Update version-at-commit.sh to handle "development-" prefix tags

Added a long comment explaining how versioning relates to the release
process.

no issue #

Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <peter.amstutz@curii.com>

diff --git a/build/version-at-commit.sh b/build/version-at-commit.sh
index 50b1e300e79d65d17a5ca2ccc8d07adcfd32543d..d1a63c61c34e5e76cf1e166062dd3b8745a383ee 100755 (executable)
--- a/build/version-at-commit.sh
+++ b/build/version-at-commit.sh
@@ -5,7 +5,6 @@
 
 set -e -o pipefail
 commit="$1"
 
 set -e -o pipefail
 commit="$1"

-versionglob="[0-9].[0-9]*.[0-9]*"

 devsuffix="~dev"
 
 # automatically assign *development* version
 devsuffix="~dev"
 
 # automatically assign *development* version


@@ -20,24 +19,65 @@ devsuffix="~dev"
 #    than the commit where this branch joins main.
 #    -> take nearest tag X.Y.Z and assign X.Y.(Z+1)
 
 #    than the commit where this branch joins main.
 #    -> take nearest tag X.Y.Z and assign X.Y.(Z+1)
 

+# X.Y.Z releases where Z=0 are called major
+# releases and X.Y.Z releases where Z>1 are called point releases.
+#
+# The development process distinction is that X.Y.0 releases are
+# branched from main and then subsequent X.Y.Z releases cherry-pick
+# individual features from main onto the "X.Y-staging" branch.
+#
+# In semantic versioning terminology an "X.Y.0" release which only
+# increments Y is called a "minor" release but typically these
+# releases have significant changes that calling them "minor" in
+# communications with users feels misleading.
+#
+# Incrementing X is reserved for times when a release has significant
+# backwards-incompatible changes, which we don't do very often and try
+# to avoid.
+#
+# In order to assign a useful development version, we need to
+# determine if we're on the main branch (or a development branch off
+# main) or on a release branch.  We do this by looking at the point
+# where the current commit history branched from main.
+#
+# If a new major version appeared on a branch (not directly in the
+# history of main), the merge-base between main and the release should
+# be tagged as "development-X.Y.Z" so that version-at-commit can
+# figure out what to do.
+

 # 1. get the nearest tag with 'git describe'
 # 2. get the merge base between this commit and main
 # 3. if the tag is an ancestor of the merge base,
 #    (tag is older than merge base) increment minor version
 #    else, tag is newer than merge base, so increment point version
 
 # 1. get the nearest tag with 'git describe'
 # 2. get the merge base between this commit and main
 # 3. if the tag is an ancestor of the merge base,
 #    (tag is older than merge base) increment minor version
 #    else, tag is newer than merge base, so increment point version
 

-nearest_tag=$(git describe --tags --abbrev=0 --match "$versionglob" "$commit")
+nearest_tag=$(git describe --abbrev=0 "$commit")

 merge_base=$(git merge-base origin/main "$commit")
 
 if git merge-base --is-ancestor "$nearest_tag" "$merge_base" ; then
 merge_base=$(git merge-base origin/main "$commit")
 
 if git merge-base --is-ancestor "$nearest_tag" "$merge_base" ; then

+    # the nearest tag appears before the merge base with main (the
+    # branch point), so assume this is a tag for the previous major
+    # release (or a tag with the "development-" prefix indicating the
+    # point where a major release branched off).  Subsequent
+    # development versions are given the anticipated version for the
+    # next major release.
+    #

     # x.(y+1).0~devTIMESTAMP, where x.y.z is the newest version that does not contain $commit
     # grep reads the list of tags (-f) that contain $commit and filters them out (-v)
     # this prevents a newer tag from retroactively changing the versions of everything before it
     v=$(git tag | grep -vFf <(git tag --contains "$merge_base") | sort -Vr | head -n1 | perl -pe 's/(\d+)\.(\d+)\.\d+.*/"$1.".($2+1).".0"/e')
 else
     # x.(y+1).0~devTIMESTAMP, where x.y.z is the newest version that does not contain $commit
     # grep reads the list of tags (-f) that contain $commit and filters them out (-v)
     # this prevents a newer tag from retroactively changing the versions of everything before it
     v=$(git tag | grep -vFf <(git tag --contains "$merge_base") | sort -Vr | head -n1 | perl -pe 's/(\d+)\.(\d+)\.\d+.*/"$1.".($2+1).".0"/e')
 else

+    # the nearest tag comes after the merge base with main (the branch
+    # point).  Assume this means this is a point release branch,
+    # following a major release.
+    #

     # x.y.(z+1)~devTIMESTAMP, where x.y.z is the latest released ancestor of $commit
     v=$(echo $nearest_tag | perl -pe 's/(\d+)$/$1+1/e')
 fi
     # x.y.(z+1)~devTIMESTAMP, where x.y.z is the latest released ancestor of $commit
     v=$(echo $nearest_tag | perl -pe 's/(\d+)$/$1+1/e')
 fi

+
+# strip the "development-" prefix
+v=$(echo $v | perl -pe 's/^development-//')
+

 isodate=$(TZ=UTC git log -n1 --format=%cd --date=iso "$commit")
 ts=$(TZ=UTC date --date="$isodate" "+%Y%m%d%H%M%S")
 echo "${v}${devsuffix}${ts}"
 isodate=$(TZ=UTC git log -n1 --format=%cd --date=iso "$commit")
 ts=$(TZ=UTC date --date="$isodate" "+%Y%m%d%H%M%S")
 echo "${v}${devsuffix}${ts}"