Reply To cazfi
in S3_0. That does not happen with later branches. I don't know what makes the difference
The difference comes from lua5.3 vs lua5.4 headers. The way macros are defined in the latter make it clear to the analyzer that there's no problem.
Patch meant for S3_0 only (not to later branches)
Clang analyzer gives some NULL dereference warnings about luasql code in S3_0. That does not happen with later branches. I don't know what makes the difference - we have newer version of luasql in S3_1 & master, but the changes don't seem relevant.
It's easy enough to silence those warnings by checking against NULL pointer in the very spots the warnings are given about, but if there's real problem (i.e. if NULL pointers can really make it there) that would only mean that the crash would happen a bit later where the very same pointer would be dereferenced again, even though analyzers heuristics don't catch that.