FreeTDS Development Group

[christos AT zoulas.com (Christos Zoulas)]

On Aug 24,  3:14pm, entropy AT freetds.org (entropy AT freetds.org) wrote:
-- Subject: Re: [freetds] tsql bug fixes that got lost between 0.64 and 0.65

| Christos Zoulas wrote:
| 
| > Have you tried strace'ing/truss'ing tsql when it reads an input file?
| > Using readline(3), it issues a read(2) per character, not counting the 
large
| > number of failed ioctl(2)'s in the beginning.
| 
| Can't say that I have, but I'm not too surprised to hear that it could 
| be more efficient.
| 
| > We have the code in tsql to
| > do the readline, why not just use it when we are not connected to the tty?
| > This way, in the future, when readline(3) decides to do something else 
bogus
| > with the input stream we are not going to be affected.
| 
| I don't think readline(3) just decided to do this.  readline(3) has been 
| doing tab completion for as long as I can remember.  I'm not aware of a 
| history of surprising behavior changes in readline(3).  So, as for any 
| future unexpected behavior changes, I'm inclined to worry about it when 
| it happens instead of worrying about it now.

Readline has been doing tab completion interactively for a long
time. I don't know when it started doing this non-interactively.
Anyway, I think that it is incredibly stupid for readline to work
the way it does. What happens if I enter another editing character
in a script, such as escape or control-p, or even meta-x? I don't
know the answer, but my guess is that it will do the editing action.
Also what happens when you enter text in a non ascii-c locale, and
that character you enter happens to be bound to a readline function?
Readline should recognize that it is not connected to a tty and
turn character at a time input off and disable all the editing. It
currently does not do that, so it is unsafe for processing input
from files.

| I personally don't like the idea of switching to a different readline(3) 
| implementation based on whether the terminal is a tty or not.  It will 
| cause surprising and stupid behavior, like my SQL input mysteriously 
| failing if it contains a line longer than 1024 characters.

That is a problem with the current implementation of readline(3) in tsql,
and can be easily fixed. On systems that have fgetln() use that instead,
on systems that don't have it use:

    http://cvsweb.netbsd.org/bsdweb.cgi/src/tools/compat/fgetln.c

christos