I've been reading archives of Wandering Thoughts, the blog of a UToronto sysadmin. Lots of interesting ideas. Reading PersistentVsDisposableUsage, in which he claims that most people either open a new shell for each task or use the same set of long running shells for multiple tasks, I realized I fall pretty squarely into the second category. I keep shells open that fill the desktop space, and use whichever one is free. But where I should be, really, is "shell per task", where when doing one class of thing I use a specific shell instance. This would keep history straight.

I also started thinking about what Sameer does with history: all commands typed go both into a full history file somewhere and a .local_history file in the current directory. The local_history idea I don't like, as it clutters things up a lot, but a full history that's more robust than the shell history would be nice. So I added the lines below to my prompt function in bash:

echo "$(date +%Y-%m-%d--%H-%M-%S) $(hostname) $PWD $(history | tail -n 1)" >> ~/.full_history

I think this should work, but it's possible that with appending from multiple places at once it will get clobbered at some point.

I also thought about what commands do I run:

    $ cat combined-history | awk '{print $2}' | susrn | head
    619 ls
    516 find
    413 cat
    304 python
    275 cd
    216 svn
    215 rm
    199 more
    130 for
    122 svc
    

Alternately, taking into account all the compound commands I run by looking at the commands following pipe and semicolon in addition to line starters:

    $ cat combined-history | sed s/'[;|]'/'\nNNN '/g | awk '{print $2}'| susrn | head -n 12
    881 grep
    620 ls
    544 do
    536 done
    535 sed
    524 find
    501 python
    465 cat
    412 while
    315 head
    283 cd
    291 more
    

(combined-history is the result of running history in each shell)

Notes:

• I invoke most python scripts with python instead of a shebang. I use a lot of python.
• I leave emacs running most of the time in its own windows. So it doesn't show up.
• The command svc is svn commit
• The do and done entries are higher than while because I sometimes use for. The do entry is higher than the done one because I leave off final done more than I'd like. I extended the second one to 12, in case you don't want to look at do and done.
• While I realize there are performance reasons not to, I use cat $fname | cmds all the time. It's much more consistent because then everything follows the same pattern. Bash could consider optimizing "cat single-arg | cmds" to "< single-arg cmds".