At Fri, 10 Feb 2023 16:43:15 +0900, Michael Paquier <mich...@paquier.xyz> wrote in > On Wed, Feb 08, 2023 at 09:42:13PM -0500, Tom Lane wrote: > > Hm. On the one hand, if it is in fact not in postgresql.conf.sample, > > then that flag should be set for sure. OTOH I see that that flag > > isn't purely documentation: help_config.c thinks it should hide > > GUCs that are marked that way. Do we really want that behavior? > > Not sure. I can see an argument that you might want --describe-config > > to tell you that, but there are a lot of other GUC_NOT_IN_SAMPLE > > GUCs that maybe do indeed deserve to be left out. > > I am not sure to follow. help_config() won't show something that's > either marked NO_SHOW_ALL, NOT_IN_SAMPLE or DISALLOW_IN_FILE, hence > config_file does not show up already in what postgres > --describe-config prints, because it has DISALLOW_IN_FILE, so adding > NOT_IN_SAMPLE changes nothing.
I think currently the output by --describe-config can be used only for consulting while editing a (possiblly broken) config file. Thus I think it's no use showing GIC_DISALLOW_IN_FILE items there unless we use help_config() for an on-session use. On the other hand, don't we need to remove the condition GUC_NOT_IN_SAMPLE from displayStruct? I think that help_config() should show a value if it is marked as !GUC_DISALLOW_IN_FILE even if it is GUC_NOT_IN_SAMPLE. I'm not sure whether there's any variable that are marked that way, though. regards. -- Kyotaro Horiguchi NTT Open Source Software Center