On 2017-12-31 I added a new regression test to exercise some formerly
untested code in plpgsql, and it turns out that the addition exposed a
compiler-dependent bug that would send the test into an infinite loop.
Buildfarm member fulmar, for one, has been stuck since then; we only
noticed when its test output file filled up the disk :-(.
I've now pushed a patch that should fix the underlying problem.
If you don't routinely monitor your buildfarm member(s), please check to
see if any have been stuck building HEAD since January. A look at the
buildfarm webpage suggests that at least aholehole, dunlin, leech,
markhor, mastodon, and treepie might be suffering effects of this, as
they've not reported in for roughly the right amount of time.
regards, tom lane