Package search and install specifications
Conda supports the following specifications for conda search
and conda install
.
Package search
conda search
for a specific package or set of packages can be accomplished in several ways. This section includes information on the standard specification and the use of key-value pairs.
Standard specification
![../../_images/conda_search.png](../../_images/conda_search.png)
- channel
(Optional) Can either be a channel name or URL. Channel names may include letters, numbers, dashes, and underscores.
- subdir
(Optional) A subdirectory of a channel. Many subdirs are used for architectures, but this is not required. Must have a channel and backslash preceding it. For example:
main/noarch
- name
(Required) Package name. May include the
*
wildcard. For example,*py*
returns all packages that have "py" in their names, such as "numpy", "pytorch", "python", etc.- version
(Optional) Package version. May include the
*
wildcard or a version range(s) in single quotes. For example:numpy=1.17.*
returns all numpy packages with a version containing "1.17." andnumpy>1.17,<1.19.2
returns all numpy packages with versions greater than 1.17 and less than 1.19.2.- build
(Optional) Package build name. May include the
*
wildcard. For example,numpy 1.17.3 py38*
returns all version 1.17.3 numpy packages with a build name that contains the text "py38".
Key-value pairs
Package searches can also be performed using what is called "key-value pair notation", which has different rules than the Standard specification example image. The search below will return the same list of packages as the standard specification.
$ conda search "numpy[channel=conda-forge, subdir=linux-64, version=1.17.*, build=py38*]"
This notation supports the following key-value pairs:
- build # validated via GlobStrMatch
- build_number # validated via BuildNumberMatch
- channel # validated via ChannelMatch
- features # validated via FeatureMatch
- fn # validated via ExactStrMatch
- license # validated via CaseInsensitiveStrMatch
- license_family # validated via CaseInsensitiveStrMatch
- md5 # validated via ExactStrMatch
- name # validated via GlobLowerStrMatch
- sha256 # validated via ExactStrMatch
- subdir # validated via ExactStrMatch
- track_features # validated via FeatureMatch
- url # validated via ExactStrMatch
- version # validated via VersionSpec
Key-value pair notation can be used at the same time as standard notation.
$ conda search "conda-forge::numpy=1.17.3[subdir=linux-64, build=py38*]"
Warning
Any search values using the key-value pair notation will override values in the rest of the search string. For example, conda search numpy 1.17.3[version=1.19.2]
will return packages with the version 1.19.2
.
Package installation
When you're installing packages, conda recommends being as concrete as possible. Using *
wildcards and version ranges during an install will most likely cause a conflict.
However, *
wildcards can still be helpful in an install command when used sparingly.
Installing with wildcards
Let's say you are working on a project that requires version 2.3 of a package. If you upgrade to 2.4 or 3.0, your project will break. You're also using an environment file to create your environment.
In the version 2.3.1
, 2
is the major version, 3
is the minor version, and 1
is the patch. Patches typically contain bug fixes, so if you want to keep version 2.3
in your environment without updating to 2.4
or 3.0
, but want to take advantage of any bug fixes, using 2.3.*
in your environment file would be helpful to you.
Concrete install example
Let's take the search from the Package search section.
$ conda search "conda-forge/linux-64::numpy 1.17.* py38*"
This returns the following:
Loading channels: done
# Name Version Build Channel
numpy 1.17.3 py38h95a1406_0 conda-forge
numpy 1.17.5 py38h18fd61f_1 conda-forge
numpy 1.17.5 py38h95a1406_0 conda-forge
You can then choose a specific version and build, if necessary, and edit your conda install
command accordingly.
$ conda install "conda-forge/linux-64::numpy 1.17.5 py38h95a1406_0"