conda-forge Cached
+ asciitree 0.3.3 py_2 conda-forge Cached
+ blosc 1.21.6 h85f69ea_0 conda-forge 50kB
+ boto3 1.39.5 pyhd8ed1ab_0 conda-forge Cached
+ botocore 1.39.5 pyge310_1234567_0 conda-forge Cached
+ bzip2 1.0.8 h2466b09_7 conda-forge Cached
+ ca-certificates 2025.7.14 h4c7d964_0 conda-forge 156kB
+ copernicusmarine 2.0.1 pyhd8ed1ab_0 conda-forge Cached
+ deprecated 1.2.18 pyhd8ed1ab_0 conda-forge Cached
+ docutils 0.21.2 pyhd8ed1ab_1 conda-forge Cached
+ fasteners 0.19 pyhd8ed1ab_1 conda-forge Cached
+ freexl 2.0.0 h8276f4a_0 conda-forge 77kB
+ geos 3.13.0 h5a68840_0 conda-forge 2MB
+ geotiff 1.7.4 h887f4e7_0 conda-forge 123kB
+ h5netcdf 1.6.3 pyhd8ed1ab_0 conda-forge Cached
+ hdf4 4.2.15 h5557f11_7 conda-forge Cached
+ jmespath 1.0.1 pyhd8ed1ab_1 conda-forge Cached
+ lcms2 2.17 hbcf6048_0 conda-forge 511kB
+ lerc 4.0.0 h6470a55_1 conda-forge 165kB
+ libarchive 3.7.4 haf234dc_0 conda-forge 958kB
+ libcurl 8.14.1 h88aaa65_0 conda-forge 368kB
+ libexpat 2.7.0 he0c23c2_0 conda-forge 141kB
+ libgcc 15.1.0 h1383e82_3 conda-forge Cached
+ libgdal-core 3.9.2 h042995d_7 conda-forge 8MB
+ libgomp 15.1.0 h1383e82_3 conda-forge Cached
+ libiconv 1.18 h135ad9c_1 conda-forge Cached
+ libjpeg-turbo 3.1.0 h2466b09_0 conda-forge 838kB
+ libkml 1.3.0 h538826c_1021 conda-forge 2MB
+ librttopo 1.1.0 hd4c2148_17 conda-forge 405kB
+ libspatialite 5.1.0 h939089a_11 conda-forge 8MB
+ libsqlite 3.50.2 hf5d6505_2 conda-forge Cached
+ libssh2 1.11.1 h9aa295b_0 conda-forge 293kB
+ libwebp-base 1.6.0 h4d5522a_0 conda-forge 279kB
+ libwinpthread 12.0.0.r4.gg4f2fc60ca h57928b3_9 conda-forge Cached
+ libxcb 1.17.0 h0e4246c_0 conda-forge Cached
+ libzlib 1.3.1 h2466b09_2 conda-forge Cached
+ lzo 2.10 hcfcfb64_1001 conda-forge 143kB
+ mget3 3.4.0 py311h3e6a449_0 conda-forge Cached
+ minizip 4.0.6 hb638d1e_0 conda-forge 85kB
+ numcodecs 0.15.1 py311hcf9f919_0 conda-forge Cached
+ openjpeg 2.5.3 h4d64b90_0 conda-forge 240kB
+ pcre2 10.44 h99c9b8b_2 conda-forge 817kB
+ proj 9.5.1 h4f671f6_0 conda-forge 3MB
+ pthread-stubs 0.4 h0e40799_1002 conda-forge Cached
+ pyhdf 0.11.6 py311hc008540_0 conda-forge Cached
+ pystac 1.13.0 pyhd8ed1ab_0 conda-forge Cached
+ python-tzdata 2025.2 pyhd8ed1ab_0 conda-forge Cached
+ python_abi 3.11 2_cp311 conda-forge Cached
+ s3transfer 0.13.0 pyhd8ed1ab_0 conda-forge Cached
+ scikit-fmm 2025.6.23 py311h11fd7f3_0 conda-forge Cached
+ semver 3.0.4 pyhd8ed1ab_0 conda-forge Cached
+ tk 8.6.13 h2c6b04d_2 conda-forge 3MB
+ tzlocal 5.3 py311h1ea47a8_0 conda-forge Cached
+ ucrt 10.0.22621.0 h57928b3_1 conda-forge Cached
+ uriparser 0.9.8 h5a68840_0 conda-forge 49kB
+ vc14_runtime 14.44.35208 h818238b_26 conda-forge Cached
+ xerces-c 3.2.5 he0c23c2_2 conda-forge 4MB
+ xorg-libxau 1.0.12 h0e40799_0 conda-forge Cached
+ xorg-libxdmcp 1.1.5 h0e40799_0 conda-forge Cached
+ zarr 2.18.7 pyhd8ed1ab_0 conda-forge Cached
Remove:
--------------------------------------------------------------------------------------------
- arcpy 3.5 py311_arcgispro_57294 esri 6kB
- arcpy-base 3.5 py311_57294 esri 5MB
- jpeg 9e 0 esri 278kB
Change:
--------------------------------------------------------------------------------------------
- libtiff 4.7.0 1 esri 926kB
+ libtiff 4.7.0 hfc51747_1 conda-forge 979kB
- pillow 11.1.0 py311_0 esri 956kB
+ pillow 11.1.0 py311h43e43bb_0 conda-forge 42MB
- zlib 1.3.1 1 esri 115kB
+ zlib 1.3.1 h2466b09_2 conda-forge Cached
Upgrade:
--------------------------------------------------------------------------------------------
- cryptography 43.0.1 py311_0 esri 1MB
+ cryptography 45.0.5 py311h5e0b3ae_0 conda-forge 1MB
- libdeflate 1.20 1 esri 73kB
+ libdeflate 1.22 h2466b09_0 conda-forge 156kB
- libpng 1.6.43 1 esri 346kB
+ libpng 1.6.50 h95bef1e_0 conda-forge 352kB
- openssl 3.0.16 0 esri 3MB
+ openssl 3.5.1 h725018a_0 conda-forge 9MB
- vs2015_runtime 14.38.33130 0 esri 1MB
+ vs2015_runtime 14.44.35208 h38c0c73_26 conda-forge Cached
Downgrade:
--------------------------------------------------------------------------------------------
- gdal 3.10.2 arcgispro_py311_17888 esri 1MB
+ gdal 3.9.2 py311h689aae6_7 conda-forge 2MB
Summary:
Install: 60 packages
Remove: 3 packages
Change: 3 packages
Upgrade: 5 packages
Downgrade: 1 packages
Total download: 91MB
--------------------------------------------------------------------------------------------
Confirm changes: [Y/n]
That's obviously a bad thing. My best guess is that this occurs due to a difference in the choice of MSVC C/C++ compiler and runtime libraries chosen by Esri for Pro 3.5's Python environment vs. those chosen by conda-forge for packages released there, including MGET. While MGET itself currently contains little C/C++ code that would be affected by this, it relies on a number of other Python packages that have substantial amounts of C/C++. Alternatively, rather than being the MSVC compilers and runtimes, it might be a common library such as jpeg or zlib that is the problem. In any case, it appears that when micromamba tries to resolve MGET's dependencies, it determines that the only way get them installed is to install something that arcpy is incompatible with, and therefore arcpy will no longer work so micromamba wants to remove it as well.
We can fix it by adding the esri channel to the install command. When we do that, micromamba appears to switch out the arcpy that comes with Pro 3.5 for another one available on the esri channel:
(arcgispro-py3-mget) C:\Users\jjr8\AppData\Local\ESRI\conda\envs\arcgispro-py3-mget>micromamba install --channel esri --channel conda-forge mget3
conda-forge/noarch 21.2MB @ 28.2MB/s 0.7s
conda-forge/win-64 29.0MB @ 35.0MB/s 0.8s
Pinned packages:
<List of pinned packages deleted from this GitHub issue for brevity>
Transaction
Prefix: C:\Users\jjr8\AppData\Local\ESRI\conda\envs\arcgispro-py3-mget
Updating specs:
- mget3
warning libmamba Extracted package cache 'C:\Users\jjr8\AppData\Roaming\mamba\pkgs\python_abi-3.11-2_cp311' has invalid url
Package Version Build Channel Size
-----------------------------------------------------------------------------------
Install:
-----------------------------------------------------------------------------------
+ asciitree 0.3.3 py_2 conda-forge Cached
+ boto3 1.39.5 pyhd8ed1ab_0 conda-forge 84kB
+ botocore 1.39.5 pyge310_1234567_0 conda-forge 8MB
+ copernicusmarine 2.0.1 pyhd8ed1ab_0 conda-forge Cached
+ deprecated 1.2.18 pyhd8ed1ab_0 conda-forge Cached
+ docutils 0.21.2 pyhd8ed1ab_1 conda-forge Cached
+ fasteners 0.19 pyhd8ed1ab_1 conda-forge Cached
+ h5netcdf 1.6.3 pyhd8ed1ab_0 conda-forge Cached
+ hdf4 4.2.15 h5557f11_7 conda-forge Cached
+ jmespath 1.0.1 pyhd8ed1ab_1 conda-forge Cached
+ libjpeg-turbo 3.0.3 0 esri Cached
+ libzlib 1.3.1 h2466b09_2 conda-forge Cached
+ mget3 3.4.0 py311h3e6a449_0 conda-forge Cached
+ numcodecs 0.15.1 py311hcf9f919_0 conda-forge Cached
+ pyhdf 0.11.6 py311hc008540_0 conda-forge Cached
+ pystac 1.13.0 pyhd8ed1ab_0 conda-forge Cached
+ python-tzdata 2025.2 pyhd8ed1ab_0 conda-forge Cached
+ python_abi 3.11 2_cp311 esri Cached
+ s3transfer 0.13.0 pyhd8ed1ab_0 conda-forge Cached
+ scikit-fmm 2025.6.23 py311h11fd7f3_0 conda-forge Cached
+ semver 3.0.4 pyhd8ed1ab_0 conda-forge Cached
+ tzlocal 5.3 py311h1ea47a8_0 conda-forge Cached
+ ucrt 10.0.22621.0 h57928b3_1 conda-forge Cached
+ vc14_runtime 14.44.35208 h818238b_26 conda-forge Cached
+ zarr 2.18.7 pyhd8ed1ab_0 conda-forge Cached
Change:
-----------------------------------------------------------------------------------
- arcpy 3.5 py311_arcgispro_57294 esri 6kB
+ arcpy 3.5 py311_arcgispro_57257 esri 74kB
- arcpy-base 3.5 py311_57294 esri 5MB
+ arcpy-base 3.5 py311_57257 esri 5MB
- zlib 1.3.1 1 esri 115kB
+ zlib 1.3.1 h2466b09_2 conda-forge Cached
Upgrade:
-----------------------------------------------------------------------------------
- vs2015_runtime 14.38.33130 0 esri 1MB
+ vs2015_runtime 14.44.35208 h38c0c73_26 conda-forge Cached
Downgrade:
-----------------------------------------------------------------------------------
- gdal 3.10.2 arcgispro_py311_17888 esri 1MB
+ gdal 3.9.2 arcgispro_py311_17882 esri 1MB
Summary:
Install: 25 packages
Change: 3 packages
Upgrade: 1 packages
Downgrade: 1 packages
Total download: 14MB
-----------------------------------------------------------------------------------
Confirm changes: [Y/n]
I did that and MGET's automated tests related to ArcGIS all passed, except #46. That looks like a behavior change in ArcGIS Pro 3.5, but could conceivably be a bug present in the py311_arcgispro_57257 build of arcpy that micromamba used as a replacement for the py311_arcgispro_57294 build that came with Pro 3.5. I will investigate that as a separate issue. For this overall installation issue, we should update the installation instructions to this command:
micromamba install --channel esri --channel conda-forge mget3