investigate swprintf call failure (#37735)

[vtjnash]

No description provided.

[vtjnash]

That's odd:

ERRORS
    The printf() family of functions may fail if:
     [EILSEQ]           An invalid wide character code was encountered.

[vtjnash]

Hm, apparently this might be a known bug in FreeBSD (that %ls doesn't follow the posix spec)? https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=114578

[vtjnash]

FWIW, it appears that tabularasa_freebsd-x86_64-openstack_2 has always failed, and tabularasa_freebsd-x86_64-openstack_3 has always succeeded at this test, so I suspect a configuration issue compounded by that FreeBSD bug. (this current CI run will hopefully help clarify that)

https://build.julialang.org/#/builders/33/builds/4490 says we have on the broken bot:

      From worker 4:	LANG=
      From worker 4:	LC_CTYPE="C"
      From worker 4:	LC_COLLATE="C"
      From worker 4:	LC_TIME="C"
      From worker 4:	LC_NUMERIC="C"
      From worker 4:	LC_MONETARY="C"
      From worker 4:	LC_MESSAGES="C"
      From worker 4:	LC_ALL=

and on the working bot:

      From worker 4:	LANG=
      From worker 4:	LC_CTYPE="en_US.UTF-8"
      From worker 4:	LC_COLLATE="en_US.UTF-8"
      From worker 4:	LC_TIME="en_US.UTF-8"
      From worker 4:	LC_NUMERIC="en_US.UTF-8"
      From worker 4:	LC_MONETARY="en_US.UTF-8"
      From worker 4:	LC_MESSAGES="en_US.UTF-8"
      From worker 4:	LC_ALL=en_US.UTF-8

[vtjnash]

It appears that the macOS libc also inherited the same issue from BSD:

~/julia/test$ LANG= make ccall
    JULIA test/ccall
Test  (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
ccall      (1) |        started at 2020-10-09T14:32:01.183
ccall      (1) |         failed at 2020-10-09T14:32:41.973
Error During Test at /Users/jameson/julia/test/ccall.jl:1713
  Got exception outside of a @test
  InexactError: check_top_bit(UInt64, -1)
  Stacktrace:
    [1] throw_inexacterror(f::Symbol, #unused#::Type{UInt64}, val::Int32)
      @ Core ./boot.jl:602

[vtjnash]

Ah, sometimes it seems you just can't win. Looks like the tabularasa_linux-x86_64-nanosoldier2_1 has utf-8 support disabled?? and thus the warning message I added for that situation then causes some of the tests to fail.