| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| |
|
|
|
|
|
|
|
| |
This fixes volatile cache validation that did not get fixed in commit
a63c163a50bf8a4e5ca8673cd8bebae29c36643a. Now all volatile cache
problems that have been introduced since commit
2ed1cb53cc4158af08c22d466b15b9a9a7767212 should be fixed.
|
|
|
|
|
|
|
|
|
| |
The metadata_overlay usage, that was triggered automatically for users
without depcachedir write access, has been broken since commit
2ed1cb53cc4158af08c22d466b15b9a9a7767212. The metadata_overlay class
is not compatible with the new validation code, so remove it. If users
have metadata_overlay configured as their cache module, fall back to
flat_hash.
|
|
|
|
|
| |
The deepcopy in catch/volatile.py misbehaved when it tried to copy
attributes like __DEEPCOPY__ that didn't correspond to known functions.
|
| |
|
| |
|
|
|
|
| |
This code only needs to exist in _metadata_callback.
|
|
|
|
|
| |
Allows forced ChangeLog generation even when the vcs has detected that
the ChangeLog has already been modified.
|
|
|
|
|
| |
Like echangelog does, show ChangeLog or Manifest if there are no other
changes to display.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
These tests only run if the git binary is found. They include testing
of new --echangelog=y option.
|
| |
|
|
|
|
|
| |
This feature got dropped on commit
b655137b7b5f87c11a7e742b7ef0e028a1139677.
|
| |
|
|
|
|
|
|
|
|
| |
Also:
* copy the old header from the old ChangeLog if it exists, in case it
contains a non-gentoo header
* don't add a header if the old ChangeLog exists and doesn't contain
a header
|
|
|
|
|
|
|
|
|
|
| |
Instead of calling echangelog, which on its turn has to query the VCS
again, use the existing information on changes made to the current
directory, and update the ChangeLog from Python itself.
This avoids a call to echangelog, and avoids again retrieving the same
VCS information as repoman already did. It makes repoman independent
from external tools it didn't install itself, and should be faster in
general.
|
|
|
|
|
| |
This makes it consistent with perform_checksum call which also raises
FileNotFound instead of OSError.
|
| |
|
| |
|
|
|
|
|
| |
This options disables creation of binary packages, no matter
what enabled it in the first place. See bug 386903.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
This removes a conversion to bytes that should have been removed in
commit 2ed1cb53cc4158af08c22d466b15b9a9a7767212. The hashed_path
location attribute should always be unicode, for consistency will path
handling throughout portage. The hashed_path class internally uses the
portage portage.os wrapper to ensure that paths are encoded as utf_8
bytes when they are passed to the real os module for stat calls.
|
| |
|
| |
|
|
|
|
|
| |
For parse-eapi-ebuild-head, we want to assign self.metadata from the
return value, for conformity with usage elsewhere.
|
|
|
|
|
|
|
| |
This fixes a regression in FEATURES=parse-eapi-ebuild-head support
for egencache since commit 2ed1cb53cc4158af08c22d466b15b9a9a7767212.
If the metadata does not contain _mtime_ here, then _setitem from
cache.metadata.database will raise KeyError.
|
|
|
|
|
| |
This fixes bugs that can only be triggered by egencache since
other callers handle parse-eapi-ebuild-head earlier.
|
|
|
|
|
| |
This fixes a regression in FEATURES=parse-eapi-ebuild-head support
for egencache since commit 2ed1cb53cc4158af08c22d466b15b9a9a7767212.
|
| |
|
|
|
|
| |
This removes the last reference to PORTAGE_CALLER.
|
| |
|
|
|
|
|
| |
This fixes a regression in FEATURES=parse-eapi-ebuild-head support
since commit 2ed1cb53cc4158af08c22d466b15b9a9a7767212.
|
| |
|
|
|
|
|
|
| |
Also, copy repoman_settings.categories to each individual
profile's dep_settings.categories attribute, in order to ensure
that portdbapi.cp_list() uses the correct category set.
|
| |
|
| |
|
|
|
|
|
| |
Since subversion 1.7, only a top-level .svn dir is used. Make FindVCS
also deal with that. (Tested with repoman on Prefix SVN-based tree.)
|
|
|
|
|
|
|
|
| |
Since changes to eclass validation code in commit
2ed1cb53cc4158af08c22d466b15b9a9a7767212, validation of cache for
packages with empty INHERITED metadata has been broken due to failure
to distinguish between None and empty dict return values from
eclass.cache.validate_and_rewrite_cache().
|
|
|
|
|
|
| |
The hunk that got removed from action_metadata in commit
d4ea29bf6a3ce35d49e0f54f9173e3a6e42da2d6 is actually needed when
FEATURES=metadata-transfer is enabled.
|
|
|
|
|
| |
We can re-use the vcs_files_to_cps() function from commit
9c9145a9e3c2d8d1bcebf791ce2188add656fee2.
|
|
|
|
| |
Avoid unnecessary nested loops.
|
| |
|
| |
|
|
|
|
|
|
| |
This triggered a regression since commit
2ed1cb53cc4158af08c22d466b15b9a9a7767212 that caused cache entries
containing eclass paths to appear invalid.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Enabled via cache-format = md5-dict
This format is essentially just flat_hash, using md5 rather than mtime,
and dropping the path component from _eclasses_ entries.
From a speed standpoint, the md5 overhead is ~16% in comparison to mtime,
timed on a modern sandybridge; specifically, validating 29k nodes takes
~8.8s for flat_md5, while the pms norm is ~7.7s.
That said, the cache is /usable/ in places PMS is not; in those cases,
it can definitely be a win since even if the cache is partially old,
it's better than regenerating everything from scratch.
(cherry picked from commit 95ddf97e2f7e7d3f6a072604b2df5f77e9298558)
Change-Id: Ic3561369b7a8be7f86480f339ab1686fddea6dff
|
|
|
|
|
|
|
|
|
|
| |
Specifically, the cache can use any portage supported checksum method,
or use the standard mtime approach. In addition, support controlling
whether or not paths are stored, and generally try to restore some
of the centralization/encapsulation that was in place originally.
(cherry picked from commit bc1aed614fb588f0ade5bcb5d1265a8db0f8d247)
Change-Id: Ic38057e7dbb15063c64a93c99e66e113a7d4c70e
|