Re: texi2dvi: Support bibunits

[Akim Demaille]

>>> "Eli" == Eli Zaretskii <address@hidden> writes:

 > Although you posted an improved ChangeLog entry, and although Karl
 > accepted the changes, I'd still like to ask for some elaborate
 > explanations for this.  In particular, why the change from

 >> -  for this_file in "$1".?o? "$1".aux "$1".?? "$1".idx; do

 > to

 >> +  # Use `ls' because some patterns are likely not to match.
 >> +  # - .?o?
 >> +  #   .toc, .log, LaTeX tables and lists, FiXme's .lox, maybe more.
 >> +  # - bu[0-9]*.aux
 >> +  #   Sub bibliography when using the LaTeX bibunits package.
 >> +  for this_file in `(ls "$1".?o? "$1".aux bu[0-9]*.aux "$1".?? "$1".idx) 
 >> 2>&6`
 >> +  do

 > What does it mean ``some patterns are likely not to match''?

It means that I added bu[0-9]*.aux, which might not match at all.  In
which case, depending upon the shell you might have an error, get the
pattern back as result, or have the pattern expand to nothing.  ls
maps all these behavior to the latter.

 > Anyway, I'm worried that using `ls' could fail, if, for example, `ls'
 > is aliased to "ls -F".

Well, configure never bothered with this, so I guess it is pretty
safe.  And anyway, I fail to see what -F would change here.

 > As for replacing >/dev/null with >&6, is that portable enough to
 > various flavors of Bourne shell?

It was already use elsewhere in the script.  Maybe you are missing the
part where the fd 6 is defined.