Don't let all the bmo processes hang

http://unix.stackexchange.com/a/18443/9365

Another proof that 640KB won't be enough for everybody.

This should enable reading upper parts of window even while the output
is still being produced (tail -f).

IOW, the auto-scroll is disabled after scrolling back manually using
S-PageUp or S-PageDown, and re-enabled on another keypress such as
typing when a prompt is expected at the bottom.

Hey, those numbers are *rough* estimates...

Turns out low_threshold is in minutes, not in percent, by default.
Needless to say, time estimates are grossly unreliable here and 10
minutes left is too late anyway.

... which means "delete last word" in Bashese/Emacese so I often press
it several times in succession,

Although I should not tolerate own typos, I'm also really getting tired
of accidentally tearing down random windows.  Sorry, this is just too
destructive.

Code 57 normally means N, so this layout started (rightfully, I admit)
generating conflict errors.  Assigning key is probably more reliable,
though, so we'll keeo that one.

azzgoat is not accessible this way anymore.  The nothing that was there
is now at http://netvor.vornet.cz/.

Seems to do the same as kurkale6ka/vim-pairs but better.

This is a pretty ugly hack to work around the problem that any attempts
to influence highlight colors in Vim using the `highlight` command were
ignored by Urxvt (not by Xfce4 terminal and few others I tried).

Vim under urxvt does not seem to react to highlight settings, keeping
defaults that under my favorite scheme make Search unreadable and
MatchParen confusing.

The dirty hack is to re-set the colors it happens to use: Yellow and
Cyan.  Ugly, bad, but will work for now.

*Remote* branch name must be listed along with 'PUSH' keyword.

This requires predictable name of the remote, but as this whole slurping
thing is only meant for very dirty, very local scenarios, that should
not be problem.  (Other, more open scenarios are still possible, just
harder, as they probably should be.)

Using grep directly could result in "binary file matches" without any
useful info.

This would happen e.g. after renaming some files with badly-encoded
filenames.  Bash history builtin seems to address such cases better.

(Return cursor after pasting; not sure how well it works for lines but
it's ok for chars.)

In that case (i.e. , the `${full#git_root}` would not work and print the full
basically full absolute path to git root.

Use "meta templates" that ate actually Bash scripts.

 *  Add --force-* options to skipt .git-slurp-ok file checks.

 *  Allow passing path to slurp.

    You can slurp only a particular sub-tree.

 *  Improve commit message.

    Host has been dropped, and slurped paths are logged (as relative
    to worktree root).

Makes more sense to have darker cursor when "inside" in insert mode.

Apparently with some distros, raching 0 will turn off the display
completely.  Step of 49 makes it easy to avoid that while still keeping
the advantage of huge step.

 *  Die by dedicated function.

 *  Push only if allowed (a line containing single word PUSH found in
    .git-slurp-ok file)

 *  Allow working when deeper in git repo (but not if outside).

 *  Make permisson code readable (self-doc).

 *  Enclose in main() to allow for better hygiene.

Now host/user overrides don't have to be created.