msys2

Disclaimer: I might have done things in not the best way - I'll happily
accept correction.

Here is some information about building on msys2, the rewrite of msys,
plus a proposal for a couple of tiny changes to support it.

The simplest way to install msys2 is via the chocolatey package manager
(See <https://chocolatey.org/docs/installation>)

    choco install -y msys2

Alternatively, download the installer and run it.

Once that's done, fire up an Msys2 shell and use its package manager to
install the required tools:

    pacman -S msys/bison \

        msys/flex \

        msys/make \

        msys/git \

        msys/perl \

        msys/ccache \

        msys/diffutils \

        mingw-w64-x86_64-toolchain \

        mingw-w64-i686-toolchain

If you want to run a buildfarm animal, there are a few other things you
will also need:

    pacman -S msys/perl-libwww msys/perl-Crypt-SSLeay msys/tar

If you want to run TAP tests, you need to install IPC::Run somewhere and
point PERL5LIB at it. There is no pacman package available. You will
also need:

    pacman -S msys/perl-Test-Simple msys/perl-Test-Harness

And there are a few very useful things it makes sense to install:

    pacman -S msys/vim msys/patch

Pre-build setup:

For 64 bit builds:

    unset MSYSTEM_CHOST # or export MSYSTEM_CHOST=x86_64-w64-mingw32

    export MSYSTEM=MINGW64

    export PATH=/mingw64/bin:$PATH

For 32 bit builds

    unset MSYSTEM_CHOST # or export MSYSTEM_CHOST=i686-w64-mingw32

    export MSYSTEM=MINGW32

    export PATH=/mingw32/bin:$PATH

build:

    ./configure --with-template=win32 ...

Things that fail:

* configure without a tamplate - there is a simple fix for this,
included in the attached patch
* pg_upgrade test - I was clearly not thorough enough with my fix in
commit 608a71095. It occurred to me that rather than using `uname
-s` and  putting "MINGW*|MSYS*" everywhere, it might be better to
use the "host_os" value from Makefile.global. The patch does it that
way, falling back to uname if necessary. But I'm with doing it the
other way of people prefer. Yes, I know I should use backticks
instead of $().
* the pg_dump TAP test 010_dump_connstr.pl chokes badly on $dname3 and
$dbname4. The commands complain about too many arguments. It seems
weird because jacana is doing this just fine.
* 32 bit builds only fail the circle regression test like this:

*** C:/tools/msys64/home/Administrator/bf/root/HEAD/pgsql/src/test/regress/expected/circle.out 2018-10-03 18:52:44.137775400 +0000
--- C:/tools/msys64/home/administrator/bf/root/HEAD/pgsql.build/src/test/regress/results/circle.out 2018-10-05 20:59:45.424014300 +0000
***************
*** 109,116 ****
WHERE (c1.f1 < c2.f1) AND ((c1.f1 <-> c2.f1) > 0)
ORDER BY distance, area(c1.f1), area(c2.f1);
five | one | two | distance
! ------+----------------+----------------+------------------
! | <(3,5),0> | <(1,2),3> | 0.60555127546399
| <(3,5),0> | <(5,1),3> | 1.47213595499958
| <(100,200),10> | <(100,1),115> | 74
| <(100,200),10> | <(1,2),100> | 111.370729772479
--- 109,116 ----
WHERE (c1.f1 < c2.f1) AND ((c1.f1 <-> c2.f1) > 0)
ORDER BY distance, area(c1.f1), area(c2.f1);
five | one | two | distance
! ------+----------------+----------------+-------------------
! | <(3,5),0> | <(1,2),3> | 0.605551275463989
| <(3,5),0> | <(5,1),3> | 1.47213595499958
| <(100,200),10> | <(100,1),115> | 74
| <(100,200),10> | <(1,2),100> | 111.370729772479

cheers

andrew