Make the test data source archive mostly-reproducible

This should keep us from having to upload a fresh archive to the
lookaside cache for a new version when the data hasn’t actually
changed.

get_test_data.sh

@@ -11,8 +11,13 @@ Usage: $1 VERSION
 
 Generate a source archive containing the test data files for pyMapVBVD that are
 tracked in git LFS and are not included in the default source archive. The
-result will be named pyMapVBVD-\${VERSION}-test-data.tar.xz and will be written
+result will be named pyMapVBVD-test-data.tar.xz and will be written
 into the current working directory.
+
+The script attempts to produce a reproducible archive to avoid uploading
+near-duplicate archives to the lookaside cache when the test data does not
+change from one version to the next. Accordingly, the top-level directory
+pyMapVBVD-\${VERSION}/ is omitted from the archived paths.
 EOF
 }
 
@@ -41,7 +46,7 @@ GITDIR="$(echo "${FORGEURL}" | sed -r 's@.*/@@')"
 #SOURCE0="${FORGEURL}/archive/${VERSION}/pdfminer.six-${VERSION}.tar.gz"
 #TARNAME="$(basename "${SOURCE0}")"
 TARDIR="pyMapVBVD-${VERSION}"
-NEWTAR="${TARDIR}-test-data.tar.xz"
+NEWTAR='pyMapVBVD-test-data.tar.xz'
 
 SAVEDIR="${PWD}"
 XDIR="$(mktemp -d)"
@@ -61,8 +66,20 @@ git lfs ls-files -n |
 		ln "${fn}" "${DESTDIR}/$(basename "${fn}")"
 	done
 
-cd "${XDIR}"
-tar -cv "${TARDIR}/" | xz -9e > "${NEWTAR}"
+cd "${XDIR}/${TARDIR}"
+# https://www.gnu.org/software/tar/manual/html_section/Reproducibility.html
+# We discard mtimes entirely, since git doesn’t store mtimes and they are just
+# the time we did the git clone.
+TZ=UTC LC_ALL=C tar \
+		--create --verbose \
+		--sort=name \
+		--format=posix \
+		--numeric-owner --owner=0 --group=0 \
+		--mode=go+u,go-w \
+		--pax-option='delete=atime,delete=ctime' \
+		--clamp-mtime --mtime='@0' \
+		. |
+	xz -9e > "${XDIR}/${NEWTAR}"
 
 cd "${SAVEDIR}"
 mv -v "${XDIR}/${NEWTAR}" .