Обсуждение: debug_sortsupport GUC?
The text abbreviation code has a compile-time option to emit DEBUGn elogs. I made no attempt to add these to the numeric abbreviation code because I find such things completely unhelpful; when you need to investigate such things other than in initial development, it's unlikely that you will be in a position to recompile. Also, as I showed at some length earlier, even recompiling with an apparently irrelevant change can have enough of an effect on performance to make investigation more complex. So if these debugging elogs are to be kept at all, I propose that rather than being compile-time options they should be controlled by a debug_sortsupport GUC. Opinions? -- Andrew (irc:RhodiumToad)
On Sat, Mar 21, 2015 at 8:28 PM, Andrew Gierth <andrew@tao11.riddles.org.uk> wrote: > So if these debugging elogs are to be kept at all, I propose that rather > than being compile-time options they should be controlled by a > debug_sortsupport GUC. Opinions? This seems like a reasonable idea. Why wouldn't it just be under the existing trace_sort GUC, though? That's been enabled by default since 8.1. It's already defined in pg_config_manual.h. -- Peter Geoghegan
>>>>> "Peter" == Peter Geoghegan <pg@heroku.com> writes: >> So if these debugging elogs are to be kept at all, I propose that>> rather than being compile-time options they shouldbe controlled by>> a debug_sortsupport GUC. Opinions? Peter> This seems like a reasonable idea. Why wouldn't it just be underPeter> the existing trace_sort GUC, though? That'sbeen enabled byPeter> default since 8.1. It's already defined in pg_config_manual.h. Ungh... yes, it's defined by default, but it clearly still requires keeping the #ifdefs in there in order to still build if someone manually undefines it. Was hoping to avoid the #ifdefs entirely - perhaps the existing #ifdefs should just be removed? If it's been enabled since 8.1 it seems unlikely to be causing anyone any issues. -- Andrew (irc:RhodiumToad)
On Sun, Mar 22, 2015 at 6:59 PM, Andrew Gierth <andrew@tao11.riddles.org.uk> wrote: > Ungh... yes, it's defined by default, but it clearly still requires > keeping the #ifdefs in there in order to still build if someone manually > undefines it. Was hoping to avoid the #ifdefs entirely - perhaps the > existing #ifdefs should just be removed? If it's been enabled since 8.1 > it seems unlikely to be causing anyone any issues. I'd vote for removing the #ifdefs, and using trace_sort. -- Peter Geoghegan