Re: pg_dump misses comments on NOT NULL constraints

On Wed, Jun 18, 2025 at 11:05 PM Álvaro Herrera <alvherre(at)kurilemu(dot)de> wrote:
>
> I agree that this is roughly the right approach, but I think you're
> doing it harder than it needs to be -- it might be easier to add a JOIN
> to pg_description to the big query in getTableAttrs(), and add a pointer
> to the returned string in tbinfo->notnull_comments[j] (for versions
> earlier than 18, don't add the join and have it return constant NULL).
> Then in dumpTableSchema, in the place where you added the new query,
> just scan that array and print COMMENT ON commands for each valid
> constraint where that's not a null pointer.
>

Previously I was worried about print_notnull, shouldPrintColumn.
if there is a not-null constraint that is not dumped separately, it has comments
then we should dump these comments, then no need to worry about print_notnull.

using notnull_comments saves us one more query.
however, in determineNotNullFlags we have:

char *default_name;
/* XXX should match ChooseConstraintName better */
default_name = psprintf("%s_%s_not_null", tbinfo->dobj.name,
tbinfo->attnames[j]);
if (strcmp(default_name,
PQgetvalue(res, r, i_notnull_name)) == 0)
tbinfo->notnull_constrs[j] = "";

then we can not blindly use tbinfo->notnull_constrs as the not-null
constraint name.
if tbinfo->notnull_constrs is an empty string, we need to use the above
"%s_%s_not_null" trick to get the default no-null constraint name.