> It is a slightly disappointing that the loops for > the buffers are duplicated, particularly for the relation vs the all > case.
Yes, and we got another two loops in pg_buffercache_evict functions, and more loops in Drop/Flush relation/database buffers functions. Maybe we can abstract them into a generic loop function and it takes a buffer handler function pointer to avoid duplication? > One thing that would make more sense to me is to > split the patch in two: one for the changes in bufmgr.c/h and one for > pg_buffercache. There can be an argument that these new APIs could be > useful for out-of-core code, as well, to let developers play with the > state of the buffers. I mean, why not, the thread is also about API > extensibility, as well, to enforce dirty states. :) Agreed. It might be three, one for generic loop function, one for API and one for pg_buffercache. If we put new API out-of-core, the latter two can be merged. Best regards, Yuhang Qiu
