Whenever using `clone` command inside YADM_WORK sub-directory, we need
to checkout the correct repo contents.
Steps to reproduce:
```bash
mkdir $HOME/subdir
cd $HOME/subdir
yadm clone --bootstrap "<repo-url>"
yadm status
```
The new test Docker image has a newer gnupg which does not behave the
same way, handling invalid passwords. This type of error is simulated
using an ICP error in the pinentry protocol.
A local class is set with:
$ yadm config local.class cls1
More classes can be added with:
$ yadm config --add local.class cls2
$ yadm config --add local.class cls3
Any of cls1, cls2 and cls3 can be used in an alternate condition.
For templates, the existing variable yadm.class/YADM_CLASS is set to
the last class (i.e. cls3) to remain compatible with how it works
today and with what the following command gives:
$ yadm config local.class
For the default template processor there is no explicit yadm.classes
variable. Instead a yadm.class condition will check against all
classes.
For the other processors, a new template variable YADM_CLASSES will be
set to all classes separated by newline. For jinja2 templates a class
can be checked with: {%- if "cls" in YADM_CLASSES.split("\n") %}
For esh templates the logic is a bit more complex, but it is possible
to do.
Fixes#185.
Instead of duplicating the permissions on the temp file, the permissions
are duplicated on the output file directly. If the output file exists as
read-only, it is made writeable first.
There are some environments which don't allow the mv to work if the file
itself is read-only.
Instead of doing work to find the default branch just to be able to
set up the repository before doing a fetch, do a "git clone" and let
git handle it.
Use -c core.sharedrepository=0600 to get the same result as
--shared=0600 passed to init.
Use --separate-git-dir to get the git directory in $YADM_REPO. Use a
temporary dir as work tree and remove it right after the clone is
done.
When the clone is done, iterate over all missing files in $YADM_WORK
and perform a checkout. If local files exists that differ compared
with the cloned ones the local files are left intact and the user is
instructed to deal with the conflicts.
This makes it possible to run e.g. "yadm -Y foo introspect repo
2>/dev/null" and get an empty output instead of getting the error
message about foo not being fully qualified.
Start with doing "submodule absorbgitdirs" as otherwise "submodule
deinit" will fail if a module has been cloned first and later added as
a submodule (as it will then contain the .git dir instead of it being
under the superprojects .git dir).
Then try to deinit the submodules before moving the repo and abort the
upgrade if it fails for any submodule. Then do the move and finally
initialize the submodules that were initialized before the upgrade.
See #285
to make it easier to use yadm in the container. Set it in the system
config (i.e. /etc/gitconfig) to avoid getting a clash if one wants to
clone an existing yadm repo that contains a .gitconfig.
