Bash – How to debug and fix slow autocomplete in bash

autocompletebash

After a recent update (Ubuntu 12.04 LTS), TAB complete on the command line is slow. After entering a partial command (e.g evi [TAB]) or partial filename (e.g. evince somedocu[TAB]) the shell, sometimes though not always, hangs for several seconds.

Personally, I'd prefer a less powerfull autocomplete to a slow one. Is there a simple fix?

Edit: Additional information related to comments:

PATH is pretty standard. ~/bin has some bash scripts

$ echo $PATH
/home/USERNAME/bin:/usr/local/bin:/usr/bin:/bin:/usr/games

The number of files in the working directory is less than 100.
The autocomplete feature was especially slow after unusual disk activity (system upgrade). It is, thus, possible, that rereading /usr/bin and other directories caused the lag.

Best Answer

I don't know about fixing — there are all kinds of things that could go cause delays. But I can offer a few tips to investigate.

Just as a guess, maybe there's a directory somewhere in a search path ($PATH, or some place where bash looks for completion data) that's on a filesystem which is slow to respond. Usually it's remote filesystems that are slow, but it could also be a failing hard disk, a hung FUSE driver, etc.

The first step to investigate is to run set -x to get a trace of the commands that the shell executes to generate the completions. Watch where it pauses.

If that doesn't give enough information, bring in the big guns. Note the shell's process ID (echo $$). In another terminal, run strace -f -s9999 -p$$ (or the equivalent of strace if running on another unix flavor). Strace lists the system calls performed by the process. See if it seems to be accessing files that it shouldn't, or if access to some files is slow. Adding the option -T to the strace command line makes it show the time spent in each system call.

Related Solutions

Bash – How to Enable Autocomplete in SSH Session

In short: source /etc/bash_completion should do the trick (run it in the shell within the SSH session).

Long story: in order for bash completion to work, you have to tell bash how to complete each command's arguments. This requires a long sequence of invocations of the bash built-in command complete; therefore, they are usually collected in a separate script (or several ones in /etc/bash.complete.d/*) that loads them all.

Being a regular shell script, you can always load the bash_completion in any shell startup script (~/.bash_profile, ~/.bash_login, ~/.bashrc)

Bash – Custom bash autocomplete for git breaks other git autocomplete features

You could use __git_complete (defined in git-autocompletion.bash) to install your own function, and have your function fall back to the original function. Possibly like this:

function _ptid_git_complete_()
{
  local line="${COMP_LINE}"                   # the entire line that is being completed

  # check that the commit option was passed to git 
  if [[ "$line" == "git commit " ]]; then 
    # get the PivotalTracker Id from the branch name
    ptid=`git branch | grep -e "^\*" | sed 's/^\* //g' | sed 's/\-/ /g' | awk '{ print $(NF) }'`
    nodigits=$(echo $ptid | sed 's/[[:digit:]]//g')

    if [ ! -z $nodigits ]; then
      : # do nothing
    else
      COMPREPLY=("-m \"[#$ptid]")
    fi
  else
    __git_main
  fi
}

__git_complete git _ptid_git_complete_

Best Answer

Related Solutions

Bash – How to Enable Autocomplete in SSH Session

Bash – Custom bash autocomplete for git breaks other git autocomplete features

Related Question