Oh ... duh. We've been assuming that the strtoint change broke it,
but that's wrong. The test case that is failing is new as of yesterday,
and the correct answer is that it's never worked on Windows. See
https://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=3b7ab4380440d7b14ee390fabf39f6d87d7491e2
I think what's wrong is that src/tools/msvc/ecpg_regression.proj
needs to be taught that tests under ecpg/test/compat-oracle need
to be run with "-C ORACLE". Neither that directory nor that
switch existed before yesterday. There's already stuff in there
that knows about "-C INFORMIX", but beyond seeing the switch it
looks like line noise to me, so I'm not volunteering to fix it.
regards, tom lane