On Fri, 20 Oct 2017, I wrote:
> A quick look at this suggests that the problem is in pcretest. This is a
> test program for the PCRE library, and as such is not written to be
> robust against all possible inputs. I will look at this some more in due
> course, but as the 8.xx series is now in maintenance-only mode, I think
> I will probably decide not to do anything, unless there is a very
> trivial fix.
There is no trivial fix, and I do not think it is worth doing anything
about this. The comment in the pcretest code says this: "These vectors
store, end-to-end, a list of zero-terminated captured substring names,
each list itself being terminated by an empty name. Assume that 1024 is
plenty long enough for the few names we'll be testing."
The test program for current releases (the 10.xx series) is pcre2test,
which was completely re-written for PCRE2, and has a different input
syntax. It does not suffer from this problem, and is a lot more robust
than pcretest (which was only ever a quick hack).
