| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Our instructions for making a custom Cygwin package server rely on that.
|
|
At the cost of keeping more versions around, this avoids instantly vaulting
the old previous version when a new current version is uploaded, which makes
it a lot easier to restore things if it was meant to be a test version...
|
|
Fix thinkos introduced in 39136ce0 which broke the check that all packages
from a source package have the same version.
Also revise the exceptions list to list install packages, not source
packages
|
|
Actually permit the key in override.hint
Warn if we specify a non-existent version to keep
|
|
|
|
It's often wrong (e.g. in case of debuginfo package) and packages often have
a more useful subpackage indication (e.g. 'bindings', 'common data',
'development', 'documentation', 'runtime', 'samples', 'tools', 'utilities'
etc.), which we shouldn't be involved in guessing about.
|
|
Rename variable used to hold html listing filename to avoid clash with html
module
Add test
|
|
|
|
|
|
Fix an exception trying to determine best_version when a package erroneously
has no versions
|
|
With hindsight, override.hint seems to have been a bit of a mistake, as it
suffers from the same problem as setup.hint, that a careless update to it
unintentionally affects other versions.
So, as an alternate mechanism, add support for an explicit 'test:' in
pvr.hint, marking a version as a test version.
If multiple versions have are marked thus, the highest version will be
assigned to the test stability level. Versions marked as test cannot be
assigned to any other stability level.
|
|
|
|
|
|
|
|
|
|
|
|
Track the removed files in a package upload, so we can take different action
for absent and remove override.hint
If override.hint is absent in an upload, the values in any existing
override.hint apply
If override.hint is being removed in an upload, the values in any existing
override.hint should be ignored
|
|
This prevents bogus reports of a change when the legacy hint for a package
doens't exist, because a pvr.hint is used instead.
|
|
Absent stability level hints are now explicitly noted, so they can be
handled more correctly when merging a package.
|
|
Keep a copy of legacy hints so we can report if they change
|
|
|
|
Also remove a <br/> used to compensate for negative h2 margin
Also update relevant tests
|
|
|
|
The processing of them was implemented, but not the parsing...
|
|
Needed now that the comment contains a human-readable timestamp
|
|
We count the number of packages vaulted, not files
|
|
|
|
* calm/package.py (write_setup_ini): In addition to the epoch
timestamp write that time in a comment in human-readable form so
that it can be more easily deciphered.
|
|
We probably want to told when we are uploading a package which is
immediately going to be automatically removed as stale.
Uploading a package at the same time as a marking it for removal doesn't
make any sense, so report that as an error.
Update per-version testpackage to avoid this error.
|
|
Add an additional staleversion package to test the handling of stale
versions in the release area
|
|
Move stale package versions to the vault. A package version is not stale if
any of the following is true:
* The version is used as curr:, prev: or test: (due to either being
explictly listed in override.hint or implictly chosen)
* The version is listed as keep: in override.hint
* It's one of the highest n versions
* The first version which is less than t days old, and all higher versions.
where:
n is given by keep-count: in override.hint
t is given by keep-days: in override.hint
These tunables currently default to 0, so no more packages are kept than
would be kept by stalepkgs, although these defaults might change in future.
We track the freshness or staleness of install and source packages
separately. Unlike stalepkgs, a fresh empty install file does not cause the
corresponding source package to be considered fresh. Also remove any
pvr.hint which is stale.
Add the --no-stale option to disable.
First we check for stale pkgs in release area. stale pkgs arising due to
more fresh pckages being uploaded are reported as a part of the upload.
Avoid doing unnecessary work if nothing is stale.
Update tests
|
|
If there is an existing override.hint, it should be retained rather than be
discarded when there is an upload
|
|
Also check the setup.ini produced by the 'process' test against the expected output
|
|
Selection of the best version should take place after assuring there are
some versions, in case there are none.
|
|
|
|
|
|
|
|
Use the information from the 'best' version
(Previously we would have been using the same information from setup.hint
for all versions, so the precise version chosen perhaps doesn't make a big
difference)
|
|
|
|
Tests for packages with per-version and override hints
|
|
Update processing for the hints being stored for each package version,
rather than each package.
'skip' becomes a package flag, set if any version has it. (It doesn't seem
terribly useful to have this in the hint file, since it's implied by the
absence of any install tarfiles.)
When writing setup.ini
- sdesc:, ldesc:, category: and message: are taken from the curr version
- requires: is the union of all versions
- source: is controlled by external-source: per-version
|
|
For every package version, either setup.hint or pvr.hint must exist
Read version overrides from override.hint or setup.hint
Factor out reading hints and checking for errors in read_package
|
|
|
|
|
|
Fix reporting of problems with the package set after uploads are merged to
correctly identify the arch which has the problem.
|
|
If this option is specified, suppress warning that a package doesn’t
have a current version.
|
|
|
|
Also, make that check, and check for self-requires even if
-okmissing=required-package is used.
|
|
Don't import dirq unless we actually need it, i.e. if the experimental
option --queuedir is used.
Also remove dirq from setuptools requires
|
