Re: patch for parallel pg_dump

On Wed, Mar 14, 2012 at 2:02 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> I think we should get rid of die_horribly(), and instead have arrange
> to always clean up AH via an on_exit_nicely hook.

Good. The only exit handler I've seen so far is
pgdump_cleanup_at_exit. If there's no other one, is it okay to remove
all of this stacking functionality (see on_exit_nicely_index /
MAX_ON_EXIT_NICELY) from dumputils.c and just define two global
variables, one for the function and one for the arg that this function
would operate on (or a struct of both)?

We'd then have the current function and AHX (or only &AH->connection
from it) in the non-parallel case and as soon as we enter the parallel
dump, we can exchange it for another function operating on
ParallelState*. This avoids having to deal with thread-local storage
on Windows, because ParallelState* is just large enough to hold all
the required data and a specific thread can easily find its own slot
with its threadId.

>>> Sure, but since all the function does is write to it or access it,
>>> what good does that do me?
>>
>> It encapsulates the variable so that it can only be used for one
>> specific use case.
>
> Seems pointless to me.

Not so much to me if the alternative is to make ParallelState* a
global variable, but anyway, with the concept proposed above,
ParallelState* would be the arg that the parallel exit handler would
operate on, so it would indeed be global but hidden behind a different
name and a void* pointer.

(I will address all the other points you brought up in my next patch)