Possible regression in 2.13 series? "no viable configuration" warning

[Valentin Villenave]

Greetings everybody,

I've been pulling my hair all day long on this one, I may be getting
crazy or perhaps you'll be able to reproduce this as well;

Have a look at the following snippet, pretty straightforward:

<<
  \new Staff = "md" \relative c' {
    r4 r8 s s2
    r4 r8 s s2
    r4 e, r2 |
  }
  \new Staff = "mg" \relative c'' {
    b8 f' c' \change Staff = "md" <g, c f>~
    <g c f>2 \change Staff = "mg" |
    bes8 fes' ces' \change Staff = "md" <g, c f>~
    <g c f>2 \change Staff = "mg" |
    R1
  }
>>

This compiles fine, but with any recent build you'll get a "no viable
configuration found" warning. Since I didn't remember having
encountered such a warning with older versions, I tried other GUB
builds.

2.12.* and below: produces no warning
2.13.0 - 2.13.6: crashes with a segfault. (WTF?)
2.13.7 and above: "no viable configuration" warning.

Changing any note in the snippet prevents makes the warning disappear.
As does removing the last measure (although there is no cross-staff
voodoo in it).

Consider me a control freak, but I just hate it when my scores
produces compilation warnings that I can't fix.

Cheers,
Valentin.