[petsc-dev] reproducibility

Karl Rupp rupp at mcs.anl.gov
Sun Jun 9 17:32:56 CDT 2013


 >         With -citations, or in addition to it, we should have some flag
>     that causes the print out of the unique version information for each
>     package used in a computation. Since computations are often done
>     "between" version releases or with "randomly" patched releases how
>     can we capture that information?
> I think getting version information from other packages is pretty
> worthless (how many times has SuperLU
> had a shadow release). However, for all the packages that we control, we
> can get the SHA1 very simply.
> So a list of SHA1s for all packages would be meaningful.

Yeah, SHA1 is fine. One problem, however, is the reverse-lookup, i.e. 
trying to 'find' the correct packages with given SHA1s in order to e.g. 
reproduce some results. If we provide all external packages out of our 
own download repositories, we can keep track of the SHA1s in use. With 
external download URLs, however, such a trace-back might be fairly hard 
or even impossible. Thus, we might want to consider some date 
information in addition to the SHA1 just to have a hint in case a SHA1 
cannot be matched.

Best regards,

More information about the petsc-dev mailing list