Re: Use \if/\endif to remove non-libxml2 expected output in regression tests

Daniel Gustafsson <daniel(at)yesql(dot)se> writes:
>> On 12 Jun 2026, at 09:49, Michael Paquier <michael(at)paquier(dot)xyz> wrote:
>> I'd like to backpatch that down to v14, to ease future maintenance.
>> Any comments and/or objections?

> Skipping tests when the underlying library/platform support is missing is in
> line with how the TAP tests already do it. We would lose coverage of invoking
> XML functions without libxml support, which risk hiding bugs in the error-
> handling (which is already the case with the TAP tests etc).

Yeah. Also, I think this isn't really moving the support goalposts
very far, because xml_2.out will still require maintenance whenever
we touch xml.out, and that's the harder file to update if you don't
have a suitable version of libxml2 at hand. So I'm not in favor of
this as it stands. If we could find a way to get rid of xml_2.out
as well, then maybe dropping coverage of the error paths would be a
good tradeoff.

Diff'ing xml.out and xml_2.out, I see the differences are omissions
of input fragments in some error reports, eg:

@@ -335,8 +323,6 @@
<twoerrors>&idontexist;</unbalanced>
^
line 1: Opening and ending tag mismatch: twoerrors line 1 and unbalanced
-<twoerrors>&idontexist;</unbalanced>
- ^
SELECT xmlparse(document '<nosuchprefix:tag/>');
xmlparse
---------------------

I wonder if anyone's ever looked into exactly why that happens and
whether we could do something to restore the missing context.

regards, tom lane