WT-11634 Provide optimized versions of set_key/value for MongoDB index use cases

[raviprakashgiri29]

The changes include the following:.

• Added new API void __wt_cursor_set_raw_key_value(WT_CURSOR *cursor, WT_ITEM *key, WT_ITEM *value) and void_wt_cursor_set_raw_key_value_notsup(WT_CURSOR *cursor, WT_ITEM *key, WT_ITEM *value)
• Added a cppsuite unittest to test the API's
• Added a csuite test wt11634_get_set_key_value/main.c to test the API's

[agorrod]

Did you consider refactoring and using set_raw_key and set_raw_value internally? I know we are chasing performance here, but the duplication of complex and subtle code worries me.

There also doesn't seem to be enough checking to make sure the API is used as expected. I'd expect that to be present in at least diagnostic builds, and preferably everywhere: This is a primary mechanism for applications to give context to WiredTiger, which it will operate on. Ensuring the data is in an expected state/format on entry is necessary to limit how much further checking/safety we need inside the library.

[jeremythorp]

Hi @agorrod , as I understand it, part of the challenge here is that __wt_cursor_set_raw_key and __wt_cursor_set_raw_value call set_key and set_value respectively, and those functions take va args parameters. Removing the va args usage (and perf cost) is a main benefit from this proposed change.
Obviously, if there are opportunities to refactor the code and/or take a simpler approach then we should do so.

[jeremythorp]

Hi Ravi, do we need this new test file? I'd consider performing similar tests using the Catch2 framework instead, and adding them to those you have added in test_cursor_set_raw_key_value.cpp.

[evergreen-ci-prod]

Test coverage is too low, this change probably shouldn't be merged as-is.

Metric (for added/changed code)	Coverage
Line coverage	68% (27/40)
Branch coverage	33% (22/66)