Re: Logical file order of records

Neill Harper wrote:

The reason I ask is that I have a very hard bug / difference to
find during some parallel regression testing.

The subtle difference was finally tracked down to the order that
the logical files were returning records.

The records in both physicals are in the same order (i.e. rrn)
but the logicals return the records differently. One library
seems to honour the rrn theory for records with the same partial
key, but the other logical seems to be pretty random.

Unfortunately for me this is a details file and there is special
extra processing performed on the last detail record, hence the
difference in the regression.

Both files were compiled from the same dds using the same
commands params, they don't reuse deleted records

Working from a flawed theory is not a good start. Without the attribute specified to define the duplicate-key-value processing, anything that could be described as /random/ is just as correct as any other valid result. The /copy/ of the file really should be a restored version which was both saved as a network with access paths and then restored without access path rebuild. Any other method to effect a copy leaves too many issues which might impact the access path; e.g. built [and when] versus built [when] & then maintained [since when].

Regards, Chuck