Re: modular database backends

[Milan Zamazal <pdm@zamazal.org>]

>>>>> "PN" == Peter Novodvorsky <nidd@altlinux.ru> writes:

    PN> Milan Zamazal wrote:
    >> - Maybe it's not *necessary* to force backends to implement full
    >> query handling.  We probably can agree on that writing new
    >> backends should be as simple as possible.  I can imagine that
    >> your `query_pr' function could be only optional and there could
    >> be available simpler versions of query functions that can get a
    >> list of all problem IDs, a particular PR and maybe also
    >> optionally some index (similar to the current one).

    PN> We can implement simplier functions in library that will open
    PN> this module. And these functions will use query_pr as more low
    PN> level function. Any objections?

I'm not sure we understand each other.  To clarify the things: For
instance, we could make a backend library that defines three functions
query_pr, get_list_of_pr_ids, get_pr.  query_pr is implemented in the
library using the other two functions and those are implemented using
query_pr.  Every backend must provide its own implementation of either
query_pr or get_list_of_pr_ids and get_pr, the other functions may (but
needn't) be implemented using the library.

So a backend powered by a sophisticated query engine (e.g. SQL) can
implement the complex query_pr function itself and needn't bother to
implement the simpler functions get_*.  On the other hand, a file system
backend would implement the get_* functions and use the library function
for query_pr.

OK?

Milan Zamazal

-- 
And why?