Ron Johnson <ronljohnsonjr@gmail.com> writes:
> On Mon, Dec 18, 2023 at 12:02 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Looking at the v14 pg_ctl code, I see it has an extra level of
>> looping around getopt_long to try to cope with implementations
>> that don't permute the arguments. So in theory it should work as
>> documented everywhere. But I could see how things might get
>> confused, if the subroutine misidentifies what is a switch.
> There must be some PEBKAC involved that pg_ctl error handling didn't
> explain well, because _this_ unreformatted command works.
After studying the code some more, I think I can reproduce it:
$ pg_ctl restart foo -o "whatever"
pg_ctl: too many command-line arguments (first is "restart")
Try "pg_ctl --help" for more information.
It should be complaining about "foo", but the message mentions
"restart" instead. I think you must have had something in your
non-working command that did not get interpreted as being a
switch or a switch argument, leading to the misleading report.
That's with v14. HEAD gets this right:
$ pg_ctl restart foo -o "whatever"
pg_ctl: too many command-line arguments (first is "foo")
Try "pg_ctl --help" for more information.
The change in behavior is evidently due to commit 03d1080d8, which
got rid of the extra level of looping in pg_ctl's use of getopt_long.
I surmise that after we loop around and re-invoke getopt_long,
glibc's getopt_long resets "optind" to the first non-option argument,
rather than honoring the increment from the bottom of the loop.
This seems like a bug to me, but I'm not sure if I want to
propose back-patching 03d1080d8 (and 411b72034) to deal with
it. That code hasn't made it into a stable branch yet, so
a back-patch would be a little bit risky.
regards, tom lane