hstow: use find(1)'s -path argument rather than filtering its output

The -path argument is included in POSIX.1-2017.

Also improve the awk code in the case where .hstow-always-adopt doesn't exist.

1 files changed, 59 insertions, 29 deletions

diff --git a/bin/hstow b/bin/hstow
index 7dbcf0ec..9e2e42c2 100755
--- a/bin/hstow
+++ b/bin/hstow
@@ -78,11 +78,37 @@ readlinks () {
 }
 
 disjoin_file () {
-    if [ -e "$DIR/$1" ]; then
-	while read -r line; do
-	    [ -n "$line" ] && printf "|$2" "$line"
-	done <"$DIR/$1" | sed 's#^.##; s#/#\\/#g'
-    fi
+    while read -r line; do
+	[ -n "$line" ] && printf "|$2" "$line"
+    done <"$DIR/$1" | sed 's#^.##; s#/#\\/#g'
+}
+
+globs_to_find_args () {
+    local file="$DIR/$1"; shift
+    printf '%s\n' "$@" | cat - $([ -e "$file" ] && echo "$file") \
+	| awk -F'\n' -vOFS='\t' \
+	      '/\/\*$/ { sub(/..$/, ""); prune[++c] = "./" $0; next }
+	       $0 { notpath[++d] = "./" $0 }
+	       # We want to prevent find(1) recursing into directories
+	       # of these names, but not prevent the remainder of the
+	       # find(1) expression from matching the names themselves.
+	       # This is what is correct for globs of the form "dir/*".
+	       # While it is true that neither of the find(1) commands
+	       # which use globs_to_find_args() match directories, such
+	       # that it would not make sense to add a line to one of
+	       # .hstow-*-ignore with the aim of excluding the contents
+	       # of a directory but not the directory itself, it could
+	       # be that the name is currently a symlink that we should
+	       # unstow.  (Also note that we do not pass -L to find(1),
+	       # so we will not recurse into symlinks to dirs anyway.)
+	       END { if (c) {
+			 print "(", "(", "-path", prune[1]
+			 for (i = 2; i <= c; i++)
+			     print "-o", "-path", prune[i]
+			 print ")", "-prune", "-o", "-name", "*", ")"
+		     }
+		     for (j in notpath) print "!", "-path", notpath[j]
+		   }'
 }
 
 usage () {
@@ -104,22 +130,24 @@ stow () {
 }
 
 stow1 () {
-    ignores="$(disjoin_file .hstow-local-ignore "./%s")"
-
-    # Files that (i) always/often have their symlinks replaced with
-    # regular files when applications access them; and (ii) we don't
-    # ever want to edit the copy under $DIR directly, but only via the
-    # link/copy under $HOME.
-    $always_adopt \
-	&& adoptp=1 \
-	|| adoptp="rel ~ /^($(disjoin_file .hstow-always-adopt "%s"))/"
-
-    find . ! -name . ! -type d ! -name "$cchars" \
-	 ! -name .gitignore \
-	 ! -name .hstow-local-ignore \
-	 ! -name .hstow-always-adopt \
-	 ! -name .hstow-unstow-ignore \
-	| awk -F'\n' -vOFS='\t' '! /^(\.\/\.git\/|'"$ignores"')/ \
+    if $always_adopt; then
+	adoptp=1
+    elif ! [ -e .hstow-always-adopt ]; then
+	adoptp=0
+    else
+	# EREs matching files that (i) always/often have their symlinks
+	# replaced with regular files when applications access them; and
+	# (ii) we don't ever want to edit the copy under $DIR directly,
+	# but only via the link/copy under $HOME.
+	# We might list globs in this file & convert them to EREs here.
+	adoptp="$(printf 'rel ~ /^(%s)/' \
+			 "$(disjoin_file .hstow-always-adopt "%s")")"
+    fi
+    find . $(globs_to_find_args .hstow-local-ignore ".git/*") \
+	 ! -name . ! -type d ! -name "$cchars" ! -name .gitignore \
+	 ! -name .hstow-local-ignore ! -name .hstow-always-adopt \
+	 ! -name .hstow-unstow-ignore -print \
+	| awk -F'\n' -vOFS='\t' '
 { rel = $1; gsub(/\/dot[-.]/, "/.", rel); gsub(/^\.\//, "", rel)
   dotdotslashes = rel
   sub(/[^\/]*$/, "", dotdotslashes)
@@ -157,11 +185,9 @@ stow1 () {
 unstow () {
     cd "$HOME"
 
-    # For speed, skip directories into which we'll never stow anything.
-    ignores="$(disjoin_file .hstow-unstow-ignore "./%s")"
-
-    dir_pat=".$(echo $DIR | cut -c$(echo $HOME | wc -m | tr -d ' ')-)/"
-    dirs_pat="$(echo "^($dir_pat|$ignores)" | sed -e 's#\.#\\.#g')"
+    # .hstow-unstow-ignore is a list of globs matching dirs into which
+    # we'll never stow anything.  We have this for the sake of speed.
+    #
     # awk's close() calls pclose(3), completing all the link deletions.
     # POSIX.1 "Utility Description Defaults", "Consequences of Errors"
     # implies that should rmdir(1) encounter a non-empty directory, it
@@ -169,9 +195,13 @@ unstow () {
     # Thus, here, -p means that we do not need to sort the operands.
     # We don't know the code with which rmdir(1) will exit, and if it is
     # 255 then xargs will give up.  So we wrap in a call to sh -c.
-    find . ! \( -user "$(id -un)" -o -group "$(id -gn)" \) \
-	 -prune -o -type l ! -name . ! -name "$cchars" -print \
-	| grep -Ev "$dirs_pat" | readlinks 0 true \
+    find . ! \( -user "$(id -un)" -o -group "$(id -gn)" \) -prune -o \
+	 $(globs_to_find_args \
+	       .hstow-unstow-ignore \
+	       "$(echo "$DIR" \
+		      | cut -c$((1 + $(echo "$HOME" | wc -m)))-)/*") \
+	 -type l ! -name . ! -name "$cchars" -print \
+	| readlinks 0 true \
 	| awk -F'\t' -vOFS='\t' '$2 ~ /^(\.\.\/)*\.STOW\/'"$NAME"'\// \
 { gsub(/"/, "\"'"'"'\"'"'"'\"", $1)
   printf "\"%s\"\n", $1 | "xargs -E '' -- rm -f"