Track reproducibility of packages

[avdgrinten]

xbbs should track whether packages could be (byte-wise) reproduced consistently (e.g., by the last full rebuild or the last X full rebuilds).

[avdgrinten]

Here are some more thoughts on this:

Definitions:

• Call a package n-times reproduced if the last (n+1) builds produced the same file hash. Note that is important that we demand (n+1) builds here (i.e., the initial build does not count).
• We say a package is m-times reproduced by full builds if it is n > m times reproduced and among these n builds, there are m full builds. Note that this is not the same as demanding that the last (m+1) full builds produced the same file hash (e.g., if the initial build of the package is not done in a full build). We can do the same definition for incremental builds.

I think what we should display is the number of times a package was reproduced by (i) full builds (second definition) and (ii) all builds (first definition).

[avdgrinten]

Actually, the following might be simpler:

For each package, we build a table with one row per build and (build number, version, hash) as columns. Then, if there are two rows for the current version that have different hashes, that's a reproduction failure.

[ArsenArsen]

That is indeed what I wanted to do. Just collect build artifact hashes, find discrepancies

EDIT: The above statistic can very well be calculated from a history of build hashes