Takahiro Itagaki <itagaki.takahiro@oss.ntt.co.jp> writes:
> I'd like to apply the patch to HEAD and previous releases because
> the issue seems to be a bug in the core. Any comments or objections?
The proposed patch seems quite ugly to me; not only the messy coding,
but the fact that it might return either the segment containing the
XLOG_BACKUP_END record or the next one.
I think an appropriate fix might just be s/XLByteToSeg/XLByteToPrevSeg/,
so that the result is always the segment containing the XLOG_BACKUP_END
record even when the record ends exactly at a segment boundary.
regards, tom lane