Inline index_offset_buffer_t methods

[senderista]

There were a number of trivial index_offset_buffer_t methods that showed up in profiles of the server, which turned out to be defined in a .cpp file, preventing inlining. Here is the most expensive one (about twice the overhead of gaia::db::server_t::sort_log):

Total time   Self time       Calls  Function
  ==========  ==========  ==========  ====================
    1.453  s   41.722 us           4  gaia::db::server_t::session_handler
    ...
    46.535 ms   46.535 ms     1000601  gaia::db::index::index_offset_buffer_t::size
    22.155 ms    3.122 ms          18  gaia::db::server_t::sort_log
    ...

Measurements using test_insert_perf.simple_table_insert do not show a clear improvement, but profiling confirms that server overhead is reduced.

[LaurentiuCristofor]

Can you put these in the order in which they are listed in the header, or change that order to match this one? The latter is probably even better, because methods like insert should precede simple getters like the other methods.

[senderista]

done

[LaurentiuCristofor]

Just a minor request to list the methods in the same order during definition as during declaration.

[simone-gaia]

From my understanding, the inline keyword has almost no effect on the compiler's decision on whether to inline or not. The semantic of the keyword has drastically changed from "you should inline this method" to "allow multiple definitions of the function".

[simone-gaia]

Indeed in the int_type_t change the inline keyword hasn;t been specified.

[senderista]

You have to specify inline somewhere or you'll violate the ODR.

[simone-gaia]

LGTM