daveg <daveg@sonic.net> writes:
> On Sun, Jul 20, 2008 at 02:50:50PM -0400, Tom Lane wrote:
>> In most cases our policy has been that pg_dumpall should accept and pass
>> through any pg_dump option for which it's sensible to do so. I did not
>> make that happen but it seems it'd be a reasonable follow-on patch.
> I'll remember that next time.
Er .. actually that was a direct request for you to do it.
> Finally, you changed the option value and case from 1 to case 2. getopt_long
> only returns the value argument when there is no flag to set, so 1 was unique
> and could have been read as "the first no-short option with an argument".
Yeah. The code *worked* as you submitted it, but what was bothering me
was that the "val = 1" table entries worked in two completely different
ways for the different argument types. I first thought that you'd
broken the existing long argument options --- you hadn't, but I had to
go re-read the getopt_long source to convince myself of that. So it
seemed like using a different "val" value might help clarify the
difference in behavior for future readers of the code. In particular
the next guy who wants to add a long option with parameter value would
certainly not be able to use val = 1; but I thought the code as you
had it wouldn't give him any clue what to do. If you've got a better
idea about how to deal with that, feel free...
regards, tom lane
