Parse and solve a workspace manifest#

Watch the Discover, solve and maintain a workspace demo and follow the commands below.

Discover the environments in a workspace, inspect their requirements and solve selected targets into a combined lock. Parse mode reads declarations without solving. Workspace parsing requires conda-workspaces 0.10.0 or later.

Create a manifest#

Save this as conda.toml:

[workspace]
name = "parse-demo"
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64"]

[dependencies]
python = "3.13.*"

[target.linux-64.dependencies]
strace = "*"

[feature.test.dependencies]
pytest = ">=8"

[environments]
default = []
test = { features = ["test"] }

The test environment inherits the top-level Python requirement and adds pytest. The Linux target also includes strace.

Discover and select from the CLI#

List the environment declarations and their platform mappings:

conda presto --parse --file conda.toml

The JSON response reports format: "conda-toml", the default and test environments, and selected: []. Discovery does not select the host platform.

Inspect the test environment for Linux:

conda presto --parse --file conda.toml \
  --environment test --platform linux-64

The selected entry contains the Python, pytest and strace requirements, channels and system requirements for that target. Repeat --environment or --platform to select more targets. Selecting only --environment test returns both declared platforms, with strace only in the Linux requirements.

Use the HTTP API#

Start the service in another terminal:

conda presto --serve

These examples use curl and jq. Submit the same manifest for discovery:

jq -n --rawfile file conda.toml \
  '{file: $file, filename: "conda.toml"}' |
  curl --fail-with-body --silent --show-error \
    --header 'Content-Type: application/json' --data-binary @- \
    http://127.0.0.1:8000/parse

Select the test environment for Linux:

jq -n --rawfile file conda.toml \
  '{file: $file, filename: "conda.toml", environments: ["test"], platforms: ["linux-64"]}' |
  curl --fail-with-body --silent --show-error \
    --header 'Content-Type: application/json' --data-binary @- \
    http://127.0.0.1:8000/parse

The HTTP and CLI responses have the same shape. If either selector is supplied, omitted environments mean all environments and omitted platforms mean each selected environment’s declared platforms. Supply a platform explicitly when the manifest declares none. Empty, unknown or ambiguous selectors produce an error.

Solve selected environments#

Create one lock containing the default and test environments on both platforms:

conda presto --file conda.toml \
  --environment default --environment test \
  --platform linux-64 --platform osx-arm64 \
  --format conda-workspaces-lock-v1 > conda.lock

The lock contains four environment/target solutions. A failed pair prevents a successful incomplete lock. Omitting both selectors solves all declared environments and targets.

The HTTP API accepts the same selection through repeated query parameters when uploading raw TOML:

curl --fail-with-body --silent --show-error \
  --header 'Content-Type: application/toml' \
  --data-binary @conda.toml \
  'http://127.0.0.1:8000/resolve?filename=conda.toml&environment=default&environment=test&platform=linux-64&platform=osx-arm64&format=conda-workspaces-lock-v1' \
  --output conda.lock

JSON requests use environments and platforms arrays alongside file and filename. Successful eligible responses include a Location for the retained exact output. See HTTP API.

Export normalized dependencies#

Select one environment to write its composed requirements as a new manifest:

mkdir -p exported
conda presto --export --file conda.toml --environment test \
  --platform linux-64 --platform osx-arm64 \
  --format conda-toml > exported/conda.toml

Export mode composes declarations through conda-workspaces without running the solver. The exporter separates shared and target-specific dependencies. It does not preserve the original feature organization, comments, tasks or every workspace setting. This is a normalized dependency manifest, not a fully pinned lock. The pixi-toml and pyproject-toml formats provide equivalent output in their supported syntax. Selected targets must have distinct concrete subdirectories and identical ordered channels.

The HTTP operation uses the same selection:

curl --fail-with-body --silent --show-error \
  --header 'Content-Type: application/toml' \
  --data-binary @conda.toml \
  'http://127.0.0.1:8000/export?filename=conda.toml&environment=test&platform=linux-64&platform=osx-arm64&format=conda-toml' \
  --output exported/conda.toml

Unsolved declarations cannot produce lockfiles, explicit package URLs or SBOMs. Use the solve operation above when exact package records are required.

Understand the result#

platform preserves a workspace target’s logical name and subdir identifies its conda platform. This matters when a manifest declares multiple variants of the same subdirectory. Choose the logical name when the subdirectory is ambiguous.

Selected local, Git and URL PyPI sources are rejected. Version requirements that need missing optional conda-pypi support are reported as errors. The number of selected environment/platform combinations uses the configured platform limit, with specs and channels limits applied to each target.

Omit the solve format to receive native JSON with an environment, logical platform, concrete subdir, package list and error field for each target. Workspace solves use the manifest’s requirements and channels, so additional inline specs, channel overrides and multiple input files are rejected.

/sbom rejects direct workspace manifests. After creating conda.lock, use the saved-lock SBOM workflow to render selected environments without solving again. /capabilities reports workspace_parse: true and workspace_solve: true. See HTTP API for response fields, CLI reference for options and Use conda-presto in GitHub Actions for CI use.

On this page
Show Source