Re: A small test case for the AC_CONFIG_xxx problem

[Paolo Bonzini]

| echo > test1
| testsrc=test1
| testdest=test2
| AC_CONFIG_LINKS($testdest:test1 test3:$testsrc, [],
|   [testsrc=$testsrc
|    testdest=$testdest])

> I.e., test3 is OK, but not test1.

There are two parts in it.  One bug is this controversial use
of AC_CONFIG_LINKS, which I only introduced for symmetry
and to check whether the bug only involved one purpose or
both.

The other bug is that `./config.status test3' won't work,
while `CONFIG_LINKS=test3 ./config.status' or simply
`./config.status' will.  This is indeed a bug and is because
the processing of command line arguments in config.status
must be after the evaluation of the INIT-CMDS; you can now:

- defer the processing of all command line arguments, including
for example --help, like in the patch I sent to autoconf-patches,

- or process options and defer the processing of the rest of the
command line, like in the patch that I sent you privately
(by mistake, I wanted to address autoconf-patches actually).

Note that allowing both ways to use AC_CONFIG_LINKS is a
breeze once you move the INIT-CMDS, since it is only a matter
of double-quoting instead of single-quoting the choice in
the command line parsing case statement.

I hope this clarification will help fixing the bug.

Paolo