gosora/common/utils.go

534 lines
13 KiB
Go
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/*
*
* Utility Functions And Stuff
* Copyright Azareal 2017 - 2020
*
*/
package common
import (
"crypto/rand"
"encoding/base32"
"encoding/base64"
"encoding/json"
"errors"
"fmt"
"html"
"io/ioutil"
"math"
"os"
"strconv"
"strings"
"time"
"unicode"
)
// Version stores a Gosora version
type Version struct {
Major int
Minor int
Patch int
Tag string
TagID int
}
// TODO: Write a test for this
func (ver *Version) String() (out string) {
out = strconv.Itoa(ver.Major) + "." + strconv.Itoa(ver.Minor) + "." + strconv.Itoa(ver.Patch)
if ver.Tag != "" {
out += "-" + ver.Tag
if ver.TagID != 0 {
out += strconv.Itoa(ver.TagID)
}
}
return
}
// GenerateSafeString is for generating a cryptographically secure set of random bytes which is base64 encoded and safe for URLs
// TODO: Write a test for this
func GenerateSafeString(len int) (string, error) {
rb := make([]byte, len)
_, err := rand.Read(rb)
if err != nil {
return "", err
}
return base64.URLEncoding.EncodeToString(rb), nil
}
// GenerateStd32SafeString is for generating a cryptographically secure set of random bytes which is base32 encoded
// ? - Safe for URLs? Mostly likely due to the small range of characters
func GenerateStd32SafeString(len int) (string, error) {
rb := make([]byte, len)
_, err := rand.Read(rb)
if err != nil {
return "", err
}
return base32.StdEncoding.EncodeToString(rb), nil
}
// TODO: Write a test for this
func RelativeTimeFromString(in string) (string, error) {
if in == "" {
return "", nil
}
t, err := time.Parse("2006-01-02 15:04:05", in)
if err != nil {
return "", err
}
return RelativeTime(t), nil
}
// TODO: Write a test for this
func RelativeTime(t time.Time) string {
diff := time.Since(t)
hours := diff.Hours()
secs := diff.Seconds()
weeks := int(hours / 24 / 7)
months := int(hours / 24 / 31)
switch {
case months > 3:
if t.Year() != time.Now().Year() {
//return t.Format("Mon Jan 2 2006")
return t.Format("Jan 2 2006")
}
return t.Format("Jan 2")
case months > 1:
return fmt.Sprintf("%d months ago", months)
case months == 1:
return "a month ago"
case weeks > 1:
return fmt.Sprintf("%d weeks ago", weeks)
case int(hours/24) == 7:
return "a week ago"
case int(hours/24) == 1:
return "1 day ago"
case int(hours/24) > 1:
return fmt.Sprintf("%d days ago", int(hours/24))
case secs <= 1:
return "a moment ago"
case secs < 60:
return fmt.Sprintf("%d seconds ago", int(secs))
case secs < 120:
return "a minute ago"
case secs < 3600:
return fmt.Sprintf("%d minutes ago", int(secs/60))
case secs < 7200:
return "an hour ago"
}
return fmt.Sprintf("%d hours ago", int(secs/60/60))
}
// TODO: Finish this faster and more localised version of RelativeTime
/*
// TODO: Write a test for this
// ! Experimental
func RelativeTimeBytes(t time.Time, lang int) []byte {
diff := time.Since(t)
hours := diff.Hours()
secs := diff.Seconds()
weeks := int(hours / 24 / 7)
months := int(hours / 24 / 31)
switch {
case months > 3:
if t.Year() != time.Now().Year() {
return []byte(t.Format(phrases.RTime.MultiYear(lang)))
}
return []byte(t.Format(phrases.RTime.SingleYear(lang)))
case months > 1:
return phrases.RTime.Months(lang, months)
case months == 1:
return phrases.RTime.Month(lang)
case weeks > 1:
return phrases.RTime.Weeks(lang, weeks)
case int(hours/24) == 7:
return phrases.RTime.Week(lang)
case int(hours/24) == 1:
return phrases.RTime.Day(lang)
case int(hours/24) > 1:
return phrases.RTime.Days(lang, int(hours/24))
case secs <= 1:
return phrases.RTime.Moment(lang)
case secs < 60:
return phrases.RTime.Seconds(lang, int(secs))
case secs < 120:
return phrases.RTime.Minute(lang)
case secs < 3600:
return phrases.RTime.Minutes(lang, int(secs/60))
case secs < 7200:
return phrases.RTime.Hour(lang)
}
return phrases.RTime.Hours(lang, int(secs/60/60))
}
*/
var pMs = 1000
var pSec = pMs * 1000
var pMin = pSec * 60
var pHour = pMin * 60
var pDay = pHour * 24
func ConvertPerfUnit(quan float64) (out float64, unit string) {
f := func() (float64, string) {
switch {
case quan >= float64(pDay):
return quan / float64(pDay), "d"
case quan >= float64(pHour):
return quan / float64(pHour), "h"
case quan >= float64(pMin):
return quan / float64(pMin), "m"
case quan >= float64(pSec):
return quan / float64(pSec), "s"
case quan >= float64(pMs):
return quan / float64(pMs), "ms"
}
return quan, "μs"
}
out, unit = f()
return math.Ceil(out), unit
}
// TODO: Write a test for this
func ConvertByteUnit(bytes float64) (float64, string) {
switch {
case bytes >= float64(Petabyte):
return bytes / float64(Petabyte), "PB"
case bytes >= float64(Terabyte):
return bytes / float64(Terabyte), "TB"
case bytes >= float64(Gigabyte):
return bytes / float64(Gigabyte), "GB"
case bytes >= float64(Megabyte):
return bytes / float64(Megabyte), "MB"
case bytes >= float64(Kilobyte):
return bytes / float64(Kilobyte), "KB"
}
return bytes, " bytes"
}
// TODO: Write a test for this
func ConvertByteInUnit(bytes float64, unit string) (count float64) {
switch unit {
case "PB":
count = bytes / float64(Petabyte)
case "TB":
count = bytes / float64(Terabyte)
case "GB":
count = bytes / float64(Gigabyte)
case "MB":
count = bytes / float64(Megabyte)
case "KB":
count = bytes / float64(Kilobyte)
default:
count = 0.1
}
if count < 0.1 {
count = 0.1
}
return
}
// TODO: Write a test for this
// TODO: Localise this?
func FriendlyUnitToBytes(quantity int, unit string) (bytes int, err error) {
switch unit {
case "PB":
bytes = quantity * Petabyte
case "TB":
bytes = quantity * Terabyte
case "GB":
bytes = quantity * Gigabyte
case "MB":
bytes = quantity * Megabyte
case "KB":
bytes = quantity * Kilobyte
case "":
// Do nothing
default:
return bytes, errors.New("Unknown unit")
}
return bytes, nil
}
// TODO: Write a test for this
// TODO: Re-add T as int64
func ConvertUnit(num int) (int, string) {
switch {
case num >= 1000000000000:
return num / 1000000000000, "T"
case num >= 1000000000:
return num / 1000000000, "B"
case num >= 1000000:
return num / 1000000, "M"
case num >= 1000:
return num / 1000, "K"
}
return num, ""
}
// TODO: Write a test for this
// TODO: Re-add quadrillion as int64
// TODO: Re-add trillion as int64
func ConvertFriendlyUnit(num int) (int, string) {
switch {
case num >= 1000000000000000:
return 0, " quadrillion"
case num >= 1000000000000:
return 0, " trillion"
case num >= 1000000000:
return num / 1000000000, " billion"
case num >= 1000000:
return num / 1000000, " million"
case num >= 1000:
return num / 1000, " thousand"
}
return num, ""
}
// TODO: Make slugs optional for certain languages across the entirety of Gosora?
// TODO: Let plugins replace NameToSlug and the URL building logic with their own
/*func NameToSlug(name string) (slug string) {
// TODO: Do we want this reliant on config file flags? This might complicate tests and oddball uses
if !Config.BuildSlugs {
return ""
}
name = strings.TrimSpace(name)
name = strings.Replace(name, " ", " ", -1)
for _, char := range name {
if unicode.IsLower(char) || unicode.IsNumber(char) {
slug += string(char)
} else if unicode.IsUpper(char) {
slug += string(unicode.ToLower(char))
} else if unicode.IsSpace(char) {
slug += "-"
}
}
if slug == "" {
slug = "untitled"
}
return slug
}*/
// TODO: Make slugs optional for certain languages across the entirety of Gosora?
// TODO: Let plugins replace NameToSlug and the URL building logic with their own
func NameToSlug(name string) (slug string) {
// TODO: Do we want this reliant on config file flags? This might complicate tests and oddball uses
if !Config.BuildSlugs {
return ""
}
name = strings.TrimSpace(name)
name = strings.Replace(name, " ", " ", -1)
var sb strings.Builder
for _, char := range name {
if unicode.IsLower(char) || unicode.IsNumber(char) {
sb.WriteRune(char)
} else if unicode.IsUpper(char) {
sb.WriteRune(unicode.ToLower(char))
} else if unicode.IsSpace(char) {
sb.WriteByte('-')
}
}
if sb.Len() == 0 {
return "untitled"
}
return sb.String()
}
// TODO: Write a test for this
func HasSuspiciousEmail(email string) bool {
lowEmail := strings.ToLower(email)
// TODO: Use a more flexible blacklist, perhaps with a similar mechanism to the HTML tag registration system in PreparseMessage()
if !strings.Contains(lowEmail, "@") || strings.Contains(lowEmail, "casino") || strings.Contains(lowEmail, "viagra") || strings.Contains(lowEmail, "pharma") || strings.Contains(lowEmail, "pill") {
return true
}
var dotCount, shortBits, currentSegmentLength int
for _, char := range lowEmail {
if char == '.' {
dotCount++
if currentSegmentLength < 3 {
shortBits++
}
currentSegmentLength = 0
} else {
currentSegmentLength++
}
}
return dotCount > 7 || shortBits > 2
}
func unmarshalJsonFile(name string, in interface{}) error {
data, err := ioutil.ReadFile(name)
if err != nil {
return err
}
return json.Unmarshal(data, in)
}
func unmarshalJsonFileIgnore404(name string, in interface{}) error {
data, err := ioutil.ReadFile(name)
if err == os.ErrPermission || err == os.ErrClosed {
return err
} else if err != nil {
return nil
}
return json.Unmarshal(data, in)
}
// TODO: Write a test for this
func CanonEmail(email string) string {
email = strings.ToLower(email)
// Gmail emails are equivalent without the dots
espl := strings.Split(email, "@")
if len(espl) >= 2 && espl[1] == "gmail.com" {
return strings.Replace(espl[0], ".", "", -1) + "@" + espl[1]
}
return email
}
// TODO: Write a test for this
func createFile(name string) error {
f, err := os.Create(name)
if err != nil {
return err
}
return f.Close()
}
// TODO: Write a test for this
func writeFile(name, content string) (err error) {
f, err := os.Create(name)
if err != nil {
return err
}
_, err = f.WriteString(content)
if err != nil {
return err
}
err = f.Sync()
if err != nil {
return err
}
return f.Close()
}
// TODO: Write a test for this
func Stripslashes(text string) string {
text = strings.Replace(text, "/", "", -1)
return strings.Replace(text, "\\", "", -1)
}
// The word counter might run into problems with some languages where words aren't as obviously demarcated, I would advise turning it off in those cases, or if it becomes annoying in general, really.
func WordCount(input string) (count int) {
input = strings.TrimSpace(input)
if input == "" {
return 0
}
var inSpace bool
for _, value := range input {
if unicode.IsSpace(value) || unicode.IsPunct(value) {
if !inSpace {
inSpace = true
}
} else if inSpace {
count++
inSpace = false
}
}
return count + 1
}
// TODO: Write a test for this
func GetLevel(score int) (level int) {
var base float64 = 25
var current, prev float64
var expFactor = 2.8
for i := 1; ; i++ {
_, bit := math.Modf(float64(i) / 10)
if bit == 0 {
expFactor += 0.1
}
current = base + math.Pow(float64(i), expFactor) + (prev / 3)
prev = current
if float64(score) < current {
break
}
level++
}
return level
}
// TODO: Write a test for this
func GetLevelScore(getLevel int) (score int) {
var base float64 = 25
var current float64
var expFactor = 2.8
for i := 1; i <= getLevel; i++ {
_, bit := math.Modf(float64(i) / 10)
if bit == 0 {
expFactor += 0.1
}
current = base + math.Pow(float64(i), expFactor) + (current / 3)
//fmt.Println("level: ", i)
//fmt.Println("current: ", current)
}
return int(math.Ceil(current))
}
// TODO: Write a test for this
func GetLevels(maxLevel int) []float64 {
var base float64 = 25
var current, prev float64 // = 0
var expFactor = 2.8
var out []float64
out = append(out, 0)
for i := 1; i <= maxLevel; i++ {
_, bit := math.Modf(float64(i) / 10)
if bit == 0 {
expFactor += 0.1
}
current = base + math.Pow(float64(i), expFactor) + (prev / 3)
prev = current
out = append(out, current)
}
return out
}
// TODO: Write a test for this
// SanitiseSingleLine is a generic function for escaping html entities and removing silly characters from usernames and topic titles. It also strips newline characters
func SanitiseSingleLine(in string) string {
in = strings.Replace(in, "\n", "", -1)
in = strings.Replace(in, "\r", "", -1)
return SanitiseBody(in)
}
// TODO: Write a test for this
// TODO: Add more strange characters
// TODO: Strip all sub-32s minus \r and \n?
// SanitiseBody is the same as SanitiseSingleLine, but it doesn't strip newline characters
func SanitiseBody(in string) string {
in = strings.Replace(in, "", "", -1) // Strip Zero length space
in = html.EscapeString(in)
return strings.TrimSpace(in)
}
func BuildSlug(slug string, id int) string {
if slug == "" || !Config.BuildSlugs {
return strconv.Itoa(id)
}
return slug + "." + strconv.Itoa(id)
}