open-vault/helper/strutil/strutil.go
2019-02-05 16:02:15 -05:00

426 lines
10 KiB
Go

package strutil
import (
"encoding/base64"
"encoding/json"
"fmt"
"sort"
"strings"
"github.com/hashicorp/errwrap"
glob "github.com/ryanuber/go-glob"
)
// StrListContainsGlob looks for a string in a list of strings and allows
// globs.
func StrListContainsGlob(haystack []string, needle string) bool {
for _, item := range haystack {
if glob.Glob(item, needle) {
return true
}
}
return false
}
// StrListContains looks for a string in a list of strings.
func StrListContains(haystack []string, needle string) bool {
for _, item := range haystack {
if item == needle {
return true
}
}
return false
}
// StrListSubset checks if a given list is a subset
// of another set
func StrListSubset(super, sub []string) bool {
for _, item := range sub {
if !StrListContains(super, item) {
return false
}
}
return true
}
// ParseDedupAndSortStrings parses a comma separated list of strings
// into a slice of strings. The return slice will be sorted and will
// not contain duplicate or empty items.
func ParseDedupAndSortStrings(input string, sep string) []string {
input = strings.TrimSpace(input)
parsed := []string{}
if input == "" {
// Don't return nil
return parsed
}
return RemoveDuplicates(strings.Split(input, sep), false)
}
// ParseDedupLowercaseAndSortStrings parses a comma separated list of
// strings into a slice of strings. The return slice will be sorted and
// will not contain duplicate or empty items. The values will be converted
// to lower case.
func ParseDedupLowercaseAndSortStrings(input string, sep string) []string {
input = strings.TrimSpace(input)
parsed := []string{}
if input == "" {
// Don't return nil
return parsed
}
return RemoveDuplicates(strings.Split(input, sep), true)
}
// ParseKeyValues parses a comma separated list of `<key>=<value>` tuples
// into a map[string]string.
func ParseKeyValues(input string, out map[string]string, sep string) error {
if out == nil {
return fmt.Errorf("'out is nil")
}
keyValues := ParseDedupLowercaseAndSortStrings(input, sep)
if len(keyValues) == 0 {
return nil
}
for _, keyValue := range keyValues {
shards := strings.Split(keyValue, "=")
if len(shards) != 2 {
return fmt.Errorf("invalid <key,value> format")
}
key := strings.TrimSpace(shards[0])
value := strings.TrimSpace(shards[1])
if key == "" || value == "" {
return fmt.Errorf("invalid <key,value> pair: key: %q value: %q", key, value)
}
out[key] = value
}
return nil
}
// ParseArbitraryKeyValues parses arbitrary <key,value> tuples. The input
// can be one of the following:
// * JSON string
// * Base64 encoded JSON string
// * Comma separated list of `<key>=<value>` pairs
// * Base64 encoded string containing comma separated list of
// `<key>=<value>` pairs
//
// Input will be parsed into the output parameter, which should
// be a non-nil map[string]string.
func ParseArbitraryKeyValues(input string, out map[string]string, sep string) error {
input = strings.TrimSpace(input)
if input == "" {
return nil
}
if out == nil {
return fmt.Errorf("'out' is nil")
}
// Try to base64 decode the input. If successful, consider the decoded
// value as input.
inputBytes, err := base64.StdEncoding.DecodeString(input)
if err == nil {
input = string(inputBytes)
}
// Try to JSON unmarshal the input. If successful, consider that the
// metadata was supplied as JSON input.
err = json.Unmarshal([]byte(input), &out)
if err != nil {
// If JSON unmarshalling fails, consider that the input was
// supplied as a comma separated string of 'key=value' pairs.
if err = ParseKeyValues(input, out, sep); err != nil {
return errwrap.Wrapf("failed to parse the input: {{err}}", err)
}
}
// Validate the parsed input
for key, value := range out {
if key != "" && value == "" {
return fmt.Errorf("invalid value for key %q", key)
}
}
return nil
}
// ParseStringSlice parses a `sep`-separated list of strings into a
// []string with surrounding whitespace removed.
//
// The output will always be a valid slice but may be of length zero.
func ParseStringSlice(input string, sep string) []string {
input = strings.TrimSpace(input)
if input == "" {
return []string{}
}
splitStr := strings.Split(input, sep)
ret := make([]string, len(splitStr))
for i, val := range splitStr {
ret[i] = strings.TrimSpace(val)
}
return ret
}
// ParseArbitraryStringSlice parses arbitrary string slice. The input
// can be one of the following:
// * JSON string
// * Base64 encoded JSON string
// * `sep` separated list of values
// * Base64-encoded string containing a `sep` separated list of values
//
// Note that the separator is ignored if the input is found to already be in a
// structured format (e.g., JSON)
//
// The output will always be a valid slice but may be of length zero.
func ParseArbitraryStringSlice(input string, sep string) []string {
input = strings.TrimSpace(input)
if input == "" {
return []string{}
}
// Try to base64 decode the input. If successful, consider the decoded
// value as input.
inputBytes, err := base64.StdEncoding.DecodeString(input)
if err == nil {
input = string(inputBytes)
}
ret := []string{}
// Try to JSON unmarshal the input. If successful, consider that the
// metadata was supplied as JSON input.
err = json.Unmarshal([]byte(input), &ret)
if err != nil {
// If JSON unmarshalling fails, consider that the input was
// supplied as a separated string of values.
return ParseStringSlice(input, sep)
}
if ret == nil {
return []string{}
}
return ret
}
// TrimStrings takes a slice of strings and returns a slice of strings
// with trimmed spaces
func TrimStrings(items []string) []string {
ret := make([]string, len(items))
for i, item := range items {
ret[i] = strings.TrimSpace(item)
}
return ret
}
// RemoveDuplicates removes duplicate and empty elements from a slice of
// strings. This also may convert the items in the slice to lower case and
// returns a sorted slice.
func RemoveDuplicates(items []string, lowercase bool) []string {
itemsMap := map[string]bool{}
for _, item := range items {
item = strings.TrimSpace(item)
if lowercase {
item = strings.ToLower(item)
}
if item == "" {
continue
}
itemsMap[item] = true
}
items = make([]string, 0, len(itemsMap))
for item := range itemsMap {
items = append(items, item)
}
sort.Strings(items)
return items
}
// RemoveEmpty removes empty elements from a slice of
// strings
func RemoveEmpty(items []string) []string {
if len(items) == 0 {
return items
}
itemsSlice := make([]string, 0, len(items))
for _, item := range items {
if item == "" {
continue
}
itemsSlice = append(itemsSlice, item)
}
return itemsSlice
}
// EquivalentSlices checks whether the given string sets are equivalent, as in,
// they contain the same values.
func EquivalentSlices(a, b []string) bool {
if a == nil && b == nil {
return true
}
if a == nil || b == nil {
return false
}
// First we'll build maps to ensure unique values
mapA := map[string]bool{}
mapB := map[string]bool{}
for _, keyA := range a {
mapA[keyA] = true
}
for _, keyB := range b {
mapB[keyB] = true
}
// Now we'll build our checking slices
var sortedA, sortedB []string
for keyA := range mapA {
sortedA = append(sortedA, keyA)
}
for keyB := range mapB {
sortedB = append(sortedB, keyB)
}
sort.Strings(sortedA)
sort.Strings(sortedB)
// Finally, compare
if len(sortedA) != len(sortedB) {
return false
}
for i := range sortedA {
if sortedA[i] != sortedB[i] {
return false
}
}
return true
}
// EqualStringMaps tests whether two map[string]string objects are equal.
// Equal means both maps have the same sets of keys and values. This function
// is 6-10x faster than a call to reflect.DeepEqual().
func EqualStringMaps(a, b map[string]string) bool {
if len(a) != len(b) {
return false
}
for k := range a {
v, ok := b[k]
if !ok || a[k] != v {
return false
}
}
return true
}
// StrListDelete removes the first occurrence of the given item from the slice
// of strings if the item exists.
func StrListDelete(s []string, d string) []string {
if s == nil {
return s
}
for index, element := range s {
if element == d {
return append(s[:index], s[index+1:]...)
}
}
return s
}
// GlobbedStringsMatch compares item to val with support for a leading and/or
// trailing wildcard '*' in item.
func GlobbedStringsMatch(item, val string) bool {
if len(item) < 2 {
return val == item
}
hasPrefix := strings.HasPrefix(item, "*")
hasSuffix := strings.HasSuffix(item, "*")
if hasPrefix && hasSuffix {
return strings.Contains(val, item[1:len(item)-1])
} else if hasPrefix {
return strings.HasSuffix(val, item[1:])
} else if hasSuffix {
return strings.HasPrefix(val, item[:len(item)-1])
}
return val == item
}
// AppendIfMissing adds a string to a slice if the given string is not present
func AppendIfMissing(slice []string, i string) []string {
if StrListContains(slice, i) {
return slice
}
return append(slice, i)
}
// MergeSlices adds an arbitrary number of slices together, uniquely
func MergeSlices(args ...[]string) []string {
all := map[string]struct{}{}
for _, slice := range args {
for _, v := range slice {
all[v] = struct{}{}
}
}
result := make([]string, 0, len(all))
for k := range all {
result = append(result, k)
}
sort.Strings(result)
return result
}
// Difference returns the set difference (A - B) of the two given slices. The
// result will also remove any duplicated values in set A regardless of whether
// that matches any values in set B.
func Difference(a, b []string, lowercase bool) []string {
if len(a) == 0 {
return a
}
if len(b) == 0 {
if !lowercase {
return a
}
newA := make([]string, len(a))
for i, v := range a {
newA[i] = strings.ToLower(v)
}
return newA
}
a = RemoveDuplicates(a, lowercase)
b = RemoveDuplicates(b, lowercase)
itemsMap := map[string]bool{}
for _, aVal := range a {
itemsMap[aVal] = true
}
// Perform difference calculation
for _, bVal := range b {
if _, ok := itemsMap[bVal]; ok {
itemsMap[bVal] = false
}
}
items := []string{}
for item, exists := range itemsMap {
if exists {
items = append(items, item)
}
}
sort.Strings(items)
return items
}