I wonder whether it was ever consciously decided what the dependency
relationship between libpgport and libpgcommon would be. When I added
asprintf(), I had intuitively figured that libpgport would be the lower
layer, and so psprintf() in libpgcommon depends on vasprintf() in
libpgport. I still think that is sound. But working through the
buildfarm issues now it turns out that wait_result_to_str() in libpgport
depends on pstrdup() in libpgcommon. That doesn't seem ideal. I think
in this case we could move wait_error.c to libpgcommon. But I would
like to know what the consensus on the overall setup is.