Re: [PATCH 1/1] mingw: allow building with an MSYS2 runtime v3.x

From: Johannes Schindelin <johannes.schindelin@gmx.de>
Date: Mar 11, 2019, 16:32 UTC
Message-ID: <nycvar.QRO.7.76.6.1903111727510.41@tvgsbejvaqbjf.bet>
In-Reply-To: <xmqqr2bfmt5z.fsf@gitster-ct.c.googlers.com>

Hi Junio,

On Sun, 10 Mar 2019, Junio C Hamano wrote:

Show 10 quoted lines

> "Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
> writes:
> 
> > So let's invert that test case to test for *anything else* than a
> > version starting with "1" (for MSys). That should safeguard us for the
> > future, even if Cygwin ends up releasing versionsl like 314.272.65536.
> 
> An obvious alternative is to do "2" -> "[23]" *and* update the else to
> an explicit check for "1", *and* add an else that fails the build when
> it sees "4" and other unknown ones.

Right.

> I do not know how compatible MSYS2 runtime v4.x will be with the
> settings the part inside the ifneq() we see here protects, but you
> probably do, so I think this would be good enough.

I would expect Cygwin v4.x (which will be the basis for MSYS2 runtime v4.x) to be as diligent as Cygwin v3.x in maintaining as much backwards compatibility as possible.

Read: I trust "2 or higher" to be the answer.

Show 12 quoted lines

> The only case that makes any difference is when v4.x _mostly_ satisifies
> but slightly differs, affecting the built binary in a subtle way that is
> not discovered right away.  But in such a case, the more cautious "fail
> on '4' that we do not recognize" will not help much, as the first
> reaction after seeing the error would be to copy the settings that used
> to be OK with v2.x and v3.x anyway (which is what is being done
> here---we start with the assumption that the support needed for v3.x is
> largely the same as v2.x).
> 
> Will queue directly on top of... 'maint'?  'master'?
> 
> Let's say 'maint' and merge that up to 'master'.

Thank you.

For the record, the really super ugly hack I plan on implementing to make Windows builds pass in our Azure Pipeline is to build the git-sdk-64-minimal artifact (which contains all of the stuff needed to compile Git for Windows, such as `make`, `gcc`, etc) is to "install" a different `uname`: as the MINGW binaries live in `/mingw64/bin/` (unlike the MSYS binaries, which live in `/usr/bin/`, like `uname.exe`), I can write the following script into `/mingw64/bin/uname`:

	#!/bin/sh

	case "$*" in -r) echo 2.bogus;; *) /usr/bin/uname "$*";; esac

This will be the stop-gap solution for branches that are based on commits *without* this here patch.

It's ugly alright, but the only hack I could think of that actually will work.

Ciao, Dscho

Show 23 quoted lines

> 
> Thanks.
> 
> >
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> > ---
> >  config.mak.uname | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/config.mak.uname b/config.mak.uname
> > index c8b0e34c31..0207e895a4 100644
> > --- a/config.mak.uname
> > +++ b/config.mak.uname
> > @@ -569,7 +569,7 @@ ifneq (,$(wildcard ../THIS_IS_MSYSGIT))
> >  	NO_GETTEXT = YesPlease
> >  	COMPAT_CLFAGS += -D__USE_MINGW_ACCESS
> >  else
> > -	ifeq ($(shell expr "$(uname_R)" : '2\.'),2)
> > +	ifneq ($(shell expr "$(uname_R)" : '1\.'),2)
> >  		# MSys2
> >  		prefix = /usr/
> >  		ifeq (MINGW32,$(MSYSTEM))
>