We add a `filtered` method on repo. This method return an instance of `repoview`
that behaves exactly as the original repository but with a filtered changelog
attribute. Filters are identified by a "name". Planned filter are `unserved`,
`hidden` and `mutable`. Filtering the repository in place what out of question
as it wont not allows multiple thread to share the same repo. It would makes
control of the filtering scope harder too. See the `repoview` docstring for
details.
A mechanism to compute filtered revision is also installed. Some caches will be
installed in later commit.
For a repository with over 400,000 commits, rebasing one revision near tip,
this avoids two treks up the DAG, speeding the operation up by around 1.6
seconds.
With the current implementation, `changelog.nodemap` is not filtered. So some
filtered changeset in common are not filtered by `n in nodemap`. This leads to
crash lower in the stack when the bundle generation try to access those node on
a filtered changelog.
The check pattern only checked for whitespace between keyword and operator.
Now it also warns:
> x = f(),7
missing whitespace after ,
> x = f()+7
missing whitespace in expression
All filecache usage on repo is for logic that should be unfiltered. The
caches should be common to all filtered instances, and computation must
be done unfiltered. A dedicated storecache subclass is created for
this purpose.
Some of the localrepo property caches must be computed unfiltered and
stored globally. Some others must see the filtered version and store data
relative to the current filtering.
This changeset introduces two classes `unfilteredpropertycache`
and `filteredpropertycache` for this purpose. A new function
`hasunfilteredcache` is introduced for unambiguous checking for cached
values on unfiltered repos.
A few tweaks are made to the property cache class to allow overriding
the way the computed value is stored on the object.
Some logic relative to _tagcaches is cleaned up in the process.
Computation of common changesets during push needs to be done on the
widest set possible. An unfiltered version of the repo is kept for
discovery and various revset calls. The discovery code itself enforces
the filtering of unserved outgoing changeset.
The current branchcache construction is not aware of filtering. We keep
the status quo, ensuring that the branch cache logic is computed as
before: without any filtering.
This decorator ensure the method in run on an unfiltered version of the
repository. See follow-up commit for details.
This decorator is not named `unfiltered` because it would clash with the
`unfilteredmethod` on `localrepo` itself.
This commit is part of the changelog level filtering effort. It returns
the main "unfiltered" version of a repo-like object. For localrepo this
means the same localrepo object. But this method will be overwritten
by the filtered versions of a repository to return the core unfiltered
version of the repo.
Introducing this simple method first allows later commits to prepare
for the use of a filtered version of a repository.
A new repo method is added because a lot of users may call it. At the
end of this series of commits, about 40 calls exist in core and hgext.
Bookmarks persistence still showed a fair amount of its legacy as a
monkeypatching extension. This encapsulates all bookmarks
serialization and parsing in a single class, and offers a single
location where other bookmarks storage engines can be substituted
in. As a result, many files no longer import the bookmarks module,
which strikes me as an encapsulation win.
This doesn't do anything to the current bookmark state yet, but I'm
hoping put that in the bmstore class as well.
This applies the same logic as used for `obsolete` and `unstable` changesets.
Refuse to push them without force.
We'll probably want to factor this logic with two new functions
`pctx.troubled()` and `ctx.troubles()`. But I'm waiting for the third
"trouble" to make it into core.
This factors out the checks from tags and bookmarks, and newly applies
the same prohibitions to branches. checknewlabel takes a new parameter,
kind, indicating the kind of label being checked.
Test coverage is added for all three types of labels.
Recomputing branch cache on clone may be expensive,
therefore if possible we fetch it along with the data.
- If the clone is performed by copying, we just copy branchcache file.
- If we localrepo.clone and streaming then we follow the procedure:
1. Fetch branchmap from the remote
2. Fetch the actual data.
3. Find the latest rev within branch heads (tip at the time of
branchmap fetch)
4. Update the cache for the revs in [remotetip+1, tip]
This way we ensure that the branchcache is correct even in case
of races with commits.
Tags initially prevented revision to be hidden. It seemed a bad idea to have
tags refer to revisions that one can't see. But proper filtering of hidden
revisions excludes them from tag computation. Coming changelog filtering will do
that. Anyway, tags that really matter will likely be public and therefore not
hidden.
The current working directory parent and bookmarked revision are still not
hidden. Bookmarks were likely automatically moved at rewrite time, bookmarks
that remain on obsolete revisions were probably moved there on purpose.
If there are filtered changesets the cache is not valid. We'll have to cache
branchmap for filtered state too, but for now recomputing the branchmap is
enough.
The forced recomputation of the branch cache was introduced by `b4909adfc093`.
Back there, `addchangegroup` did not handle any lock logic.
Later `6042c410e045` introduced lock logic to `addchangegroup`. Its description
does not explain why the `updatebranchcache` call is made outside locking. I
believe that the lock was released there because it fit well with the transaction
release already in the code.
Finally `1eda82d76f0c` moved all "unlocked" code of `addchangegroup` to an
`repo._afterlock` callback.
I do not think that the call to `updatebranchcache()` requires to be done
outside locking. That may even be a bad idea to do so. Bringing this call back
in the `addchangegroup` function makes the flow simpler and eases the following
up changelog level filtering business.
This patch also changes initialization order of "*opener" and "*vfs"
fields: first, "*vfs" fields are initialized , and then, "*opener"
ones are initialized.
We usually update bookmarks only if the new location is descendant of the old
bookmarks location. We extract this logic into a function. This is the first
step to allow more complex logic using obsolescence in this validation of the
bookmark movement.
This changeset introduces caches on the `obsstore` that keeps track of sets of
revisions meaningful for obsolescence related logics. For now they are:
- obsolete: changesets used as precursors (and not public),
- extinct: obsolete changesets with osbolete descendants only,
- unstable: non obsolete changesets with obsolete ancestors.
The cache is accessed using the `getobscache(repo, '<set-name>')` function which
builds the cache on demand. The `clearobscaches(repo)` function takes care of
clearing the caches if any.
Caches are cleared when one of these events happens:
- a new marker is added,
- a new changeset is added,
- some changesets are made public,
- some public changesets are demoted to draft or secret.
Declaration of more sets is made easy because we will have to handle at least
two other "troubles" (latecomer and conflicting).
Caches are now used by revset and changectx. It is usually not much more
expensive to compute the whole set than to check the property of a few elements.
The performance boost is welcome in case we apply obsolescence logic on a lot of
revisions. This makes the feature usable!
This restores the old behaviour of clearing the filecache when the repo is
destroyed but combines it with also clearing it on _rollback. Before, we tried
to only call it through _rollback but that ruined callers of destroyed.
Doing it on both code paths covers destroyed being called from somewhere
else, e.g. strip.
Obsolete markers are now exchanged in smaller pieces that fit in a http header.
This changes is done to avoid 400 bad request error when exchanging obsolete
puskey over http.
The last key pushed is always hold by the "dump0" key to ensure an easy place to
hook for people who need it.
Before this change, push would incorrectly fast-path the bundle
generation when extinct changesets are involved, because they are not
added to outgoing.excluded. The reason to do so are related to
outgoing.excluded being assumed to contain only secret changesets by
scmutil.nochangesfound(), when displaying warnings like:
changes found (ignored 9 secret changesets)
Still, outgoing.excluded seems like a good API to report the extinct
changesets instead of dedicated code and nothing in the docstring
indicates it to be bound to secret changesets. This patch adds extinct
changesets to outgoing.excluded and fixes scmutil.nochangesfound() to
filter the excluded node list.
Original version and test by Pierre-Yves.David@ens-lyon.org
The repo.hiddenrevs set is updated with all extinct() changesets which aren't
descendants of either:
- the current working copy,
- a bookmark,
- a tag.
This set is always accessed through the repo for now. Having this set
carried by the changelog make it complicated to:
- initialize it, computing hidden set may involve revset call
- lazy compute it, (1) only the changelog can detect someone access it,
(2) only the repo have enought knowledge to compute it.
In later version I expect he changelog to apply filtering itself and the set to
be carried by changelog again.
They were previously inside the mercurial.phases module, but obsolete
logic will need them to exclude `extinct` changesets from pull and
push.
The proper and planned way to implement such filtering is still to apply a
changelog level filtering. But we are far to late in the cycle to implement and
push such a critical piece of code (changelog filtering). With Matt Mackall
approval I'm extending this quick and dirty mechanism for obsolete purpose.
Changelog level filtering should come during the next release cycle.
This change separates peer implementations from the repository implementation.
localpeer currently is a simple pass-through to localrepository, except for
legacy calls, which have already been removed from localpeer. This ensures that
the local client code only uses the most modern peer API when talking to local
repos.
Peers have a .local() method which returns either None or the underlying
localrepository (or descendant thereof). Repos have a .peer() method to return
a freshly constructed localpeer. The latter is used by hg.peer(), and also to
allow folks to pass either a peer or a repo to some generic helper methods.
We might want to get rid of .peer() eventually.
The only user of locallegacypeer is debugdiscovery, which uses it to pose as a
pre-setdiscovery client. But we decided to leave the old API defined in
locallegacypeer for clarity and maybe for other uses in the future.
It might be nice to actually define the peer API directly in peer.py as stub
methods. One problem there is, however, that localpeer implements
lock/addchangegroup, whereas the true remote peers implement unbundle.
It might be desireable to get rid of this distinction eventually.
User should resolve unstability locally before pushing the same way we encourage
user to merge locally instead of pushing a new remote head.
If we are to push obsolete changeset, at least one set of the pushed set will be
either obsolete or unstable. The check is narrowed to only heads.
As a part of migration to vfs, this patch invokes some file API
indirectly via vfs, while ensuring repository directory in the
constructor of "localrepository" class.
New file API are added to "scmutil.abstractopener" class, because they
are also used via other derived classes than "scmutil.opener".
But "join()" is not yet defined other than "scmutil.opener" class,
because it should not be used via other opener classes yet.
As a part of migration to vfs, this patch uses "self.vfs" instead of
"self.opener", while ensuring repository directory in the constructor
of "localrepository" class.
As a part of migration to vfs, this patch uses "self.root", which can
be recognized as the path relative to "self.vfs", instead of "path"
argument.
This fix allows to make invocations of "util.makedirs()" and
"os.path.exists()" while ensuring repository directory in
"localrepository.__init__()" ones indirectly via vfs.
But this fix also raises issue 2528: "hg clone" with empty destination.
"path" argument is empty in many cases, so this issue can't be fixed
in the view of "localrepository.__init__()".
Before this patch, it is fixed by empty-ness check ("not name") of
exception handler in "util.makedirs()".
try:
os.mkdir(name)
except OSError, err:
if err.errno == errno.EEXIST:
return
if err.errno != errno.ENOENT or not name:
raise
This requires "localrepository.__init__()" to invoke "util.makedirs()"
with "path" instead of "self.root", because empty "path" is treated as
"current directory" and "self.root" becomes valid path.
But "hg clone" with empty destination can be detected also in
"hg.clone()" before "localrepository.__init__()" invocation, so this
patch re-fixes issue2528 by checking it in "hg.clone()".
As a part of migration to vfs, this patch uses "self.wvfs.join()"
instead of "os.path.join()", while initialization of fields in the
constructor of "localrepository" class.
As a part of migration to vfs, this patch moves path expansion API
invocations in the constructor of "localrepository" to the constructor
of "opener", because the root path to the repository is very important
to handle paths using non-ASCII characters correctly.
This patch also rearrange initialization order of "wvfs" field,
because it is required to initialize "self.root".
As a part of migration to vfs, this patch adds "vfs" fields to
"localrepository" class.
This allows new codes to access current "opener" objects related to
repositories via "vfs" fields, so patches referring to "vfs" will
replace referring to "opener" in time.
This patch also adds initializations for "vfs" fields to
"statichttprepository" class derived from it, because its constructor
doesn't invoke the constructor of "localrepository", so "vfs" fields
should be initialized explicitly as same as "opener" fields: it has no
working directory, so "wvfs" field is not added.
Marker are now written as soon as possible but within a transaction. Using a
transaction ensure a proper behavior on error and rollback compatibility.
Flush logic are not necessary anymore and are dropped from lock release.
With this changeset, the obsstore is open, written and closed for every single
added marker. This is expected to be highly inefficient and batched write should
be implemented "quickly".
Another issue is that every flush of the file will invalidate the obsstore
filecache and trigger a full re instantiation of the repo.obsstore attribute
(including, reading and parsing entry). This is also expected to be highly
inefficient and proper filecache operation should be implemented "quickly" too.
A side benefit of the filecache issue is that repo.obsstore object is properly
invalidated on transaction abortion.
This is the second step toward incremental writing of marker inside a
transaction. The obsstore file is now handled append only.
Header writing have been extracted from _writemarkers.
Because the _writemarkers method have been dropped, the push code
directly reuse the serialised content of local repo `listkeys`. This
is not very pretty, but this part of the protocol still need major
improvement anyway.
This is the first step toward incremental writing of obsolete marker within a
transaction.
For this purpose, obsstore is now given its repo sopener. This make it able to
handles read and write to the obsstore file itself. Most IO logic is removed
from localrepo and handled by obsstore object directly.
This function augments strip to incrementally update the branchheads cache
rather than recompute it from scratch. This speeds up the performance of strip
and rebase on repos with long history. The performance optimization only
happens if the revisions stripped are all on the same branch and the parents of
the stripped revisions are also on that same branch.
This adds a few test cases, particularly one that reproduces the extra heads
that mpm observed.
_updatebranchcache used to use revlog.reachable. After the switch to
revlog.ancestors, we can now clean it up a bit and switch the algorithm from
nodes to revs.
Accepting a variable number of arguments as the old API did is
deeply ugly, particularly as it means the API can't be extended
with new arguments. Partly as a result, we have at least three
different implementations of the same ancestors algorithm (!?).
Most callers were forced to call ancestors(*somelist), adding to
both inefficiency and ugliness.
b67b333b0d8a attempted to force the filecaches in localrepo to reload
everything after a rollback. But simply clearing _filecache isn't enough,
invalidate() needs to be called before/after. localrepo._rollback calls
invalidate() already, so we clear the map right afterwards which ensures
everything will be reread.
Destroying history via strip used to invalidate the branchheads cache,
causing it to be regenerated the next time it is read. This is
expensive in large repos. This change converts strip to pass info to
localrepo.destroyed() to enable to it to incrementally update the
cache, improving the performance of strip and other operations that
depend on it (e.g., rebase).
This change also strengthens a bit the integrity checking of the
branchheads cache when it is read, by rejecting the cache if it has
nodes in it that no longer exist.
Similar to branch heads we introduce the notion of bookmarkheads.
Bookmarkheads are changests that are bookmarked with the given bookmark
or a diverged version
The original motivation was changectx.phase() had special logic to
correctly lookup in repo._phaserev, including invalidating it when
necessary. And at other places, repo._phaserev was accessed directly.
This led to the discovery that phases state including _phaseroots,
_phaserev and _dirtyphase was manipulated in localrepository.py,
phases.py, repair.py, etc. phasecache helps encapsulating that.
This patch replaces all phase state in localrepo with phasecache and
adjust related code except for advance/retractboundary() in phases.
These still access to phasecache internals directly. This will be
addressed in a followup.
Introduce manifestdict.withflags() to get a set of all files which have any
flags set, since these are likely to be a minority. Otherwise checking .flags()
for every file is a lot of dictionary lookups and is quite slow.
Introduce match.always() to check if a match object always says yes, i.e.
None was passed in. If so, mfmatches should not bother iterating every file in
the repository.
Cosmetic cleanups. Fix comment typo referring to the notion of multiple tips.
Make variable describing a generator end in 'gen'.
Fix another var containing a node not to end with 'rev'.
The fix introduced in 3509b9cf8f86 was only partially successful. It is correct
to turn dirstate 'm' merge records into normal/dirty ones but copy records are
lost in the process. To adjust them as well, we need to look in the first
parent manifest to know which files were added and preserve only related
records. But the dirstate does not have access to changesets, the logic has to
moved at another level, in localrepo.
Here is a script illustrating the previous behaviour:
The merge brings a new file 'b' from remote
$ hg merge 1 --debug
searching for copies back to rev 1
unmatched files in other:
b
resolving manifests
overwrite: False, partial: False
ancestor: 07f494440405, local: 540395c44225+, remote: 102a90ea7b4a
b: remote created -> g
updating: b 1/1 files (100.00%)
getting b
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
(branch merge, don't forget to commit)
Delete but do not remove b
$ rm b
$ hg st
! b
The commit succeeds
$ hg commit -m merge
$ hg parents --template "{rev} {desc|firstline} files: {files}\n"
3 merge files:
$ hg st
! b
b changes were ignored, but even b existence was ignored
$ hg manifest
a
This happens because localrepo.commitctx() checks the input ctx.files(), which
is empty for workingctx.files() only returns added, modified or removed
entries, and bypass files/manifest updates completely. So the committed
revision manifest is the same as its first parent one, not containing the 'b'
file.
This patch forces the commit to abort in presence of a merge and missing files.
test-merge4.t is modified accordingly as it was introduced to check hg was not
just terminating with a traceback (5cc0d3ba11f9).
bookmarks is copied to journal.bookmarks differently from how dirstate is
copied to journal.dirstate. The different way is less robust, which can render
the repo unpushable by other users if the first pushing user aborts their
transaction.
The underlying cause is that the copyfile method attempts an unnecessary chmod,
which fails if the user is not the owner of the journal.bookmarks file.
This patch makes the bookmarks journaling more consistent with the rest of the
journaling, and will allow users to update lingering journal.bookmarks files
that they're not the owners of.
New users of filecache use different names for the function used to compute
the runtime path of the cached file.
Users should subclass filecache and provide their own version of this
function to call the appropriate join function on 'obj' (an instance
of the class that its member function was decorated).
when pattern which does not match against any files in working context
is specified, current implementation of 'localrepository.status()'
decides whether warning message about it should be shown or not by
'f not in context'
this works correctly for 'file pattern', but not for 'directory
pattern', because 'f not in context' always returns True for
directories, even if they are related to the context.
this patch uses 'changectx.dirs()' to examine whether specified
pattern is related to the context as a directory or not.
Files are being replaced by rollback but the corresponding data in localrepo
isn't actually updated for things like bookmarks, phases, etc. Then when
rollback is done, the cache is updated thinking it has the most up-to-date
data, where in fact it is still pre-rollback.
We clear _filecache to force everything to be recreated.
Currently we have the following return codes if nothing is found:
commit incoming outgoing pull push
intended 1 1 1 1 1
documented 1 1 1 0 1
actual 1 1 1 0 0
This fixes the lower-right entry.
Verify uses repo.cancopy() to detect whether a repo is a plain old
local repo, so it was giving a confusing error message when secret
changesets were present.
When the contents of .hgsubstate are stale (either because they've
manually been tweaked or partial updates have confused it), we get
confused about whether it actually needs committing.
So instead, we actively consult the parent's substate and compare it
the actual current state when deciding whether it needs committing.
Side effect: lots of "committing subrepo" messages that didn't
correspond with real commits disappear.
This change is fairly invasive for a fairly obscure condition, so it's
kept on the default branch.
When mq changeset are secret, they don't appear in outgoing and won't be
pushed. So it's not necessary to abort the push.
The checkpush call is protected by lock to prevent race on phase.
If push failed we should not expect the pushed changeset to exist on remote.
The common set before the push is used for phase related operation instead of
common + missing.
Note:
* We still pull phase data even if push fails
* We still try to push data even if push fails (same than bookmark)
The ``discovery.prepush`` function was doing multiple things not related to
discovery. This changeset move some code into the ``localrepo.push`` method. The
old ``discovery.prepush`` function jobs is now restricted to checking for
multple head creation. It was then renamed ``discovery.checkheads``.
This new ``discovery.checkheads`` function may receive several other changes in
the future but we are a bit too much near the freeze for a wider refactoring.
New tags were written to .hgtags / .hglocaltags without updating or
invalidating the localrepo cache.
Before 462e6cfb1bac a lock was acquired soon after the new tags had been
written, and that invalidated the cache so the new tags for example could be
seen in pretxncommit hooks. With 462e6cfb1bac the lock had already been
acquired at this point and the missing cache invalidation was exposed.
The tag caches will now explicitly and immediately be invalidated when new tags
are added.
This commit add a whennodata list where extension can register a callback to be
called if no phase related data are found in the repository.
The goal is to ensure the existing extension that move phase data in 2.1 can
compute consistent phase boundary for existing repo.
The code now only exchange draft root and only care about movement related to
public//draft boundary.
There is multiple reason to simplify this code:
* Secret are never discovered anymore
* We decided to not support more the three existing phase
Removing phase index from pushkey (if ever decided) will be made in another commit.
This fix the lack phase movement when a locally secret changeset without added
children was pushed to the repository. In such case, this changeset would be
present in the bundle source, but not in the ``added`` variable.
This list will contains any node see in the source, not only the added one.
This is intended to allow phase to be move according what was pushed by client
not only what was added.
test-mq-cache.t did apparently look at stale cache content.
Testing with different locking mechanism happened to update the cache more
frequently and thus caused a test failure.
Simplifies client logic in multiple places since it encapsulates the
computation of the common and, more importantly, the missing node lists.
This also allows an upcomping patch to communicate precomputed versions of
these lists to clients.
The bugs seemed to show up when element not in future common changeset should
hold new hold phase data.
The whole phase push machinery was rewritten in the process.
current "localrepository._checknested()" uses specified path itself to
compare against subrepo pathes.
it is invoked from "hgsubrepo.subrepo()" or pathauditor (as callback),
and both use "os.sep" as separator.
this causes unexpected nesting check result, if subrepo configuration
uses "/" as path separator for sub repo path.
this path uses "/" to join path components (or apply "util.pconvert()"
on path) to normalize.
The contract for repo.destroyed() is that it is called whenever
changesets are destroyed, either by strip or by rollback. That
contract was inadvertently broken in 6c30b131b2ae, when we made a
chunk of code conditional on destroying one of the working dir's
parents. Oops: it doesn't matter *which* changesets are destroyed or
what their relationship is to the working dir, we should call
repo.destroyed() whenever we destroy changesets.
On Windows, we store symlinks as plain files with the link contents.
Via user error or NFS/Samba assistance, these files often end up with
'normal' file contents. Committing these changes thus gives an
invalid symlink that can't be checked out on Unix.
Here we filter out any modified symlink placeholders that look
suspicious when computing status:
- more than 1K (looks more like a normal file)
- contain NULs (not allowed on Unix, probably a binary)
- contains \n (filenames can't contain \n, very unusual for symlinks,
very common for files)
This changeset flips the default value of ui.commitsubrepos setting
from True to False and adds a --subrepos flag to commit.
The commit, status, and diff commands behave like this with regard to
recusion and the ui.commitsubrepos setting:
| recurses | recurses
| by default | with --subrepos
--------+---------------+----------------
commit: | commitsubrepo | True
status: | False | True
diff: | False | True
By changing the default from True to False, the table becomes
consistent in the two columns:
* without --subrepos on the command line, commit will abort if a
subrepo is dirty and status/diff wont show changes inside subrepos.
* with --subrepos, all three commands will recurse.
A --subrepos flag on the command line overrides the config settin.g
Older publish=True was:
1) Content of Publishing server are seen as public by client.
2) Any changegroup *added* to a publish=True server is public.
New definition are:
1) Content of Publishing server are seen as public by client.
2) Any changegroup *pushed* to a publish=True server is public.
See mercurial/phase.py documentation for exact final behavior
What is a "publishing repository"?
==================================
Setting a repository as "publishing" alter its behavior **when used as a
server**: all changesets are **seen** as public changesets by clients.
So, pushing to a "publishing" repository is the most common way to make
changesets public: pushed changesets are seen as public on the remote side and
marked as such on local side.
Note: the "publishing" property have no effects for local operations.
Old repository are publishing
=============================
Phase is the first step of a series of features aiming at handling mutable
history within mercurial. Old client do not support such feature and are unable
to hold phase data. The safest solution is to consider as public any changeset
going through an old client.
Moreover, most hosting solution will not support phase from the beginning.
Having old clients seen as public repositories will not change their usage:
public repositories where you push *immutable* public changesets *shared* with
others.
Why is "publishing" the default?
================================
We discussed above that any changeset from a non-phase aware repository should
be seen as public. This means that in the following scenario, X is pulled as
public::
~/A$ old-hg init
~/A$ echo 'babar' > jungle
~/A$ old-hg commit -mA 'X'
~/A$ cd ../B
~/B$ new-hg pull ../A # let's pretend A is served by old-hg
~/B$ new-hg log -r tip
summary: X
phase: public
We want to keep this behavior while creating/serving the A repository with
``new-hg``. Although committing with any ``new-hg`` creates a draft changeset.
To stay backward compatible, the pull must see the new commit as public.
Non-publishing server will advertise them as draft. Having publishing repository
the default is thus necessary to ensure this backward compatibility.
This default value can also be expressed with the following sentence: "By
default, without any configuration, everything you exchange with the outside is
immutable.". This behaviour seems sane.
Why allow draft changeset in publishing repository
=====================================================
Note: The publish option is aimed at controlling the behavior of *server*.
Changeset in any state on a publishing server will **always*** be seen as public
by other client. "Passive" repository which are only used as server for pull and
push operation are not "affected" by this section.
As in the choice for default, the main reason to allow draft changeset in
publishing server is backward compatibility. With an old client, the following
scenario is valid::
~/A$ old-hg init
~/A$ echo 'babar' > jungle
~/A$ old-hg commit -mA 'X'
~/A$ old-hg qimport -r . # or any other mutable operation on X
If the default is publishing and new commits in such repository are "public" The
following operation will be denied as X will be an **immutable** public
changeset. However as other clients see X as public, any pull//push (or event
pull//pull) will mark X as public in repo A.
Allowing enforcement of public changeset only repository through config is
probably something to do. This could be done with another "strict" option or a
third value config for phase related option (mode=public, publishing(default),
mutable)
You can get into trouble if you commit, update back to an older
changeset, and then rollback. The update removes your valuable changes
from the working dir, then rollback removes them history. Oops: you've
just irretrievably lost data running nothing but core Mercurial
commands. (More subtly: rollback from a shared clone that was already
at an older changeset -- no update required, just rollback from the
wrong directory.)
The fix assumes that only "commit" transactions have irreplaceable
data, and allows rolling back non-commit transactions as always. But
when rolling back a commit, check that the working dir is checked out
to tip, i.e. the changeset we're about to destroy. If not, abort. You
can get back the old (dangerous) behaviour with --force.
If the working dir parent was destroyed by rollback, then the old
behaviour is perfectly reasonable: restore dirstate, branch, and
bookmarks. That way the working dir moves back to an existing
changeset rather than becoming an orphan.
But if the working dir parent was unaffected -- say, you updated to an
older changeset and then did rollback -- then it's silly to restore
dirstate and branch. So don't do that. Leave the status of the working
dir alone. (But always restore bookmarks, because that file refers to
changeset IDs that may have been destroyed.)
- clarify how we parse undo.desc
- fix bad grammar in an error message
- factor out ui local
- rename some local variables
- standardize string quoting
This is extremely handy for those occasional circumstances where you
need to edit .hg/sharedpath manually, since modern Unix text editors
make it surprisingly difficult to create a text file with no trailing
newline.
The usual contract is that close() makes your writes permanent, so
atomictempfile's use of close() to *discard* writes (and rename() to
keep them) is rather unexpected. Thus, change it so close() makes
things permanent and add a new discard() method to throw them away.
discard() is only used internally, in __del__(), to ensure that writes
are discarded when an atomictempfile object goes out of scope.
I audited mercurial.*, hgext.*, and ~80 third-party extensions, and
found no one using the existing semantics of close() to discard
writes, so this should be safe.
