Eli Zaretskii wrote:
> Looking at your test program, I see that you generate the seconds file
> name without deleting the first one.  When a file by the first
> generated name already exists, gen_tempname will indeed generate a
> different name.  But in the scenario I described, Emacs creates one
> temporary file, uses it, then deletes it, and only after that creates
> another file.

I'm adding this scenario to the unit test. Still, the unit test succeeds
when run once on:
  Linux, FreeBSD, NetBSD, OpenBSD, macOS, Solaris, Cygwin, native Windows.

Since this contradicts what you debugged on mingw, I ran the test 10000
times on native Windows. Result:
  - on 32-bit mingw, no failure.
  - on 64-bit mingw, around 30 failures (or around 10 failures with Paul's
    newest patch). That is, a probability of ca. 0.3% of getting the same
    file name as on the previous call.

Bruno