Re: BUG #19545: Integer truncation of `GinTuple.keylen` causes out-of-bounds read in parallel GIN index build

On 18.09.26 04:07, Tom Lane wrote:
> Peter Eisentraut <peter(at)eisentraut(dot)org> writes:
>> I have committed these, thanks.
>
> The ABI-compliance-checking buildfarm animals have all been unhappy
> since this went in:
>
> 'struct GinTuple' changed:
> type size changed from 16 to 24 (in bytes)
> 1 data member deletion:
> 'uint16 keylen', at offset 6 (in bytes)
> 1 data member insertion:
> 'union {Size keylen; double force_align_d; int64 force_align_i64;} u', at offset 8 (in bytes)
> there are data member changes:
> 'int16 typlen' offset changed from 8 to 16 (in bytes) (by +8 bytes)
> 'bool typbyval' offset changed from 10 to 18 (in bytes) (by +8 bytes)
> 'signed char category' offset changed from 11 to 19 (in bytes) (by +8 bytes)
> 'int nitems' offset changed from 12 to 20 (in bytes) (by +8 bytes)
> 'char data[]' offset changed from 16 to 24 (in bytes) (by +8 bytes)
>
> Is it really okay to change this struct in v18?
> If so, the .abi-compliance-history files need to be updated.

Yes, I think this is okay. The type is effectively for internal use
only and is only used transiently. I will add the
.abi-compliance-history entry.