Bastian Blank wrote: > > Also, I have no idea what "Pex3w7XYJ65vIFm2" is. git-annex only uses > > the hostname, or a md5sum of the hostname for the control socket. > > This seems to be neither, unless you have very strange hostnames. > > It is something ssh adds to avoid having a socket where noone listens. > So there are two bugs: > - ssh does not take this random stuff into account when checking the > length of the ControlPath option. > - git-annex assuming it have the complete 100 bytes, while ssh takes 16 > for itself.
bind(4, {sa_family=AF_FILE, path="thisfile.E4vQpVBOQLyfYY8p"}, 28) = 0
umask(022) = 0177
listen(4, 64) = 0
link("thisfile.E4vQpVBOQLyfYY8p", "thisfile") = 0
unlink("thisfile.E4vQpVBOQLyfYY8p") = 0
So ssh is correct to check the length.
This is an unfortunate behavior of ssh to take an already constrained
path and arbitrarily make it longer. So many ways it could handle
this better, like using a separate non-length-constrained lock file.
I guess I am going to have to hardcode its current 16+1 character length
of the extension it appends in order to tell if git-annex can constuct
a socket path that is short enough.
--
see shy jo
signature.asc
Description: Digital signature
