Re: Possible slight speedup to non-interactive /bin/sh startup

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Mon, 11 Feb 2019 19:56:41 +0100
    From:        =?UTF-8?B?SmFyb23DrXIgRG9sZcSNZWs=?= <jaromir.dolecek%gmail.com@localhost>
    Message-ID:  <CAMnsW54tmHhcTbfxwv64e=c=5h-E3iUobrO4kR2p5nkRxMbaDA%mail.gmail.com@localhost>

  | IMO 2% speedup is not worth this.

Entirely reasonable - I'm not sure it is worth it either.

  | What exactly does take libedit do to take it so long? Does it trigger
  | loading locales?

No, libedit does nothing, it isn't called at all in a non-interactive
shell (it just sits there).   When I was first testing the new code,
I did the infrastructure first (the stuff to call via the functiion
pointers, etc) to make sure that no remaining undefined refs to libedit
exist, before I added the dlopen() and the dlsym() loop ... which of course
made the shell dump core as soon as it attempted to execute one of those
un-init'd pointers.   An interactive sh couldn't start at all.   But
non-interactive ones worked just fine.

The extra time is all in ld.so simply finding, and doing whatever it
does with the library so that the functions it defines can be called.

Nothing that could be done to libedit itself can possibly change
anything that is being measured here.

kre