On Tue, Oct 08, 2024 at 06:57:10PM +1100, Peter Smith wrote: > As for what is left, I had made just the minimal patch to fix the > known "IntervalStyle" problem, but that assumed prior knowledge of > what were the only problem names in the first place. Your way is > better -- to always use the record name where possible.
Yes, useful for custom GUCs, even if I doubt that many of them mix character casing these days. > One thing I thought remains not so good is how > set_config_with_handle(name...) function does > record=find_option(name,...) but then all subsequent names are > supposed to be coming from the record->name. It feels confusing to > have that parameter 'name' still in scope after the find(), e.g., when > you should not be using that anymore. So, I tried to separate this > logic into 2 functions -- set_config_with_handle() and > set_config_with_handle_guts(). There are no name overlaps now, but I > wonder if the fix was overkill. See v14. I was also looking at reducing the scope of the "name" variable in this path, but that makes IMO the code a bit more confusing with the existing wrapper set_config_option() and set_config_option_ext(), while the handle is only used within fmgrs. Applied the previous patch after looking at it again. -- Michael
signature.asc
Description: PGP signature
