git patches may require copies to be handled out-of-order. For instance, take
the following sequence:
* modify a
* copy a into b
Here, we have to generate b from a before its modification. To do so,
applydiff() was scanning for copy metadata and performing the copies before
processing the other changes in-order. While smart and efficient, this approach
complicates things by handling file copies and file creations at different
places and times. While a new file must not exist before being patched a copied
file already exists before applying the first hunk.
Instead of copying the files at their final destination before patching, we
store them in a temporary file location and retrieve them when patching. The
filestore always stores file content in real files but nothing prevents adding
a cache layer. The filestore class was kept separate from fsbackend for at
least two reasons:
- This class is likely to be reused as a temporary result store for a future
repository patching call (entries just have to be extended to contain copy
sources).
- Delegating this role to backends might be more efficient in a repository
backend case: the source files are already available in the repository itself
and do not need to be copied again. It also means that third-parties backend
would have to implement two other methods. If we ever decide to merge the
filestore feature into backend, a minimalistic approach would be to compose
with filestore directly. Keep in mind this copy overhead only applies for
copy/rename sources, and may even be reduced to copy sources which have to
handled ahead of time.
The patcher has to know if a file is being created or removed to check if the
target already exists, or to actually unlink the file when a hunk emptying it
is applied. This was done by embedding the creation/removal information in the
first (and only) hunk attached to the file.
There are two problems with this approach:
- creation/removal is really a property of the file being patched and not its
hunk.
- for regular patches, file creation cannot be deduced at parsing time: there
are case where the *stripped* file paths must be compared. Modifying hunks
after their creation is clumsy and prevent further refactorings related to
copies handling.
Instead, we delegate this job to selectfile() which has all the relevant
information, and remove the hunk createfile() and rmfile() methods.
workingctx.remove(list, unlink=True) is unsuited here, because it does too
much: it also unlinks added files. But the command 'hg remove' is specified
to *never* unlink added files.
Instead, we now unlink the files at the commands.remove level (if --after was
not specified) and use workingctx.forget for all files.
As an added bonus, this happens to eliminate a wlock acquire/release pair,
since the previous implementation caused
acquire wlock
release wlock
acquire wlock
release wlock
where the first pair of acquire/release was caused by the workingctx.forget
call, and the second by the workingctx.remove call.
Currently, if there is a bare git subrepo, but it is at the "right"
revision, calling dirty() will error because diff-index does not work
on bare repos. This patch makes it so bare subrepos are always
considered dirty.
and check if we got one before creating.
note that the contents of the ui object might change after
dispatch() returns (by options passed through --config for example),
to ensure it doesn't, pass a copy() of it.
Restore the previous diffstat behaviour of scaling by the maximum number of
changes to a single file. Changeset 7bb0e22a7988 modified the diffstat to be
scaled by the total number of changes. This seems to have been unintentional.
Firstly, I think we should do this for all new wire commands, just
to be on the safe side. So I want to get this into the 1.9 release.
Secondly, there actually is potential here that sometimes the server
can know that the number of its nodes which can possibly still be
undecided on the client is small. It might then just send them along
directly (cutting short the end game). This, however, requires
walking the graph on the server, which can be expensive, so for the
moment we're not actually doing it.
It has substantially different semantics from forget at the command
layer, so change it to avoid confusion.
We can't simply combine it with remove because we need to explicitly
drop non-added files in some cases like commit.
This new Python code should be equivalent in behavior to the if
statement at line 312 of parsers.c. Without this, the pure-python
parsers improperly ignore truncated revlogs as created in
test-verify.t.
Changeset 20b319765bcf introduced the unbundlehash capability and
unconditionally hashed the heads on the client side. By mistake, the
heads were also cased in the heads == ['force'] case.
extensions that depend on other extensions (such as record) use this pattern
to check if the dependant extension is available:
try:
mq = extensions.find('mq')
except KeyError:
return
but since if an error occurs while loading an extension it leaves its entry
in the _extensions map as None, we want to raise in that situation too.
(rather than adding another check if the return value is None)
requires ctypes
Why is posixfile a class?
Because the implementation needs to use the Python library call os.fdopen [1],
which sets the 'name' attribute on the Python file object it creates to the
mostly meaningless string '<fdopen>', since file descriptors don't have a name.
But users of posixfile depend on the name attribute [2] being set to a proper
value, like Python's built-in 'open' function sets it on file objects.
Python file's name attribute is read-only, so we can't just assign to it after
the file object has alrady been created.
To solve this problem, we save the name of the file on a wrapper object,
and delegate the file function calls to the wrapped (private) file object
using __getattr__.
[1] http://docs.python.org/library/os.html#os.fdopen
[2] http://docs.python.org/library/stdtypes.html#file.name
Extensions can hook discovery.findcommonincoming to filter out unwanted remote
changesets. This patch makes getremotechanges respect the changed remote heads
returned by such extensions.
gitpatch objects emitted by iterhunks() were referencing file paths unmodified
from the input patch. _applydif() made them usable by modifying the gitpatch
objects in-place with specified path strip level. The same modified objects
were then reused by iterhunks() generator. _applydiff() now copies and update
the paths which completely decouples both routines.
As a side effect, the "git" event now receives only metadata about
copies/renames to perform the necessary copies ahead of time. Other actions are
handled in the "file" event.
Patch changes are emitted by iterhunks() in two separate events: 'file' when
hunks have to be applied and 'git' to describe other modifications like copies
or mode changes. Note that a file which mode is changed and which content is
modified by the same patch will be emitted in both events. It is more
convenient to handle all file modifications in a single event. This patch
"zips" git actions with regular changes so both kinds can be emitted at the
same place.
git afile/bfile are extracted twice, once when reading a 'diff --git', and
again when reading a unified hunk. The problem is not all git blocks have
unified hunks (renames just have metadata) and they were not extracted the same
way. This is what this patch unifies.
gitpatch objects emitted by iterhunks() are modified in place by applydiff().
Processing them earlier improves iterhunks() isolation. applydiff() modifying
them should still be fixed though.