We encountered a bug in the FLCS deprecation work where I forgot to initialize my WT_CONFIG_ITEM prior to passing it to __wt_config_getones. This resulted in a surprisingly rare segfault which was harder to debug than expected. Part of what enabled the segfault was that the calling code wrapped the config call in WT_RET_NOTFOUND_OKAY.

Using WT_RET_NOTFOUND_OKAY in combination with an uninitialized WT_CONFIG_ITEM is likely required to see any negative consequences, or undefined behaviour as a result.

But a brief glance around the codebase suggests numerous locations where an uninitialized WT_CONFIG_ITEM is passed into the config API, and at least once occurrence where it is done in combination with WT_RET_NOTFOUND_OKAY (__wt_session_get_btree_ckpt).

Scope:

• Review WT_CONFIG_ITEM usages.
• Add necesssary initialization.
• Consider ways to prevent this in a future? Perhaps a macro to use WT_CONFIG_ITEM. Similar to WT_DECL_ITEM.