building ghc once or twice?

[Jens Petersen]

I have been packaging ghc for a long time...

In older times I think it was recommended to first do a (quick) build
of a new version of ghc (with the previous version) and then to do a
(perf) rebuild of the new version against itself.
In fact I am still building ghc this way for Fedora: though it seems
like this is overhead nowadays...?
(I think one major reason was to get stable ABI hashes for the core
library packages.)

These days should I just do a single default or perf build of a new
ghc version against a previous stable release, or does it still make
sense to continue to build in two steps like I have been doing?
Any pros or cons?

Jens