Merge tip from an unrelated repository with Mercurial - mercurial

I have two unrelated repositories 'public' and 'develop':
prj/
public/
develop/
'develop' has lots of commits, 'cause here is where I work. Maybe even multiple heads
From time to time I want publish a snapshot of the development repository.
From the public folder I could do this:
>hg pull -f ../develop
>hg merge
>hg commit -m "alpha2"
But this will also pull the complete changeset history from 'develop' to 'public' (which is not what I want).
I could also delete all files from 'public', except for the '.hg' subfolder.
Then manually copy all files from the 'develop' directory and do a
>hg commit -m "alpha2"
But then I have to 'add' new files, 'remove' obsolete files and 'rename' moved files again.
Using the -A option with commit would blindly add/remove all files, even if they were uncontrolled in the 'development' repository.
There must be a more efficient way to do this ;-)

I'd suggest you just move over the full history, but if you can't stomach that for some reason then you could do any of these things:
use ConcatenatingChangesets to join all the changesets from develop into a single resulting changeset in public -or-
delete all files in public, make a 'hg archive' in develop, and expand it in public -- that will get you controlled files only -or
use 'hg manifest' in develop to power xargs to move over only files you want -or-
do a 'hg diff -r last-changeset-in-public -r tip' in develop, and then apply that resulting diff using 'hg import' in public and you'll have the full changes as a single changeset in public
Any of those should do the same thing, but all of them are a little ugly because, in general, throwing away history is hard to do.

Related

Hg clone only public commits

We have a big Hg repo, hosted in a remote location. Performing an hg clone from this master repo takes about an hour. What we generally do to speed things up is to hg serve a local repo of a colleague, hg clone http://colleague-machine, and then change de default path in .hg/hgrc to the address of the master repo.
This is all well and good, but this workaround has one drawback: because we are cloning the repo of a developer, some draft commits can be cloned along with the public ones. Moreover, these commits become public in the cloned repo, making them indistinguishable from the others.
One possibility I found is to make the developer's repo non publishing, in order to preserve the phases of the commits and to remove them later on. Another possibility is to create a bundle containing only the public commits, instead of cloning directly.
These methods are more complex to explain and to document. Is there an option for hg clone to clone only the public commits? I tried with hg clone -r "public()", but clone does not take a revset, just a regular commit identifer. Alternatively, is there an option for hg serve to serve only the public commits?
Throw disk space at the problem: just keep a local mirror clone that you update regularly.
Cloning the "true master" is slow because it's far away over a slow link. But updating the mirror is fast because, while the true master is far away over a slow link, little data needs to traverse it; and cloning the mirror is fast, and gets you the state of the true master as of the last time the mirror was updated.
As you mention, you can then just replace the default path (and maybe run a subsequent hg pull to pick up anything not-yet-mirrored, if needed). Your new clone is then the same as it would have been, had you cloned from the far-away slow true master, except that it went fast.
Git has this kind of cloning built in, as what's called a reference clone. You point your git clone process at two repositories: the true source, and the "close and fast" reference. It gets hash IDs from the true source but then uses the close-and-fast reference's storage for its data. You can then choose to continue to rely on the reference (default) or "dissociate" from the reference so that your clone is independent. It needs this dissociate operation because it can do a somewhat dangerous path-name-based "link" (not really a link in the sense of hard links; more an in-Git analogue to symbolic links) to the original, and does so by default here.
I don't think Mercurial has anything equivalent "out of the box". I imagine it should be relatively easy to write as an extension, though, if you are up for that sort of thing. You wouldn't need --dissociate at all, it would be the default wherever hard links are not feasible.
One way to do this is to use hg clone -r <rev> where <rev> is public. That will ensure that you won't get any draft commits, although you will miss any branches that aren't ancestors of <rev>.
I don't think there's a generic way to clone only public changes. It might be possible via a server-side extension or in-process hook though.
I ended up using a combination of hg serve option and hg strip.
On the existing repository:
hg serve --config phases.publish=False --port 0 --prefix repo-name
On the target machine:
hg clone <address printed by `hg serve`>
cd repo-name
hg strip -r "draft()"
The phases.publish=False config makes the repo non-publishing, and thus preserves the phase of the commits that are cloned. Now that the phases are kept on the target machine, it is easy to strip them off after the clone.

Add a parent to the original changeset in Mercurial

I have a project with 24 months of source control history in a Mercurial repository.
I've recently found some old tarballs of the project that predate source control, and i think they would be useful to import into the repository as "pre-historic" changesets.
Can i somehow add a parent to my initial commit?
Alternatively, is it possible to re-play my entire repository history on top of the tarballs, preserving all metadata (timestamps etc)?
Is it possible to have the new parent commits use the timestamps of these old tarballs?
You can use the convert extension to build a new repository where the tarballs are imported as revisions before your current root revision.
First, you import the tarballs based on the null revision:
$ hg update null
$ tar -xvzf backup-2010.tar.gz
$ hg addremove
$ hg commit -m 'Version from 2010'
$ rm -r *
$ tar -xvzf backup-2011.tar.gz
$ hg addremove
$ hg commit -m 'Version from 2011'
I'm using addremove above to give Mercurial a chance to detect renames between each tarball (look at the --similarity flag to fine-tune this and use hg rename --after by hand to help Mercurial further). Also, I remove all the files in the working copy before importing a new tarball: that way the next commit will contain exactly the snapshot present in the tarball you unpack.
After you've imported all the tarballs like above, you have a parallel history in your repository:
[c1] --- [c2] --- [c3] ... [cN]
[t1] --- [t2] --- [tM]
Your old commits are c1 to cN and the commits from the tarballs are t1 to tM. At the moment they share no history — it's as if you used hg pull -f to pull an unrelated repository into the current one.
The convert extension can now be used to do a Mercurial to Mercurial conversion where you rewrite the parent revision of c1 to be tM. Use the --splicemap flag for this. It needs a file with
<full changeset hash for c1> <full changeset hash for tM>
Use hg log --template '{node} ' -r c1 -r tM > splicemap to generate such a file. Then run
$ hg convert --splicemap splicemap . spliced
to generate a new repository spliced with the combined history. The repository is new, so you need to get everybody to re-clone it.
This technique is similar to using hg rebase as suggested by Kindread. The difference is that convert wont try to merge anything: it simply rewrites the parent pointer in c1 to be tM. Since there is no merging involved, this cannot fails with weird merge conflicts.
You should look at using rebase. This can allow you to make the changes the 2nd changeset on your repo ( you have to rebase from the 1st ).
https://www.mercurial-scm.org/wiki/RebaseExtension
However, note that if there are other clones of this repo existing ( such as for fellow developers, or on a repo server ), you will have issues with them pulling the revised repo. You will probably have to co-ordinate with the owners of those clone's to get all work into a single clone, rebase that clone, and then have everyone re-clone from the revised clone. You will also have to change the phase the of the changesets.
https://www.mercurial-scm.org/wiki/Phases
Honestly though, I would just add them to your 'modern-day' repo, I don't think making them pre-historic would give you any notable advantage over adding them to the top.

repo cloned from one remote branch needs to be commited to a different remote branch

We have a master repository located on a separate server. I originally cloned the default branch and made my changes locally. I have locally commited those changes. However, there has been a branch created on the master repository that I would like to push my changes to. Below is the description of my attempt at getting this accomplished.
I have cloned the branch. I am trying to export my changes from local default like so:
C:\hg\default>hg export -g -o mypatch -r tip
and when trying to import them into the clone of the new branch, I get the following:
C:\hg\newBranch>hg import C:\hg\default\mypatch
applying C:\hg\Fill1\mypatch
patching file .hgignore
Hunk #1 FAILED at 11
1 out of 1 hunks FAILED -- saving rejects to file .hgignore.rej
abort: patch failed to apply
I can manually fix the .hgingore.rej file just fine. The problem is that the patch also contains files that were moved. Instead of the files showing as moved, I get the following when running hg status:
C:\hg\newBranch>hg status -C
M someOtherFilesThatLookAsExpected.txt
! originalLocaion\fileA.txt
? newLocation\fileA.txt
This missing and new status is for all files that were moved in the commit contained the applied patch. Am I doing something wrong? Do I always have to manually move files when applying a patch? Is there an easier way to accomplish this branch transfer?
That's a bit difficult to answer without knowing more about your repository structure, but here's how I'd go about it without knowing more. I'm assuming that the reason for the conflict is that there are conflicting changes in the same branch of the repository.
First, get the contents of the newBranch repository:
cd c:\hg\default
hg pull c:\hg\newBranch
Then, either merge or rebase your changes on top. If you are working on the same branch, then just using
hg pull --rebase c:\hg\newBranch
in lieu of the regular pull should do (assuming you have rebasing enabled). Otherwise, do an explicit merge or rebase of the two heads that you need to reconcile. Finally, do:
hg push -r tip c:\hg\newBranch
in order to get your (now reconciled) changes back into newBranch.
Unless you have very specific and unusual requirements, push and pull should be your normal way to sync repositories or part of them (note that using -r will only push/pull the respective branch). Export/import are rather low-level mechanisms that may not give you the benefits of the standard machinery that handles renames, three-way merging logic, etc.

switch branches in hg while preserving files

While developing my software i create a set of test-cases which an be automatically done via make tests. To keep things organized i put those in a sub-directory tests. To keep my commits organized as well and not to clutter my development branches, i created an extra branch tests where i commit the test units.
My problem is: when i commit the units to the test branch and switch back to development, said tests are deleted
D:\Project>hg branch test
D:\Project>edit...
D:\Project>hg add
D:\Project>hg commit
D:\Project>hg up dev
2 files updated, 0 files merged, 3 files removed, 0 files unresolved
How can i preserve those files?
(i tried the solution for Mercurial: Switch working directory to branch without losing changes? but it still deletes the files)
EDIT See my own answer below
Those test files don't exist in your development branch, so when you check out the development branch they won't be in the working directory. DVCSs like Mercurial and git don't let you have one directory checked out at one revision/branch and another one at another revision/branch.
The answers to which you linked are for bringing the changed and their files over into another branch, they're not what you're asking for.
So the way you want to do it (separate branches) isn't going to work, but there are plenty of other/better options
One choice would be to make your tests a patch that's managed by a mq (Mercurial Queues) repository, which can itself be versioned. Then when you where in your dev branch and wanted to run the tests you'd do:
hg qpush # tests show up
... run tests, edit tests ..
hg qrefresh # save the changes you made to the test
hg qpop # tests vanish again
MQ is poweful, but sometimes a little hard to wrap your head around.
Another choice would be to make your tests a parent repository and your actual code a child repository of that parent. Your disk layout would look like:
parent/
tests/
existingrepo/ # <-- the repo you already have w/o tests
Then people could clone and you could push the existingrepo w/o tests, but the outer repo would include a pointer to it and the two would be versioned in lockstep. Again, sort of tricky, but has some nice results.
The third option, and my preferences is to get over the "To keep my commits organized as well and not to clutter my development branches" mentality. Tests are just as important as the primary code, they should be versioned with the code, and they're not cluttering anything, they're providing valuable tools to help comprehend what the code is doing. You can always use hg log --exclude tests/ to see a history that excludes them when that's convenient.
To take that plunge just do:
hg update development
hg merge tests
and you're good to go.
when i commit the units to the test branch and switch back to development, said tests are deleted
It's expected and correct result: branches store diverged lines of development, you add something only in test branch, before merge this data will exist only in branch on creation
In order "To keep my commits organized as well and not to clutter my development branches...", but have tests in development branch you can use at least two ways:
Periodically merge tests branch to development (all changes in tests will appear in development), but use log only for changesets in development branch, maybe without mergesets from tests. It's rather simply revset, you can even write log command with revset into aliases and use aliased command when needed
As Ry4an mention, you can instead of branch-separtion, perform repository-separation with subrepo|guestrepo technique - if tests are stored in subdir of main repo, convert /tests into nested repository -> subrepository and have two independent, but linked repos (future reading: Subrepository in Mercurial wiki and Subrepositories in Mercurial Kick Start Exercises)
to achive what i wanted i eventually created a subdirectory for the repo:
project/
main.cpp
makefile
.repo/
.hg/
and created a new make target:
REPO := .repo
init: $(REPO) $(REPO)/.hg
$(REPO):
mkdir -p $(REPO)
$(REPO)/.hg:
hg init $(REPO)
commit : | $(REPO) $(REPO)/.hg
ifdef BRANCH
hg -R $(REPO) update -C $(BRANCH)
endif
find . -regex "^\./[^.].*" -exec cp --parents {} ./$(REPO) \;
# the regex prevents "hidden" dot-folder from copying
# one could use --link to save time and drive usage/space
# but i am concerned about hg auto merging and overriding (per hardlink)
# my actual changes
hg -R $(REPO) commit $(EXTRA)
so i could just issue make commit BRANCH=tests to commit to arbitrary branch without losing all changes not relevant to the branch. Implementing of hg add as wipe .repo, copy file there hg -R .repo add $(filename) is left as exerciser for the reader

Mercurial Patch Creation and Usage

I have come across a problem that I "think" can only be resolved using patches.
I cloned a project from our main repository, made quite a few changes (updates, deletion of files & directory and additions) to it. These changes are not even committed. The problem is, project from the main repository has been deleted/removed and recreated as a new project (name is same, all the directory structures everything is same as before). I cloned that project again from the main repository and would like to transfer all my uncommitted changes to it.
I am still exploring the hg patch to resolve that. It would be helpful if someone could confirm that creating and adding a patch IS the right approach to this, any resources explaining the process would be of great help.
You're correct — a patch is what you need to transfer the information from one repository to another (unrelated) repository. This will work since the files are the same, as you note.
So, to transfer your uncommitted changes from your old clone, you do
$ hg diff -g > uncommited.patch
$ cd ../new
$ hg import --no-commit ../old/uncomitted.patch
That will restore the information saved in the patch. This includes information about files that are added or renamed in the old clone.
The following steps can be performed with a standard Mercurial install:
Commit the changes in your local repository. Note the revision number.
Use "hg export -r REV >patch.diff" to create a patch.
Clone the new repository.
Use "hg import patch.diff" to apply the patch to the new repository.
Example
C:\>hg init example
C:\>cd example
C:\example>echo >file1
C:\example>hg ci -Am file1
adding file1
C:\example>hg clone . ..\example2
updating to branch default
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
C:\example>rd /s/q .hg
C:\example>hg init
C:\example>hg ci -Am same-but-different
adding file1
At this point example and example2 have identical contents, but the repositories are unrelated to each other due to deleting and reinitializing the .hg folder.
Now make some changes and commit them in one of the repositories, then export them as a patch:
C:\example>echo >>file1
C:\example>echo >file2
C:\example>hg ci -Am changes
adding file2
C:\example>hg export -r 1 >patch.diff
Below shows that the other repository can't pull the changes, because of the reinitialization. It can, however, apply the patch successfully:
C:\example>cd ..\example2
C:\example2>hg pull
pulling from c:\example
searching for changes
abort: repository is unrelated
C:\example2>hg import ..\example\patch.diff
applying ..\example\patch.diff
I would first make copies of everything so you have a way of backtracking.
Then, in the working copy with the changes, I would first delete the .hg directory, then copy in the .hg directory from the new repo. This basically transfers all of the changed files into the new repo without the need to delete any files and directories.
You will still need to tell the repo about whether to remove any files marked as missing. You will also have to handle renames manually. If this is a small number of operations, it's easier than trying to use the patch method.
Once this is done, commit your changes and push, if necessary.
seems like what you want is patch queues. In that you have uncommitted changes, and you want to pull from the new repo before committing them....
$ hg qinit -c # initialize mq for your repo containing the uncommitted changes
$ hg qnew name_of_patch # create patch that contains your uncommitted changes
$ hg qpop # resets your working dir back to the parent changeset
no worries though, your changes are safe and sound in .hg/patches/name_of_patch to see for yourself.....
$ cat .hg/patches/name_of_patch
now pull in the new repo
$ hg pull -u http://location.of.new/repo # pull in changes from new repo update working dir
$ hg qpush # apply your uncommitted changes to new repo
If you are lucky you will have no merge conflicts and you can go ahead and commit the patch by....
$ hg qfinish -a # change all applied patches to changeset
And then if you want....
$ hg push http://location.of.new/repo
If the repos are unrelated, just init a patch repo on your new repo. and manually copy the patch in and add it to .hg/patches/series file.
assuming patch was created. clone new repo
$ hg clone http://location.of.new/repo ./new_repo
init patch repo
$ cd ./new_repo && hg qinit -c
copy patch
$ cp ../old_repo/.hg/patches/name_of_patch .hg/patches/
edit series file using an editor of some sort
$ your_favorite_editor .hg/patches/series
name_of_patch # <---put this in the series file
apply your patch to new repo
$ hg qpush
if no merge conflicts and you are convinced it works
$ hg qfinish -a
If the layout is the same, you can just copy all the files over (excluding .hg) and then use hg addrem.
Try to look into the MQ plugin, it does exactly this if I recall. I've never had a use for that though, so I can't say.
If the old repository was simply moved/cloned to a new URL then you could simply change the remote repository you talk to the new one.
If, however, it was recreated from the ground up (even with the same structure) then I don't believe Mercurial has any built-in functionality to help you here. Mercurial patches reference specific changesets which won't exist in your new repository.
You could use a merge tool to perform the diff and bring across any changes you made.
Edited To answer the question in the comment:
When you clone the repository you are taking a complete snapshot of the entire change history - along with the associated change-set IDs, etc.
Mercurial tracks changes by change-sets to the repository, rather than at the file level like Subversion.
If you clone, then you can easily push/merge into another repository that was also cloned from the same source.
If you recreated the repository then the change IDs won't match, and can't be merged in Hg.
The only option in this scenario would be to use a Merge tool which will let you see mismatches in files/folder structure.
Also: Worth pointing out http://hginit.com/ because it explains (indirectly) some of this.