Releases: lmstudio-ai/venvstacks
Releases Β· lmstudio-ai/venvstacks
0.4.1 - 2025-04-25
β‘οΈΒ 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
β‘οΈΒ 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
andlock
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
0.2.1 β 2024-12-05
Fixed
- Fix Typer 0.14.0+ incompatibility when setting app name (reported by Rugved Somwanshi in #96).
0.2.0 β 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 topython_implementation
(part of #78). - Renamed
runtime_name
toruntime_layer
in the layer metadata (to align with therequired_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 β 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)
0.1.0 β 2024-10-31
Changed
- Further documentation fixes and improvements
0.1.0rc1.post0 β 2024-10-30
Changed
-
Included project URLs in project metadata
-
Added installation instructions to
README.md
-
Linked to main documentation from
README.md
-
Improved the content split between the project
overview page and the top level docs landing page
0.1.0rc1 β 2024-10-29
Added
-
Initial export of
venvstacks
from Project Amphibian. -
Adopted
scriv
forCHANGELOG
management.