During the migration of git repositories from v1 to v2, we have to
create an "scmm" config section with the repository id of the current
repository. If this does not happen, further write requests to this
repository will fail, because the hooks cannot determine the id.
However, the migration failed to write this configuration for non-bare
repositories. Therefore this fix checks beforehand, whether a '.git'
folder exists in the date directory. If this is the case, we assume that
this is a non-bare repository and write the config file inside this
folder.
Git uses LF for line breaks in diffs, not CR or other delimiters. When
we are using other delimiters for diving diff output into lines, too,
we can get errors because diff lines can contain CRs. When we try to
split such lines, we get exceptions because these lines cannot be parsed
The name 'workdir' is now only used for a concrete directory, where a
clone can be created in. The new name 'working copy' is used for a
concrete instance of such a directory, where a clone has been created in
for a repository stored in SCM-Manager. Such a working copy can be
cached and reused for further changes.
This fixes an error with a repository, where an added file has \r\r\n
as line breaks. This interesting combination is handled as a single
line break by the diff operation, while Java's Scanner implementation
handles this as two lines (the first one delimited by the first \r,
the second one delimited by \r\n. This led to empty lines inside the
diff, where we only expect lines that contain at least one character
(' ', '+' or '-'), and this in turn led to an index out of bounds
exception.
Now we handle each combination of any kind of new line delimiter
characters as a single delimiter. This should be safe, because, as
mentioned earlier, we always expect at least one character in a line
for a diff output.
Fixes the 'scm' protocol uri for Windows. Have to replace all
backslashes with forward slashes, add one more slash at the
beginning and remove the colon from the dirve.
