I should have added in my last reply that if you want to investigate
field splitting, and how it works in various cases, use the expansion
of a simple variable
x=whatever
IFS=something
echo $x
or better:
args() {
printf '#=%d' "$#"
[ "$#" -gt 0 ] && printf ' <%s>' "$@"
printf '\n'
}
and use "args" instead of echo, it makes things clearer (just don't
use < or > in the tests anywhere).
Then, once that is fully understood, you can go on to dealing with
field split $* and $@ but when you're doing that, don't forget this
part of XCU 2.5.2 (describing the '*' special parameter).
When the expansion occurs in a context where field splitting will
be performed, any empty fields may be discarded
That "may" was mentioned before, but it continues:
and each of the non-empty fields shall be further split as described
in Section 2.6.5.
and there note that any empty fields that are not discarded are not
subject to field splitting. The reason for that is simple: POSIX
describes what the shells actually do, not what would make rational
sense, and if empty fields were also to be field split, that "may"
just before would turn into a "shall" (regardless of IFS, unless it
is empty) which isn't what shells actually do. That's because field
splitting requires that any empty resulting word be discarded (regardless
of whether the word changed in any way because of the field splitting).
That's why, and is the only reason why
A=
echo $A $A $A $A
prints just a newline, not 3 spaces, and then a newline. echo there is
(and is required to be) given no args at all. Replace that with
set -- '' '' '' ''
echo $*
and all bets are off, nothing specific is required (except the
trailing newline). Any number of spaces, 0..3, is permitted,
and what you get depends on what is in IFS even though you'd
expect it shouldn't, as there is nothing anywhere to match an
IFS character, or not unless you know how an implementation
actually implements that, which isn't something any user should
be expected to know, and certainly never to depend upon.
You have to combine all of that together, and also copy the implementation
technique bash uses for unspecified cases, if you really want to generate
the same results that bash does.
kre
ps: to anyone out there still reading this, when writing sh code, don't
lapse into writing C instead, in sh, when using quotes, use '' (or \)
whenever the quoted text is to be treated literally, and "" only when
you are expecting expansions in the included string - doing it the C
way ('c' for a single character, and "str" when it is a string) makes
no sense in sh at all, in sh everything is a string, quoting is only ever
needed to avoid some character(s) from having some other meaning then
simply representing themselves (either in sh parsing, or in one of the
expansions that happen). Use $'' when you want something that is more
or less identical to a C string (including the transformations that C
allows inside one, like \n becoming a newline, etc).
