tree 253b25125920feb844db3d8c1ffc8a1b2e737696
parent d63a44c2eca88bffbbe126071fd51cdc0aa76990
author Neels Hofmeyr <neels@hofmeyr.de> 1542059048 +0100
committer Neels Hofmeyr <neels@hofmeyr.de> 1542728969 +0100

gits: use @{u} to get upstream branch, not origin/%s

git has an internal concept of a branch's upstream branch, and the remote need
not be 'origin', and also, the upstream branch name may differ completely. So,
use git's {u} keyword to obtain the upstream branch name.

This removes all 'origin' strings from gits. Also, git_branch_exists() is no
longer needed, drop it.

Also remove a couple of default arguments which aren't ever used because we
always pass arguments anyway.

In the case of git_ahead_behind(), we would have use for a default
branch_upstream=None to imply calling git_branch_upstream(), but then during
rebase, if no upstream exists, we would invoke it twice to get None a second
time. So just call the function explicitly. I thought about returning an empty
string instead of None, but it's too convoluted.

In git_output(), pipe STDERR to STDOUT, because every time we parse a remote
ref (git ref-parse --abbrev-ref '%s{u}') and there is no remote branch, git
prints 'Fatal: there is no remote bla bla' on stdout, and that error is
expected / ok to happen, so it just clutters the 'gits' output. The easiest way
to achieve silence then is to pipe to STDOUT, IIUC otherwise we'd have to use
Popen() and communicate()... In the case of error, subprocess raises an
exception and we see that an error happens. In the ref-parse case we can simply
catch the exception and be quiet.

Change-Id: Ife146903ae1323a4e568587ccfd4018725e9d719