I am not entirely sure yet if the below is a problem with my
understanding, with tramp, or with my expectations of tramp, so
I'm going to go through every step.
Firstly I am only talking about the default behaviour (no
customisation).
Secondly I think the default behaviour is sensible for what I'll
call "tramp shells"; i.e. shells that the user does not have the
opportunity to interact with, but rather are used by tramp to do
tests and start up remote processes. By default the history of
these is "diverted" to .tramp_history and IMO that both works fine
and is sensible.
However, when the remote process is a shell that is a subprocess
of shell-mode/comint, the intention is for the user to now be
issuing shell commands, and I do not believe that history should
end up in the tramp history. And currently it seems to, due to
tramp setting HISTFILE in a parent environment and that being
inherited all the way down to a shell that is not in fact used by
tramp.
Furthermore I think it makes sense for shell-mode to read the
non-tramp history on startup, and also for this non-tramp history
to be the same as the shell history used by a shell started
outside Emacs, so that everything works together nicely by
default.
In the case of the explicit shell being bash, specifically, I am
able to get the behaviour that I think makes sense with the
following, but it seems like something easier should be possible.
`shell-history-file-name' set to "~/.bash_history"
init_bash.sh containing:
if [ "${INSIDE_EMACS/*tramp*/tramp}" == "tramp" ] ; then
export HISTFILE=~/.bash_history history -cr $HISTFILE
fi
This causes shell-mode to load its history from the "true
interactive" (non-tramp) history, re-initialises the
remote-shell-processes history from the same so that builtin
history commands and the like work, and also sets the shell to the
non-tramp file so that it saves this "human" history separately on
exit.
But it feels like there should be an easier way of doing
this. Something like unsetting HISTFILE immediately before
starting the remote shell process (but after everything else that
tramp has done, that should still go to .tramp_history)
Thoughts/corrections?
Thanks and regards,
- Joel
