| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Sorting any permutation of Version objects should always yield the same
result, even if those hold some build metadata.
To that end, the "precedence_key" is now used exclusively for sorting;
direct comparisons between Version objects still ignores the "build"
metadata, using a different precedence key.
For performance improvements, both precedence keys are cached.
Closes: #132
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
- The README is now a standalone document, also included as an
"introduction" page;
- A new "guide" section provides more details on most features;
- A couple of typos were fixed.
The main goal was to make the README file perfectly suitable for
rendering on PyPI, while keeping its content available on the standard
documentation on ReadTheDocs.
|
|
|
|
|
|
| |
Including docs and tests.
Closes #98.
|
| |
|
|
|
|
| |
Write clear, explicit rules for the `SimpleSpec` syntax.
|
|
|
|
|
| |
Their comparison semantics were ill-defined, and mostly an
implementation detail for the old, 'native' specs.
|
|
|
|
|
| |
- Use proper theme in development
- Use ReadTheDocs version numbers in titles
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This will be used in `sort(..., key=lambda v: v.precedence_key)`.
Remove previous comparison/ordering implementation.
The current implementation relies on a 4-tuple:
- major, minor, patch (as integers) - natural order matches precedence
rules
- a tuple of identifiers for the prerelease component.
The identifiers for the prerelease components are based on:
- A `NumericIdentifier` class, that compares number using their natural
order, and always compares lower than non-numeric identifiers
- A `AlphaIdentifier` class for non-numeric identifiers; it compares
versions using ASCII ordering.
- A `MaxIdentifier` class, that compares higher to any other identifier;
used to ensure that a non-prerelease version is greater than any of
its prereleases.
|
|
|
|
|
|
|
|
|
| |
The internal features from those classes will be removed in future
versions:
- The `Spec` class is incompatible with the support of multiple syntaxes
- The `SpecItem` class was an implementation detail, but doesn't support
complex `Range` combinations.
|
|
|
|
|
|
|
|
| |
The code follows closely the specification available at
https://docs.npmjs.com/misc/semver.html.
Despite similarities, the matching logic is fully separate from the
`native` code, since both might evolve at their own scales.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Instead of choosing the comparison on each `.match()` call, the
expression is converted to a combination of `Range()` expressions
(simple comparison to a semver-compliant `Version`).
`Range()` objects can be combined with `And` and `Or` through the
`AnyOf` and `AllOf` clauses (sticking to Python's naming scheme).
Some specific flags have been provided to those range, allowing users to
subtly alter the matching behaviour - thus accomodating different
versioning schemes:
- `<0.1.2` won't match `0.1.2-rc1`, unless the prerelease_policy flag is
set to either `always` or `same-patch`
- `<0.1.2` will match `0.1.1-rc1`, unless the `prerelease_policy` flag
is set to `same-patch`
- `==0.1.2` will always match `0.1.2+build44`, unless the `build_policy`
is set to `strict`.
The `Spec` item has been updated, alongside `SpecItem`.
Those objects keep the original expression as attributes, but don't use
them for comparisons.
|
| |
|
|
|
|
|
|
|
|
| |
According to the stated goal of "intuitive" behaviour, we want:
``Version('0.1.1-a1') not in Spec('<0.1.1')``.
Tests, code and docs have been fixed.
|
|
|
|
| |
This simplifies computing neighbouring versions.
|
|
|
|
|
|
| |
Eases the creation of version objects from existing versions.
We still validate the type and structure of each component.
|
|
|
|
|
|
|
|
|
|
| |
Those functions had been forgotten in the docs.
Clarify that `Version('1.0.1-alpha').next_patch()` is
`Version('1.0.1')`, since that version is the smallest non-prerelease
version greater than `Version('1.0.1-alpha')`.
Closes: #67
|
| |
|
|
|
|
| |
The PR was broken through fixed in ``next_minor()`` / ``next_major()``.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
SemVer 2.0.0 states that "Build metadata SHOULD be ignored when
determining version precedence".
This means that, when comparing ``0.1.0+1`` to ``0.1.0+bcd``::
>>> Version('0.1.0+1') == Version('0.1.0+bcd')
False
>>> Version('0.1.0+1') != Version('0.1.0+bcd')
True
>>> Version('0.1.0+1') < Version('0.1.0+bcd')
False
>>> Version('0.1.0+1') > Version('0.1.0+bcd')
False
>>> Version('0.1.0+1') <= Version('0.1.0+bcd')
False
>>> Version('0.1.0+1') >= Version('0.1.0+bcd')
False
>>> compare(Version('0.1.0+1'), Version('0.1.0+bcd'))
NotImplemented
This change also has the following effects:
- When including build metadata in a ``Spec``, the only valid options
are ``Spec('==0.1.0+sth')`` and ``Spec('!=0.1.0+sth')``
- The meaning of ``Spec('==0.1.0+')`` is now "Only version 0.1.0 without
build metadata"
- ``Spec('==0.1.0')`` now matches ``Version('0.1.0+anything')``
|
|
|
|
| |
Spec('*') will match all valid Version objects.
|
| |
|
| |
|
|
Also normalize the package layout.
Thanks @jdowner-gb & tleach for the report.
|