packaging: Use GRASS version for non-git builds

[sebastic]

As reported in #2610, rebuilding the GRASS sources (from the release tarballs) with no changes break the gdal-grass plugins because the GIS_H_VERSION value changes.

Non-git builds use the current timestamp in GRASS_HEADERS_GIT_HASH which is also used for GRASS_HEADERS_DATE.

This PR uses GRASS_VERSION_NUMBER for GRASS_HEADERS_GIT_HASH as a better value for non-git builds, that value won't change for simple rebuilds of the source tree.

[wenzeslaus]

Assuming a release when not using Git sounds good to me.

[sebastic]

Assuming a release when not using Git sounds good to me.

When git is available the commit hash and date are used:

# get git short hash + date of last change in GRASS headers
# (and anything else in include)
GRASS_HEADERS_GIT_HASH="${GRASS_VERSION_NUMBER}"
GRASS_HEADERS_GIT_DATE=`date -u +%FT%T%z | sed 's/\(..\)$/:\1/'`
AC_PATH_PROG(GIT, git, no)
if test "$GIT" != "no" ; then
   GRASS_VERSION_GIT=`$GIT rev-parse --short HEAD 2>/dev/null`
   if test -z "$GRASS_VERSION_GIT"; then
      GRASS_VERSION_GIT="exported"
   fi
   GRASS_HEADERS_GIT_HASH=`$GIT log -1 --pretty=format:"%h" -- "${SRCDIR}/include" 2>/dev/null`
   if test -z "$GRASS_HEADERS_GIT_HASH"; then
      GRASS_HEADERS_GIT_HASH=`date -u +%FT%T%z | sed 's/\(..\)$/:\1/'`
   fi
   GRASS_HEADERS_GIT_DATE=`date -d $($GIT log -1 --pretty=format:"%cI" -- "${SRCDIR}/include") -u +%FT%T%z | sed 's/\(..\)$/:\1/'` 2>/dev/null
   if test -z "$GRASS_HEADERS_GIT_DATE"; then
      GRASS_HEADERS_GIT_DATE=`date -u +%FT%T%z | sed 's/\(..\)$/:\1/'`
   fi
fi

[nilason]

You need to run autoconf269 to update the configure file.

[sebastic]

Generated files should not be committed.

[nilason]

Generated files should not be committed.

The generated file has been there in 20 years, since version 5. As long as it is there, the two should be in sync, no?
That being said, there is no reason -- as I see it -- this (i.e. generated file being part of source) have to remain the case indefinitely...

[nilason]

If git as a program is available, but building is made from a non-git repo (e.g. release tarball), the GRASS_HEADERS_GIT_HASH will still be reset with date on L152:

grass/configure.ac

Lines 144 to 158 in 5184a91

	AC_PATH_PROG(GIT, git, no)
	if test "$GIT" != "no" ; then
	GRASS_VERSION_GIT=`$GIT rev-parse --short HEAD 2>/dev/null`
	if test -z "$GRASS_VERSION_GIT"; then
	GRASS_VERSION_GIT="exported"
	fi
	GRASS_HEADERS_GIT_HASH=`$GIT log -1 --pretty=format:"%h" -- "${SRCDIR}/include" 2>/dev/null`
	if test -z "$GRASS_HEADERS_GIT_HASH"; then
	GRASS_HEADERS_GIT_HASH=`date -u +%FT%T%z | sed 's/\(..\)$/:\1/'`
	fi
	GRASS_HEADERS_GIT_DATE=`date -d $($GIT log -1 --pretty=format:"%cI" -- "${SRCDIR}/include") -u +%FT%T%z | sed 's/\(..\)$/:\1/'` 2>/dev/null
	if test -z "$GRASS_HEADERS_GIT_DATE"; then
	GRASS_HEADERS_GIT_DATE=`date -u +%FT%T%z | sed 's/\(..\)$/:\1/'`
	fi
	fi

Now, the question is whether L152 should change to:

      GRASS_HEADERS_GIT_HASH="${GRASS_VERSION_NUMBER}"

too, as a fallback if test on L150 fails?

[sebastic]

Since there is no --without git configure option, that might be good.

[nilason]

Looks good to me.
@wenzeslaus Please go ahead and merge if you still approve after latest changes.
@sebastic Thanks !