Consider following LDAP search query example:
(&(objectClass=Person)(|(uid=%s)(mail=%s)))
Right now on first login attempt Gogs will use the text supplied on login form
as the newly created user name. In example query above the text matches against
both e-mail or user name. So if user puts the e-mail then the new Gogs user
name will be e-mail which may be undesired.
Using optional user name attribute setting we can explicitly say we want Gogs
user name to be certain LDAP attribute eg. `uid`, so even user will use e-mail
to login 1st time, the new account will receive correct user name.
This speeds up single build/rebuild rather than install & build which compiles
everything twice, we just copy installed binary back to the project root.
Now, the dump cmd uses setting.CustomPath and setting.LogRootPath
instead of setting.WorkDir which was kind of broken if the gogs
binary was in a different directory than gogs data.
Additionally, the backup of setting.CustomPath directory is only done
if it exists.
Using a tmp dir makes gogs dump more robust to concurrent runs.
It also permits an easier cleaning of the tmp files (gogs-db.sql and
gog-repo.zip) by just removing the tmp dir.
As a side effect, it partially fix bugs on workdir.
Previously, 'gogs dump' created the archives in the current directory,
and tried to include these archives from the directory where the
gogs binary lies.
ex: if gogs binary is in /usr/bin/gogs, and gogs dump is run from /tmp/,
/tmp/gog-repo.zip is created, but gogs dump tried to include
/usr/bin/gogs-repo.zip.
The dump cmd did not check the return value of the z.AddFile or
z.AddDir when building the final archive.
It caused the dump command to succeed even if an error occurred.
The resulted dump archive could be corrupted/empty.
(errors could be various: removal by a concurrent process, disk full,
bugs in the dump cmd itself)
