I wrote:
> Would it be possible to dial down the amount of runtime consumed by
> the regression tests for this module?
Hmm ... "perf" says that a full 50% of the runtime of contrib/bloom's
regression test is consumed by GenericXLogFinish, and of that, the vast
majority is consumed by the extremely inefficient one-byte-at-a-time
matching in writeDelta() --- even dwarfing the "exchange two page images"
memcpy's. So there seems to be good reason to expend some
micro-optimization effort in that area. But aside from the writeDelta
problems, what in the world is the reason for the page exchange logic?
I could see copying a working image into place in the shared buffer
arena, but not saving the previous contents.
Even if these costs went to zero, though, the bloom regression tests
would still be consuming a disproportionate amount of time.
regards, tom lane