summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorKerin Millar <kfm@plushkava.net>2024-06-27 18:19:40 +0100
committerKerin Millar <kfm@plushkava.net>2024-06-28 18:39:01 +0100
commit31fe87701736e117803a19211fb40bf7f518ae68 (patch)
tree0e0d8f93cf9f5765a1508f1ab3920caee07dfc2c
parentRectify a grammatical error in fetch() (diff)
downloadgentoo-functions-31fe87701736e117803a19211fb40bf7f518ae68.tar.gz
gentoo-functions-31fe87701736e117803a19211fb40bf7f518ae68.tar.bz2
gentoo-functions-31fe87701736e117803a19211fb40bf7f518ae68.zip
Align the indentation of inlined awk programs with their respective heredocs
This also makes it more pleasant to read where using the (proper) tab size of 8 spaces. Signed-off-by: Kerin Millar <kfm@plushkava.net>
-rw-r--r--functions.sh128
1 files changed, 64 insertions, 64 deletions
diff --git a/functions.sh b/functions.sh
index 4103f9d..3dd6239 100644
--- a/functions.sh
+++ b/functions.sh
@@ -152,12 +152,12 @@ hr()
length=80
fi
PATTERN=${1:--} awk -v "width=${length}" -f - <<-'EOF'
- BEGIN {
- while (length(rule) < width) {
- rule = rule substr(ENVIRON["PATTERN"], 1, width - length(rule))
- }
- print rule
+ BEGIN {
+ while (length(rule) < width) {
+ rule = rule substr(ENVIRON["PATTERN"], 1, width - length(rule))
}
+ print rule
+ }
EOF
}
@@ -227,31 +227,31 @@ is_anyof()
is_subset()
{
SENTINEL=${SENTINEL-'--'} awk -f - -- "$@" <<-'EOF'
- BEGIN {
- argc = ARGC
- ARGC = 1
- for (i = 1; i < argc; i++) {
- word = ARGV[i]
- if (word == ENVIRON["SENTINEL"]) {
- break
- } else {
- set1[word]
- }
- }
- if (i == 1 || argc - i < 2) {
- exit 1
- }
- for (i++; i < argc; i++) {
- word = ARGV[i]
- set2[word]
- }
- for (word in set2) {
- delete set1[word]
- }
- for (word in set1) {
- exit 1
+ BEGIN {
+ argc = ARGC
+ ARGC = 1
+ for (i = 1; i < argc; i++) {
+ word = ARGV[i]
+ if (word == ENVIRON["SENTINEL"]) {
+ break
+ } else {
+ set1[word]
}
}
+ if (i == 1 || argc - i < 2) {
+ exit 1
+ }
+ for (i++; i < argc; i++) {
+ word = ARGV[i]
+ set2[word]
+ }
+ for (word in set2) {
+ delete set1[word]
+ }
+ for (word in set1) {
+ exit 1
+ }
+ }
EOF
}
@@ -366,49 +366,49 @@ parallel_run()
quote_args()
{
awk -v q=\' -f - -- "$@" <<-'EOF'
- BEGIN {
- strictly_posix = length(ENVIRON["POSIXLY_CORRECT"])
- argc = ARGC
- ARGC = 1
- for (arg_idx = 1; arg_idx < argc; arg_idx++) {
- arg = ARGV[arg_idx]
- if (strictly_posix || arg !~ /[\001-\037\177]/) {
- gsub(q, q "\\" q q, arg)
- word = q arg q
- } else {
- # Use $'' quoting per Issue 8
- if (ord_by["\001"] == "") {
- for (i = 1; i < 32; i++) {
- char = sprintf("%c", i)
- ord_by[char] = i
- }
- ord_by["\177"] = 127
+ BEGIN {
+ strictly_posix = length(ENVIRON["POSIXLY_CORRECT"])
+ argc = ARGC
+ ARGC = 1
+ for (arg_idx = 1; arg_idx < argc; arg_idx++) {
+ arg = ARGV[arg_idx]
+ if (strictly_posix || arg !~ /[\001-\037\177]/) {
+ gsub(q, q "\\" q q, arg)
+ word = q arg q
+ } else {
+ # Use $'' quoting per Issue 8
+ if (ord_by["\001"] == "") {
+ for (i = 1; i < 32; i++) {
+ char = sprintf("%c", i)
+ ord_by[char] = i
}
- word = "$'"
- for (i = 1; i <= length(arg); i++) {
- char = substr(arg, i, 1)
- if (char == "\\") {
- word = word "\\\\"
- } else if (char == q) {
- word = word "\\'"
+ ord_by["\177"] = 127
+ }
+ word = "$'"
+ for (i = 1; i <= length(arg); i++) {
+ char = substr(arg, i, 1)
+ if (char == "\\") {
+ word = word "\\\\"
+ } else if (char == q) {
+ word = word "\\'"
+ } else {
+ ord = ord_by[char]
+ if (ord != "") {
+ word = word "\\" sprintf("%03o", ord)
} else {
- ord = ord_by[char]
- if (ord != "") {
- word = word "\\" sprintf("%03o", ord)
- } else {
- word = word char
- }
+ word = word char
}
}
- word = word q
- }
- line = line word
- if (arg_idx < argc - 1) {
- line = line " "
}
+ word = word q
+ }
+ line = line word
+ if (arg_idx < argc - 1) {
+ line = line " "
}
- print line
}
+ print line
+ }
EOF
}