On Sun, May 01, 2022 at 10:23:18PM +0900, Michael Paquier wrote: > On Sun, May 01, 2022 at 01:07:06AM -0700, Noah Misch wrote: > > My annual audit for executables missing Windows icons turned up these: > > > > pginstall/bin/testclient.exe > > pginstall/bin/uri-regress.exe > > > > I was going to add the icons, but I felt the testclient.exe name is too > > generic-sounding to be installed. testclient originated in commit ebc8b7d. > > I > > recommend ceasing to install both programs under MSVC. (The GNU make build > > system does not install them.)
See also: a17fd67d2f2861ae0ce00d1aeefdf2facc47cd5e Build libpq test programs under MSVC. https://www.postgresql.org/message-id/74952229-b3b0-fe47-f958-4088529a3...@dunslane.net MSVC build system installs extra executables https://www.postgresql.org/message-id/e4233934-98a6-6f76-46a0-992c0f4f1...@dunslane.net Re: set TESTDIR from perl rather than Makefile I'm not really sure what the plan is for the TESTDIR patches. Is "vcregress alltaptests" still an interesting patch to pursue, or is that going to be obsoleted by meson build ? > But MSVC works differently. vcregress.pl does a TempInstall(), which > is a simple Install(), so isn't it going to be an issue for the tests > if these two tools are not installed anymore? Andrew didn't propose any mechanism for avoiding installation of the executables, so it would break the tests. However, at least cfbot currently doesn't run them anyway. One idea is if "vcregress install" accepted an option like "vcregress install check", which would mean "install extra binaries needed for running tests". Something maybe not much more elegant than this. next if ($insttype eq "client" && !grep { $_ eq $pf } @client_program_files); + next if ($pf =~ /testclient|uri-regress/);
