0.7.0 - 2025-07-05

➡️  PyPI page: venvstacks 0.7.0.

Added

• show subcommand to display layer definitions (added in #159).
• --show option on subcommands (other than show) to display the selected layers and operations before executing the command (added in #159).
• --show-only option on subcommands (other than show) to display the selected layers and operations without executing the command (added in #159).
• --json option on subcommands to display the selected layers and operations as JSON rather than as a human-readable tree. For commands other than show, implies --show-only if --show is not passed explicitly (added in #159).

Changed

• Recursive source tree processing now excludes files excluded from version control when building from a git repository, and excludes __pycache__ folders otherwise. This exclusion affects both module hash calculations and the inclusion of files in built environments (resolves #203).
• RECORD files for installed packages are now largely retained in published artifacts and locally exported environments, with only the entries corresponding to omitted files removed (resolved in #28). This allows packages that inspect the metadata for installed packages at runtime to work correctly when deployed with venvstacks.
• Default CLI console output has been substantially reduced, with new -q/--quiet and -v/--verbose options added to adjust the message volume (changed in #5).
• Library level messages are now emitted via the logging module rather than being written directly to stdout. The CLI configures the logging subsystem appropriately based on the given verbosity options (changed in #5).

Fixed

• When using the --include filtering option for layer builds, existing "build if needed" environments are now correctly updated if they have not previously been successfully built with the current layer specification and environment lock details (reported in #222).
• Implicit versioning of runtime layers no longer breaks deployed layered environments using that layer (reported in #188).
• Implicit versioning of framework layers no longer breaks loading dynamic libraries from those layers on non-Windows systems (reported in #189)
• Layer locks are now marked as valid if the lock is successfully regenerated without changes after being marked as invalid due to a lower layer having an invalid lock (resolved in #227)

0.6.0 - 2025-06-07

➡️  PyPI page: venvstacks 0.6.0.

Added

• A new optional field, support_modules, has been added to application layer specifications. This field allows application layers to embed copies of common unpackaged support libraries without needing to duplicate that code in the source tree (proposed in #202).
• The lock subcommand now accepts an --if-needed option that skips locking layers that already have a valid layer lock (added in #200).

Changed

• Added a --lock-if-needed option to the build subcommand that ensures layers are only locked if they don't already have valid transitive environment locks. --lock is now a deprecated alias for this option rather than being equivalent to running the lock subcommand (proposed in #196).
• Changes to lock inputs that only affect the implicit layer versioning are now tracked separately from changes to the additional inputs that affect the result of the transitive dependency lock generation step. These changes are now ignored for layers that do not use implicit layer versioning (proposed in #201).
• Prefer the creation of hardlinks over full copies when locally exporting environments. Depending on the filesystem, this can make local exports significantly faster when the installed packages contain large files (proposed in #205).

Fixed

• Launch module existence checks are now skipped for layers that will not be built for the target build platform (reported in #204).

0.5.1 — 2025-05-26

➡️  PyPI page: venvstacks 0.5.1.

Changed

• Build failures for invalid layer locks now provide more details on the discrepancies that result in the lock being considered invalid (changed in #181).

Fixed

• launch_module is now correctly set in the internal venvstacks_layer.json configuration file shipped as part of application layers (resolved in #174).
• Layer locks are no longer incorrectly marked as invalid solely because the lock input cache files for the declared requirements are missing (reported in #175).
• Layer lock metadata generated by versions prior to 0.5.0 is now accepted as valid as long as the locked requirements file hasn't changed (resolved in #187).
• Resetting runtime and framework layer locks no longer prevents locking layers that depend on the affected layers (resolved in #187).
• Repeated local builds for environments using the dynamic library loading wrapper scripts no longer corrupt the base Python environment link (reported in #184).

0.5.0 - 2025-05-12

➡️  PyPI page: venvstacks 0.5.0.

Changed

• Layer locks are now invalidated for launch module changes. This also means that implicit versioning will update the layer version (resolves #89).
• The exception raised when reporting dynamic library symlink conflicts in a layer now reports all ambiguous library targets in the layer instead of only reporting the first ambiguity encountered (resolved in #158).

Fixed

• Previously defined layer locks are now correctly invalidated in the following cases (resolves #149):
  • the layer's declared input requirements have changed
  • the major Python version of the layer's base runtime has changed
  • the layer depends on a layer that does not currently have a valid layer lock
  • the relative paths from the layer to the layers it depends have changed (including additions and removals of layer dependencies)
  • implicit layer versioning is enabled or disabled for the layer
• Attempting to lock a layered environment now fails if any layer it depends on does not have a currently valid layer lock (resolves #161).
• CLI arguments on Windows are no longer unexpectedly resolved as filesystem glob patterns (resolved in #160).
• Dynamic library symlinks are now correctly removed if the dynamic library is no longer included in the built layer (resolved in #163).
• As it affects launch module execution, application layer launch module hashes now incorporate the file name in addition to the file contents (resolved in #164).
• Application layer launch packages are now consistently archived using the layer's lock timestamp, even when that is more recent than the file's local modification time (resolved in #148).

0.4.1 - 2025-04-25

➡️  PyPI page: venvstacks 0.4.1.

Added

• Locking layers now emits package summary files for each layer, which should make it easier to see what has changed when locks are updated (suggested in #108).

Changed

• The exception raised when reporting dynamic library symlink conflicts in a layer now suggests using the dynlib_exclude setting to resolve the conflict (changed in #141).

Fixed

• The --reset-lock option now propagates to derived layers as intended (reported in #137).

0.4.0 - 2025-04-11

➡️  PyPI page: venvstacks 0.4.0.

Added

• Framework layers may now specify frameworks to depend on one or more framework layers instead of depending directly on a runtime layer. Framework dependencies must form a directed acyclic graph (DAG), and framework layers must be defined after any framework layers they depend on (proposed in #18, implemented in #119).
• Application layers may now specify runtime to depend directly on a a runtime layer with no intervening framework layers (added as part of resolving #18).
• All layers may now specify dynlib_exclude to indicate dynamic libraries which should not be symbolically linked into the share/venv/dynlib/ environment subfolder on Linux and macOS (added as part of resolving #38).

Changed

• To enable loading of shared objects from other environment layers, framework and application environments on Linux and macOS now run Python via a suitably capable shell environment (bash on Linux, zsh on macOS) that can be expected to be consistently installed (changed in #38).

Fixed

• Extension modules on Linux and macOS that rely on shared objects published by their dependencies (for example, PyTorch depending on CUDA libraries) now work correctly even if those dependencies are installed in a lower environment layer. See dynamic-linking for additional details (resolved in #38).

0.3.0 - 2025-03-28

Added

• The build and lock subcommands accept a new --reset-lock
  CLI option. This multi-use option requests that any previously
  created layer lock file be removed before locking the selected
  layers (thus ignoring any previous version pins or artifact
  hashes). This option uses the same wildcard pattern matching as
  the --include option. Only layers that are locked by the given
  command will have their previous lock files removed, as excluded
  layers will be excluded from both locking and having their lock
  files reset (added in #22).
• "win_arm64" and "linux_aarch64" are now accepted as target platforms.
  ARM64/Aarch64 refer to the same CPU architecture, but Python reports it differently
  depending on the OS, and this is reflected in their respective platform tags
  (added in #107).

Changed

• A Python API instability FutureWarning is now emitted at runtime (added while resolving #22).
• The previous :func:!BuildEnvironment.get_unmatched_patterns method has been replaced
  by the new :func:BuildEnvironment.filter_layers method, which returns both the
  matching layer names and the unmatched patterns (changed in #22).
• :func:BuildEnvironment.select_layers now accepts an iterable of environment names
  rather than an iterable of filter patterns to be matched (changed in #22).

Fixed

• --only-binary ":all:" is now passed when locking the layers in addition
  to being passed when creating the layer environments. This avoids emitting
  requirements that can't be installed (resolved in #102).
• Remove directories from /bin when building layers (resolved in #103)

0.2.1 — 2024-12-05

Fixed

• Fix Typer 0.14.0+ incompatibility when setting app name (reported by Rugved Somwanshi in #96).

0.2.0 — 2024-11-14

Added

• Setting versioned = True in a layer definition will now append a
  lock version number to the layer name that automatically increments
  each time the locked requirements change for that layer (layer@1,
  layer@2, etc). (implemented in #24).
• Added documentation for the :ref:stack-specification-format (part of #78).
• Added python_implementation to the published layer metadata (part of #78).
• Added bound_to_implementation to the published layer metadata (part of #78).

Changed

• Enabled rendered previews for documentation PRs (requested in #43).
• Enabled link validity checks when rendering documentation (requested in #62).
• Renamed :class:!EnvironmentExportRequest to :class:LayerExportRequest (part of #33).
• Exposed :class:LayerSpecBase, :class:LayeredSpecBase as public classes (part of #33).
• Exposed :class:LayerEnvBase, :class:LayeredEnvBase as public classes (part of #33).
• Added leading underscores to several private functions and methods (part of #33).
• Added docstrings to all remaining public functions and methods (part of #33).
• Updated docs to actively discourage using @ in layers names (part of #78).
• Renamed fully_versioned_name runtime layer specification field to python_implementation (part of #78).
• Renamed runtime_name to runtime_layer in the layer metadata (to align with the required_layers field),
  and simplified it to always refer to the runtime layer's install target name (part of #78).

Fixed

• Post-installation scripts for layered environments now work
  correctly even when run with a Python installation other
  than the expected base runtime (resolved in #66)

0.1.1 — 2024-11-01

Changed

• Update docs URL to https://venvstacks.lmstudio.ai__

• Add OpenGraph metadata to docs landing page

• Resolved several broken links in the documentation

• Documentation is now marked as being unversioned
  (it is published directly from the main branch)