Hi, On 11/24/2017 01:14 AM, Jeremy Bicha wrote: > golang-go was briefly built on s390x a year but was disabled because > upstream go only supports relatively newer s390x and because it was > shortly before the release of Debian 9 "Stretch". > > There is conversation on https://bugs.debian.org/844258 about maybe > bumping the minimum supported baseline for s390x or asking IBM if they > want to invest in getting Go working on older s390x. That bug was > closed because the main focus of that bug was fixed. So I'm opening a > new bug.
I think it's pretty clear that IBM is not interested in supporting these new workloads on older machines (which for all points and purposes was Docker). On the other hand there was also more direct investment made by IBM into making some software perform better on newer s390x machines rather than "just" paying Linux vendors to do so. I'm at this point still not sure what the right answer here can be. [0] suggests that z10 BC/EC could be completely unsupported in two years time or so, but it is not today. I suppose from IBM's point of view the whole point of follow-on service is to keep the existing services running, not to enable new ones. I suppose there are at least three options: a) Depend on some sort of architectural support metapackage. This solves the problem of missing runtime support. It does not yet solve the problem on the builder side (where one is still a z10[1]). It might also push us into a dependency hell where packages pick up dependencies on such a metapackage and then a large part becomes uninstallable. The relatively unavailability of golang-go across various Debian ports makes this unlikely, though. b) Get resources somewhere to port the instruction generator to a different baseline. I know Aurelien has tried but I also think it's unfair to expect from him to do this work. Especially when most of the benefit here is to have it in Debian so that Ubuntu/Canonical has less work to do. c) Settle with a GCC-based Go on s390x instead of the reference implementation and make that work somehow. I'm not sure what the blockers there would be but that would seem to be theoretically beneficial to various other ports as well. Kind regards Philipp Kern [0] https://www-03.ibm.com/support/techdocs/atsmastr.nsf/5cb5ed706d254a8186256c71006d2e0a/74f007db22182c6b86257e06006f7a73/$FILE/IBM%20Mainframe%20Life%20Cycle%20History%20V2.0b%20-%20July%2017,%202017.pdf [1] Which I'd be willing to retire.
signature.asc
Description: OpenPGP digital signature