search

mnauw / git-remote-hg

Transparent bidirectional bridge between Git and Mercurial for Git
GNU General Public License v2.0
62 stars 5 forks source link

readme

'git-remote-hg' is the semi-official Mercurial bridge from Git project, once installed, it allows you to clone, fetch and push to and from Mercurial repositories as if they were Git ones:

git clone "hg::http://selenic.com/repo/hello"

To enable this, simply add the 'git-remote-hg' script anywhere in your $PATH:

wget https://raw.github.com/mnauw/git-remote-hg/master/git-remote-hg -O ~/bin/git-remote-hg chmod +x ~/bin/git-remote-hg

In Windows, you also need to put and an NTFS symbolic link named python2.exe somewhere on your $PATH pointing to your Python 2 executable:

mklink

That's it :)

Obviously you will need Mercurial installed.

At present, this "working copy"/fork <<add-features, adds the following features>> (and I would prefer it is indeed rather a "working copy" to be appropriately merged upstream):

See sections below or sidemarked notes for more details.

== Configuration ==

If you want to see Mercurial revisions as Git commit notes:

% git config core.notesRef refs/notes/hg

If you are not interested in Mercurial permanent and global branches (aka. commit labels):

% git config --global remote-hg.track-branches false

With this configuration, the 'branches/foo' refs won't appear.

If you want the equivalent of 'hg clone --insecure':

% git config --global remote-hg.insecure true

If you want 'git-remote-hg' to be compatible with 'hg-git', and generate exactly the same commits:

% git config --global remote-hg.hg-git-compat true

== Notes ==

Remember to run git gc --aggressive after cloning a repository, specially if it's a big one. Otherwise lots of space will be wasted.

The oldest version of mercurial supported is 1.9. For the most part 1.8 works, but you might experience some issues.

=== Pushing branches ===

To push a branch, you need to use the "branches/" prefix:

% git checkout branches/next

do stuff

% git push origin branches/next

All the pushed commits will receive the "next" Mercurial named branch.

Note: Make sure you don't have +remote-hg.track-branches+ disabled.

=== Cloning HTTPS ===

The simplest way is to specify the user and password in the URL:

git clone hg::https://user:password@bitbucket.org/user/repo

You can also use the http://mercurial.selenic.com/wiki/SchemesExtension[schemes extension]:

[auth] bb.prefix = https://bitbucket.org/user/ bb.username = user bb.password = password

Finally, you can also use the https://pypi.python.org/pypi/mercurial_keyring[keyring extension].

However, some of these features require very new versions of 'git-remote-hg', so you might have better luck simply specifying the username and password in the URL.

=== Submodules ===

Hg repositories can be used as git submodule, but this requires to allow the hg procotol to be used by git submodule commands:

git config protocol.hg.allow always

Or adding manually the following to your git configuration file:

[protocol "hg"] allow = always

This can be done per-repository, every time after a clone, or globally in the global .gitconfig (using the --global command-line option).

=== Caveats ===

The only major incompatibility is that Git octopus merges (a merge with more than two parents) are not supported.

Mercurial branches and bookmarks have some limitations of Git branches: you can't have both 'dev/feature' and 'dev' (as Git uses files and directories to store them).

Multiple anonymous heads (which are useless anyway) are not supported; you would only see the latest head.

Closed branches are not supported; they are not shown and you can't close or reopen. Additionally in certain rare situations a synchronization issue can occur (https://github.com/felipec/git/issues/65[Bug #65]).

[[limitations]] Limitations of the remote-helpers' framework apply. In particular, these commands don't work:

Another limitation is that if git log reports a rename, this will not survive the push and Mercurial will not be aware of a rename (and similarly so for copy). Though Mercurial would know about it if you manually ran git-format-patch followed by a hg apply -s, which is not the nice way to go obviously.

Actually, scratch the limitations above ascribed to the remote-helpers framework. They are not limitations of the framework, but are due to how the original implementation of 'git-remote-hg' interacts with it. Using the remote-helpers framework in only a slightly different way has none of the above limitations. See the <<no-limitations, relevant section>> below for more details.

== Other projects ==

There are other 'git-remote-hg' projects out there, do not confuse this one, this is the one distributed officially by the Git project (though actually no longer so nowadays):

For a comparison between these and other projects go https://github.com/felipec/git/wiki/Comparison-of-git-remote-hg-alternatives[here].

[[no-limitations]] == Limitations (or not) ==

If interested in some of technical details behind this explanation, then also see the relevant section in 'git-remote-hg' manpage. Otherwise, the general idea is presented here.

More precisely and simply, the <<limitations, mentioned limitations>> are indeed limitations of the export capability of https://www.kernel.org/pub/software/scm/git/docs/gitremote-helpers.html[gitremote-helpers(1)] framework. However, the framework also supports a push capability and when this is used appropriately in the remote helper the aforementioned limitations do not apply. In the case of export capability, git-core will internally invoke git-fast-export and the helper will process this data and hand over generated changesets to Mercurial. In the case of push capability, git informs the helper what (refs) should go where, and the helper is free to ponder about this and take the required action, such as to invoke git-fast-export itself (with suitable options) and process its output the same way as before (and over to Mercurial).

And so;

To tweak how 'git-remote-hg' decides on a copy/rename, use e.g:

% git config --global remote-hg.fast-export-options '-M -C -C'

[[add-features]] == Additional Features ==

=== Miscellaneous Tweaks === Other than <<no-limitations, removing the limitations>> as mentioned above, a number of issues (either so reported in https://github.com/felipec/git-remote-hg/issues[issue tracking] or not) have been addressed here, e.g. notes handling, fetch --prune support, correctly fetching after a strip on remote repo, tracking remote changes to import (if any) in a safe, robust and efficient way, etc. Some of these have been highlighted above.

For example, the refs/hg/... refs are really an implementation detail that need not clutter up the (visible) ref space. So, in as much as they are still relevant, these are now kept elsewhere out of sight. If somehow your workflow relies on having these in the old place:

% git config --global remote-hg.show-private-refs true

More importantly, a significantly more efficient workflow is achieved using one set of shared marks files for all remotes (which also forces a local repo to use an internal proxy clone). The practical consequence is that fetching from a newly added remote hg repo does not require another (lengthy) complete import (as the original clone) but will only fetch additional changesets (if any). The same goes for subsequent fetching from any hg remote; what was fetched and imported from some remote need not be imported again from another. Operating in this shared mode also has the added advantage of correctly pushing after a strip on a remote. This shared-marks-files behaviour is the default on a fresh repo clone. It can also be enabled on an existing one by the following setting.

% git config --global remote-hg.shared-marks true

Note, however, that one should then perform a fetch from each relevant remote to fully complete the conversion (prior to subsequent pushing).

Some Mercurial names (of branches, bookmarks, tags) may not be a valid git refname. See e.g. man git-check-ref-format for a rather involved set of rules. Moreover, while a slash / is allowed, it is not supported to have both a parent and parent/child branch (though only the latter is allowed). Even though it is not quite (bidirectionally) safe, a (percentage) URL encoding (with some additional twist) is performed to obtain sane git refnames, at least so for most cases. If some nasty cases still slip through, then likely only a few instances (out of a whole Mercurial repo) are problematic. This could be handled by a single-branch clone and/or configuring a suitable refspec. However, it might be more convenient to simply filter out a few unimportant pesky cases, which can be done by configuring a regural expression in following setting:

% git config remote-hg.ignore-name nasty/nested/name

Recall also that a config setting can be provided at clone time (command line using --config option).

% git config --global remote-hg.remove-username-quotes false

By default, for backwards compatibility with earlier versions, git-remote-hg removes quotation marks from git usernames (e.g. 'Raffaello "Raphael" Sanzio da Urbino raphael@example.com' would become 'Raffaello Raphael Sanzio da Urbino raphael@example.com'). This breaks round-trip compatibility; a git commit by an author with quotes would become an hg commit without, and if re-imported into git, would get a different SHA1.

To restore round-trip compatibility (at the cost of backwards compatibility with commits converted by older versions of git-remote-hg), turn 'remote-hg.remove-username-quotes' off.

=== Helper Commands ===

Beyond that, a 'git-hg-helper' script has been added that can aid in the git-hg interaction workflow with a number of subcommands that are not in the purview of a remote helper. This is similar to e.g. 'git-svn' being a separate program altogether. These subcommands

See the helper script commands' help description for further details. It should simply be installed ($PATH accessible) next to 'git-remote-hg'. Following git alias is probably also convenient as it allows invoking the helper as git hg:

% git config --global alias.hg '!git-hg-helper'

With that in place, running git hg gc <remote> after initial fetch from (large)

will save quite some space in the git marks file. Not to mention some time each time it is loaded and saved again (upon fetch). If the remote is ever pushed to, the marks file will similarly be squashed, but for a fetch-only the aforementioned command will do. It may also be needed to run aforementioned command after a `git gc` has been performed. You will notice the need when `git-fast-import` or `git-fast-export` complain about not finding objects ;-) In addition, the helper also provides support routines for `git-remote-hg` that provide for increased (or at least safer) git-hg bidirectionality. Before explaining how it helps, let's first elaborate on what is really meant by the above _bidirectionality_ since it can be regarded in 2 directions. From the git repo point of view, one can push to a hg repo and then fetch (or clone) back to git. Or one could have fetched a changeset from some hg repo and then push this back to (another) hg clone. So what happens in either case? In the former case, from git to hg and then back, things work out ok whether or not in hg-git compatibility mode. In the latter case, it is very likely (but ultimately not guaranteed) that it works out in hg-git compatibility mode, and far less likely otherwise. Most approaches on bidirectionality try to go for the "mapping" way. That is, find a way to map all Mercurial (meta)data somewhere into git; in the commit message, or in non-standard ways in extra headers in commit objects (e.g. the latest hg-git approach). The upside of this is that such a git repo can be cloned to another git repo, and then one can push back into hg which will/should turn out ok. The downside is setting up such a mapping in the first place, avoiding the slightest error in translating authors, timestamps etc, and maintaining all that whenever there is some Mercurial API/ABI breakage. The approach here is to consider a typical git-hg interaction workflow and to ensure simple/safe bidirectionality in such a setting. That is, you are (obviously) in a situation having to deal with some Mercurial repo and quite probably with various clones as well. The objective is to fetch from these repos/clones, work in git and then push back. And in the latter case, one needs to make sure that hg changesets from one hg clone end up *exactly* that way in another hg clone (or the git-hg bridge usage might not be so appreciated). Such pushes are probably not recommended workflow practice, but no accidents or issues should arise from any push in these circumstances. There is less interest in this setting, however, for (git-wise) cloning around the derived git repo. Now, depending on your workflow and to ensure the above behaves well, following setting can be enabled as preferred: -------------------------------------- % git config --global remote-hg.check-hg-commits fail % git config --global remote-hg.check-hg-commits push -------------------------------------- If not set, the behaviour is as before; pushing a commit based on hg changeset will again transform the latter into a new hg changeset which may or may not match the original (as described above). If set to `fail`, it will reject and abort the push. If set to `push`, it will re-use the original changeset in a Mercurial native way (rather than creating a new one). The latter guarantees the changeset ends up elsewhere as expected (regardless of conversion mapping or ABI). Note that identifying and re-using the hg changeset relies on metadata (`refs/notes/hg` and marks files) that is not managed or maintained by any git-to-git fetch (or clone). As such (and as said), this approach aims for plain-and-simple safety, but only within a local scope (git repo). === Mercurial Subrepository Support === Both Git and Mercurial support a submodule/subrepo system. In case of Git, URLs are managed in `.gitmodules`, submodule state is tracked in tree objects and only Git submodules are supported. Mercurial manages URLs in `.hgsub`, records subrepo state in `.hgsubstate` and supports Git, Mercurial and Subversion subrepos (at time of writing). Merely the latter diversity in subrepo types shows that somehow mapping Mercurial "natively" to git submodules is not quite evident. Moreover, while one might conceivably devise such a mapping restricted to git and hg subrepos, any such would seem error-prone and fraught with all sorts of tricky cases and inconvenient workflow handling (innovative robust suggestions are welcome though ...) So, rather than overtaking the plumbing and ending up with stuffed drain further on, the approach here is (again) to keep it plain-and-simple. That is, provide some git-ish look-and-feel helper script commands for setting up and manipulating subrepos. And so (if the alias mentioned above has been defined), `git hg sub` provides commands similar to `git submodule` that accomplish what is otherwise taken care of by the Mercurial subrepo support. The latter is obviously extended to be git-aware in that e.g. a Mercurial subrepo is cloned as a git-hg subrepo and translation back-and-forth between hg changeset id and git commit hash is also performed where needed. There is no support though for Subversion subrepos. As with the other commands, see the help description for the proper details, but the following example session may clarify the principle: -------------------------------------- % git clone hg::hgparentrepo # bring in subrepos in proper location: % git hg sub update # do some work % git pull --rebase origin # update subrepo state: % git hg sub update # do work in subrepo and push % ( cd subrepo && git push origin HEAD:master ) # fetch to update refs/notes/hg (or enable remote-hg.push-updates-notes) % ( cd subrepo && git fetch origin ) # update .hgsubstate to subrepo HEAD: % git hg sub upstate % git add .hgsubstate # add more, commit and push as intended -------------------------------------- Note that the refspec `HEAD:master` is needed if working with detached `HEAD` in subrepo, and that pushing such refspec is actually supported now in a git-hg subrepo as explained <>. == Contributing == Please file an issue with some patches or a pull-request.