Re: [MacPorts] #73086: Go is broken on 10.13–10.15

[MacPorts]

#73086: Go is broken on 10.13–10.15
---------------------------+------------------------------------------
  Reporter:  barracuda156  |      Owner:  herbygillot
      Type:  defect        |     Status:  assigned
  Priority:  High          |  Milestone:
 Component:  ports         |    Version:  2.11.5
Resolution:                |   Keywords:  catalina, mojave, highsierra
      Port:  go            |
---------------------------+------------------------------------------

Comment (by herbygillot):

 In [changeset:"708bc073d94e34fc95e458f7faadb0728271da42/macports-ports"
 708bc073d94e34fc95e458f7faadb0728271da42/macports-ports] (master):
 {{{
 #!ConfigurableCommitTicketReference repository="macports-ports"
 revision="708bc073d94e34fc95e458f7faadb0728271da42"
 go_toolchain, go: offer 1.24 on 10.13 through 10.15

 The four go_toolchain.min_default entries held 1.21 through 1.24 at darwin
 19 and 20 while min_darwin claimed 17, so that the toolchains could be
 built
 and tried on the older systems before `go` moved anyone onto them. They
 have
 now been built there, and the entries are deleted.

 The buildbots did it over 5f128a4f56f: go-1.22, go-1.23 and go-1.24 on
 10.13
 and 10.14, go-1.23 and go-1.24 on 10.15 where 1.22 already had an archive.
 Eight first-time builds, no failures. Each of them had to bootstrap
 through
 make.bash, which runs upstream's prebuilt binary, so 10.13 started
 go1.24.13.darwin-amd64 -- a Mach-O declaring minos 11.0 -- and built a
 toolchain with it. That is the claim the load-time floors rest on, and it
 is
 now demonstrated rather than argued from symbol tables.

 `go` therefore rises from 1.20 to 1.24 on 10.13 and 10.14, and from 1.22
 to
 1.24 on 10.15. The version increases, so it is an ordinary upgrade and the
 epoch is untouched. 11 and later are unaffected.

 The table stays, empty, which is the state to expect between series. It is
 the mechanism for lowering the next floor ahead of the evidence for it. An
 absent entry needs no placeholder: the array is simply never created, and
 info exists returns 0 for every key under both Tcl 8.6 and 9.0.

 go-1.22, go-1.23 and go-1.24 carried a comment saying `go` would not
 follow
 them this far back, which was true only while the entries existed.
 Removed,
 and the general point recorded once against go_toolchain.packaged instead:
 packaging a series is not the same as `go` selecting it. `go` takes the
 newest series a system is offered, so a series sharing a floor with a
 newer
 one is never reached by default. That is now 1.20, 1.22, 1.23 and 1.25,
 packaged so a build can pin them.

 Worth being clear about what was and was not established. These are
 builds:
 the toolchain compiles, links and bootstraps on those systems. Nobody has
 run the standard library test suite there, so this does not say every
 package behaves on 10.13 the way it does on 12. The ports builders never
 invoke `port test`, so that answer has to come from hardware we do not
 have.

 See: https://trac.macports.org/ticket/73086
 }}}

-- 
Ticket URL: <https://trac.macports.org/ticket/73086#comment:7>
MacPorts <https://www.macports.org/>
Ports system for macOS