On 2015-12-11 19:26:38 -0500, Tom Lane wrote:
> I believe it's soluble, but it's going to take something more like
> loading up all the data at once and then doing lookups as we write
> out the .bki entries for each catalog. Fortunately, the volume of
> bootstrap data is small enough that that won't be a problem on any
> machine capable of running modern Postgres ...
I think that's exactly the right approach. Just building a few perl
hashes worked well enough, in my prototype of that.
If additionally a few more plain oid fields are converted into reg*
types, the source data fields are easier to understand and the catalogs
get much nicer to query...