k8ika0s commented on PR #48212: URL: https://github.com/apache/arrow/pull/48212#issuecomment-3572953306
After stepping back and watching how this series of patches evolved, it’s clear that my original intent for this work is no longer aligned with the approach being taken here. The goal of my earlier implementation was to address endianness behavior at an architectural boundary: establish explicit canonicalization rules, normalize inputs at well-defined choke points, and remove the growing list of one-off swaps, conditionals, and ad-hoc fixes that have accumulated over the years. That approach aimed to eliminate ambiguity rather than perpetuate it. The current trajectory is moving in the opposite direction. Instead of a consistent invariant, the changes now being proposed fragment the solution across multiple PRs, each addressing isolated symptoms without resolving the underlying design assumptions. This scattershot strategy may produce incremental green tests in the short term, but it leaves the long-term behavior brittle, harder to reason about, and prone to regressions whenever a new code path encounters one of the many inconsistent byte-handling patterns that now exist in the system. For the sake of clarity in the project’s history: the architectural analysis, test failures, and rationale behind the canonicalization strategy I proposed remain documented in my earlier PR and its review discussion. That material is still relevant to anyone trying to fully understand the root causes behind s390x breakage, even if that work is no longer being pursued here. I’m adding this note so that future contributors have a clear pointer to the design context that motivated these changes, rather than having it disappear into an abandoned diff. I’m closing this from my end. I won’t be continuing to restructure or split the work further, as doing so would only reinforce a direction I believe is structurally unsound. Anyone who wants to revisit the architectural approach, or understand why piecemeal patches will continue to create inconsistent behavior, can reference the prior discussion for the full set of invariants, test results, and design implications. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
