| Age | Commit message (Collapse) | Author |
|---|
|
'Reference' implementation now shares all the code from the SymbolicReference base. This implementation allows to create any reference you'd like without requireing git calls
|
|
object. They are only constrained to be located in the refs/ directory
|
|
the item as well to enable predicates to do more proper checking
|
|
method, improved code performance by disabling the visit-once flag by default for trees, which by nature are not recursive
|
|
|
|
allowing the callee to know more about its environment
|
|
the first very pure implementation
|
|
Fixed critical bug in object code: IndexObjects now use their path as hashkey, not the data\!
|
|
including a test
|
|
appear litereally anywhere. A 'default' reference is nothing more than a detached symbolic reference.
Added method iterate the packed refs file
|
|
how it should be, but many tests are still failing as the implementation is not yet there
|
|
restriction anymore.
Added a test to assure the git commands can handle it
|
|
IndexObject._mode_str_to_int: Now uses the 6 relevant bytes of the passed in octal string
|
|
any of our changes entries as it would in fact write a possibly cached tree stored in our extension data.It was solved by simply ignoring that extension data when writing the index for tree creation. A test was added for this as well
|
|
Index.checkout: fixed bug in exception creator function
|
|
using the current index. This makes it more flexible. For convenience, the index.commit method is still available, it delgates all the work
|
|
trigger an error
|
|
index.write_tree may now use the missing_ok argument which writes trees faster and allows you to operate without objects to do a true index based merge
|
|
that always occour when dealing with working tree diffs ( for ovious reasons ). Also the implementation would previously leave empty blobs where no file actually existed
|
|
fast_forwards although. Renamed cumbersome 'commit_before_forced_updated' attribute to 'old_commit' to be en par with PushInfo
|
|
exact information about the successfully checked out files as well as the failed ones
Return value of the method is an iterator to all checked out files
|
|
messages - previously it would be more narrow and raise a GitCommandError in cases where it wasnt actually needed
|
|
have not been updated as they where modified locally or did not exist. Test was improved to check for this new case.
|
|
|
|
processing - delays could have been possible previously as it would expand (possibly) large amounts of paths in advance
|
|
code does not rely on my own git patch anymore as it is unsure whether it will make it in there or when
|
|
just in time
This adds a dependency to a git update index fix - and I hope it will be put in in time ( including windows version )
|
|
symlinks.
Added index.get_entries_key method to allow direct access to anyone willing to do so
|
|
Previously it would consider such a path a relative one and handle it incorrectly
|
|
email address specified
|
|
* win_fixes:
Fixed commit.count method which now handles the paths case properly. It appears git-rev-list uses empty paths in some way, which is quite hard to specify on a shell, but easy if the process is spawned directly
test_remote: fixed test which assumed existance of local master tracking branch, it will now create it if necessary
Index tests adopted to windows - especially the symlink test needed adjustment, but it works as expected even on systems that do not support symlinks
ARGH: wb and rb is not the same as r and w on windows, hence reading of binary files went crazy as well as binary writing
test_commit: commit.count actually returned incorrect values on linux, namely 141 instead of 143. Manual checking showed that 143 is the correct number, on linux this will have to be fixed
helper: repo creation functions now handle errors on windows during os.remove by changing the mode to 777 and delete the file again. Otherwise the whole operation would fail on read-only files. Why is windows as it is ? Why does it do that to me ?
removed large-input test as it is totally dependent on the subprocess implementation in the end whether pipeing large input works. In general , input and output pipes are used, the shell is bypassed, hence there is no reason for a problem unless we are on a very rare platform. And if so, we can't do anything about it so why should there be a possibly failing test ? Problem is that the test would fail on windows in case it is not installed on c:\windows
repo.clone: Added plenty of special handling to allow drive letters to work as expected. Its quite terrible to see a two-line method inflate to 20
Fixed config module which forgot to call the superclass's initializer, finally causing failure in python 2.6
fixed test_repo to work on windows
cmd: added clear_cache method now used by test repo decorators to be sure persistent commands are killed before trying to remove the directory. Unfortunately, it still claims someone has opened the file. handle.exe does not show anyone, so what is happening here ? Is it just a windows odity ? If nothing helps I could just keep the temp data, but lets do some more testing first
git cmd on windows now runs without the shell, see diff for explanation
Fixed windows TASKKILL so it actually does something *silently*
Added utilities helping to create proper paths either with slashes or backslashes depending on the operating system
|
|
appears git-rev-list uses empty paths in some way, which is quite hard to specify on a shell, but easy if the process is spawned directly
|
|
branch, it will now create it if necessary
|
|
adjustment, but it works as expected even on systems that do not support symlinks
|
|
binary files went crazy as well as binary writing
|
|
namely 141 instead of 143. Manual checking showed that 143 is the correct number, on linux this will have to be fixed
|
|
os.remove by changing the mode to 777 and delete the file again. Otherwise the whole operation would fail on read-only files. Why is windows as it is ? Why does it do that to me ?
|
|
implementation in the end whether pipeing large input works. In general , input and output pipes are used, the shell is bypassed, hence there is no reason for a problem unless we are on a very rare platform. And if so, we can't do anything about it so why should there be a possibly failing test ? Problem is that the test would fail on windows in case it is not installed on c:\windows
|
|
as expected. Its quite terrible to see a two-line method inflate to 20
as there is no git-daemon on windows, some tests will not work. The error message has been adjusted to be more precise for the poor people trying to run the tests on windows ( including myself )
|
|
finally causing failure in python 2.6
|
|
cmd: taskkill now pipes stderror to nul as well
|
|
persistent commands are killed before trying to remove the directory. Unfortunately, it still claims someone has opened the file. handle.exe does not show anyone, so what is happening here ? Is it just a windows odity ? If nothing helps I could just keep the temp data, but lets do some more testing first
|
|
|
|
|
|
backslashes depending on the operating system
fixed test_refs and test_trees
Many more issues remain though, this is just a first backup commit
|
|
needs to be started
|
|
head.checkout functions for greater ease of use - they work exactly as expected now.
Fixed failing test in remote tests by actually removing the line
|
|
|
|
|
|
* docs:
Converted tabs to spaces.
added paragraph about the diffing engine and how to switch branches, which once more shows the need for a real checkout for all the people who do not want to implement all the safety facilities themselves
Added Index Object paragraph and wrote the Remote Handling paragraph. Updated repo area
added paragraph about using the git-python command
Worked on the tree object handling and realized an issue with attribute naming on index objects and the way the system handles these
initial work on tutorial.rst
added all new files to reference.rst and corrected the worst mistakes. There are still a few errors left that I cannot fix as it complains about whitespace in the end ... that is exactly what I hate restructured text for, its just a ... anyway.
|
