962 lines
31 KiB
C#
962 lines
31 KiB
C#
//using System;
|
|
using System;
|
|
using System.Collections.Generic;
|
|
using System.Globalization;
|
|
using System.Security;
|
|
using System.Text;
|
|
using System.Text.RegularExpressions;
|
|
|
|
namespace Godot
|
|
{
|
|
public static class StringExtensions
|
|
{
|
|
private static int GetSliceCount(this string instance, string splitter)
|
|
{
|
|
if (instance.Empty() || splitter.Empty())
|
|
return 0;
|
|
|
|
int pos = 0;
|
|
int slices = 1;
|
|
|
|
while ((pos = instance.Find(splitter, pos)) >= 0)
|
|
{
|
|
slices++;
|
|
pos += splitter.Length;
|
|
}
|
|
|
|
return slices;
|
|
}
|
|
|
|
private static string GetSlicec(this string instance, char splitter, int slice)
|
|
{
|
|
if (!instance.Empty() && slice >= 0)
|
|
{
|
|
int i = 0;
|
|
int prev = 0;
|
|
int count = 0;
|
|
|
|
while (true)
|
|
{
|
|
if (instance[i] == 0 || instance[i] == splitter)
|
|
{
|
|
if (slice == count)
|
|
{
|
|
return instance.Substring(prev, i - prev);
|
|
}
|
|
else
|
|
{
|
|
count++;
|
|
prev = i + 1;
|
|
}
|
|
}
|
|
|
|
i++;
|
|
}
|
|
}
|
|
|
|
return string.Empty;
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file, return the path to the file without the extension.
|
|
// </summary>
|
|
public static string Basename(this string instance)
|
|
{
|
|
int index = instance.LastIndexOf('.');
|
|
|
|
if (index > 0)
|
|
return instance.Substring(0, index);
|
|
|
|
return instance;
|
|
}
|
|
|
|
// <summary>
|
|
// Return true if the strings begins with the given string.
|
|
// </summary>
|
|
public static bool BeginsWith(this string instance, string text)
|
|
{
|
|
return instance.StartsWith(text);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the bigrams (pairs of consecutive letters) of this string.
|
|
// </summary>
|
|
public static string[] Bigrams(this string instance)
|
|
{
|
|
string[] b = new string[instance.Length - 1];
|
|
|
|
for (int i = 0; i < b.Length; i++)
|
|
{
|
|
b[i] = instance.Substring(i, 2);
|
|
}
|
|
|
|
return b;
|
|
}
|
|
|
|
// <summary>
|
|
// Return a copy of the string with special characters escaped using the C language standard.
|
|
// </summary>
|
|
public static string CEscape(this string instance)
|
|
{
|
|
StringBuilder sb = new StringBuilder(string.Copy(instance));
|
|
|
|
sb.Replace("\\", "\\\\");
|
|
sb.Replace("\a", "\\a");
|
|
sb.Replace("\b", "\\b");
|
|
sb.Replace("\f", "\\f");
|
|
sb.Replace("\n", "\\n");
|
|
sb.Replace("\r", "\\r");
|
|
sb.Replace("\t", "\\t");
|
|
sb.Replace("\v", "\\v");
|
|
sb.Replace("\'", "\\'");
|
|
sb.Replace("\"", "\\\"");
|
|
sb.Replace("?", "\\?");
|
|
|
|
return sb.ToString();
|
|
}
|
|
|
|
// <summary>
|
|
// Return a copy of the string with escaped characters replaced by their meanings according to the C language standard.
|
|
// </summary>
|
|
public static string CUnescape(this string instance)
|
|
{
|
|
StringBuilder sb = new StringBuilder(string.Copy(instance));
|
|
|
|
sb.Replace("\\a", "\a");
|
|
sb.Replace("\\b", "\b");
|
|
sb.Replace("\\f", "\f");
|
|
sb.Replace("\\n", "\n");
|
|
sb.Replace("\\r", "\r");
|
|
sb.Replace("\\t", "\t");
|
|
sb.Replace("\\v", "\v");
|
|
sb.Replace("\\'", "\'");
|
|
sb.Replace("\\\"", "\"");
|
|
sb.Replace("\\?", "?");
|
|
sb.Replace("\\\\", "\\");
|
|
|
|
return sb.ToString();
|
|
}
|
|
|
|
// <summary>
|
|
// Change the case of some letters. Replace underscores with spaces, convert all letters to lowercase then capitalize first and every letter following the space character. For [code]capitalize camelCase mixed_with_underscores[/code] it will return [code]Capitalize Camelcase Mixed With Underscores[/code].
|
|
// </summary>
|
|
public static string Capitalize(this string instance)
|
|
{
|
|
string aux = instance.Replace("_", " ").ToLower();
|
|
string cap = string.Empty;
|
|
|
|
for (int i = 0; i < aux.GetSliceCount(" "); i++)
|
|
{
|
|
string slice = aux.GetSlicec(' ', i);
|
|
if (slice.Length > 0)
|
|
{
|
|
slice = char.ToUpper(slice[0]) + slice.Substring(1);
|
|
if (i > 0)
|
|
cap += " ";
|
|
cap += slice;
|
|
}
|
|
}
|
|
|
|
return cap;
|
|
}
|
|
|
|
// <summary>
|
|
// Perform a case-sensitive comparison to another string, return -1 if less, 0 if equal and +1 if greater.
|
|
// </summary>
|
|
public static int CasecmpTo(this string instance, string to)
|
|
{
|
|
if (instance.Empty())
|
|
return to.Empty() ? 0 : -1;
|
|
|
|
if (to.Empty())
|
|
return 1;
|
|
|
|
int instance_idx = 0;
|
|
int to_idx = 0;
|
|
|
|
while (true)
|
|
{
|
|
if (to[to_idx] == 0 && instance[instance_idx] == 0)
|
|
return 0; // We're equal
|
|
else if (instance[instance_idx] == 0)
|
|
return -1; // If this is empty, and the other one is not, then we're less... I think?
|
|
else if (to[to_idx] == 0)
|
|
return 1; // Otherwise the other one is smaller...
|
|
else if (instance[instance_idx] < to[to_idx]) // More than
|
|
return -1;
|
|
else if (instance[instance_idx] > to[to_idx]) // Less than
|
|
return 1;
|
|
|
|
instance_idx++;
|
|
to_idx++;
|
|
}
|
|
}
|
|
|
|
// <summary>
|
|
// Return true if the string is empty.
|
|
// </summary>
|
|
public static bool Empty(this string instance)
|
|
{
|
|
return string.IsNullOrEmpty(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return true if the strings ends with the given string.
|
|
// </summary>
|
|
public static bool EndsWith(this string instance, string text)
|
|
{
|
|
return instance.EndsWith(text);
|
|
}
|
|
|
|
// <summary>
|
|
// Erase [code]chars[/code] characters from the string starting from [code]pos[/code].
|
|
// </summary>
|
|
public static void Erase(this StringBuilder instance, int pos, int chars)
|
|
{
|
|
instance.Remove(pos, chars);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file, return the extension.
|
|
// </summary>
|
|
public static string Extension(this string instance)
|
|
{
|
|
int pos = instance.FindLast(".");
|
|
|
|
if (pos < 0)
|
|
return instance;
|
|
|
|
return instance.Substring(pos + 1, instance.Length);
|
|
}
|
|
|
|
// <summary>
|
|
// Find the first occurrence of a substring, return the starting position of the substring or -1 if not found. Optionally, the initial search index can be passed.
|
|
// </summary>
|
|
public static int Find(this string instance, string what, int from = 0)
|
|
{
|
|
return instance.IndexOf(what, StringComparison.OrdinalIgnoreCase);
|
|
}
|
|
|
|
// <summary>
|
|
// Find the last occurrence of a substring, return the starting position of the substring or -1 if not found. Optionally, the initial search index can be passed.
|
|
// </summary>
|
|
public static int FindLast(this string instance, string what)
|
|
{
|
|
return instance.LastIndexOf(what, StringComparison.OrdinalIgnoreCase);
|
|
}
|
|
|
|
// <summary>
|
|
// Find the first occurrence of a substring but search as case-insensitive, return the starting position of the substring or -1 if not found. Optionally, the initial search index can be passed.
|
|
// </summary>
|
|
public static int FindN(this string instance, string what, int from = 0)
|
|
{
|
|
return instance.IndexOf(what, StringComparison.Ordinal);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file, return the base directory.
|
|
// </summary>
|
|
public static string GetBaseDir(this string instance)
|
|
{
|
|
int basepos = instance.Find("://");
|
|
|
|
string rs = string.Empty;
|
|
string @base = string.Empty;
|
|
|
|
if (basepos != -1)
|
|
{
|
|
int end = basepos + 3;
|
|
rs = instance.Substring(end, instance.Length);
|
|
@base = instance.Substring(0, end);
|
|
}
|
|
else
|
|
{
|
|
if (instance.BeginsWith("/"))
|
|
{
|
|
rs = instance.Substring(1, instance.Length);
|
|
@base = "/";
|
|
}
|
|
else
|
|
{
|
|
rs = instance;
|
|
}
|
|
}
|
|
|
|
int sep = Mathf.Max(rs.FindLast("/"), rs.FindLast("\\"));
|
|
|
|
if (sep == -1)
|
|
return @base;
|
|
|
|
return @base + rs.substr(0, sep);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file, return the file and ignore the base directory.
|
|
// </summary>
|
|
public static string GetFile(this string instance)
|
|
{
|
|
int sep = Mathf.Max(instance.FindLast("/"), instance.FindLast("\\"));
|
|
|
|
if (sep == -1)
|
|
return instance;
|
|
|
|
return instance.Substring(sep + 1, instance.Length);
|
|
}
|
|
|
|
// <summary>
|
|
// Hash the string and return a 32 bits integer.
|
|
// </summary>
|
|
public static int Hash(this string instance)
|
|
{
|
|
int index = 0;
|
|
int hashv = 5381;
|
|
int c;
|
|
|
|
while ((c = (int)instance[index++]) != 0)
|
|
hashv = ((hashv << 5) + hashv) + c; // hash * 33 + c
|
|
|
|
return hashv;
|
|
}
|
|
|
|
// <summary>
|
|
// Convert a string containing an hexadecimal number into an int.
|
|
// </summary>
|
|
public static int HexToInt(this string instance)
|
|
{
|
|
int sign = 1;
|
|
|
|
if (instance[0] == '-')
|
|
{
|
|
sign = -1;
|
|
instance = instance.Substring(1);
|
|
}
|
|
|
|
if (!instance.StartsWith("0x"))
|
|
return 0;
|
|
|
|
return sign * int.Parse(instance.Substring(2), NumberStyles.HexNumber);
|
|
}
|
|
|
|
// <summary>
|
|
// Insert a substring at a given position.
|
|
// </summary>
|
|
public static string Insert(this string instance, int pos, string what)
|
|
{
|
|
return instance.Insert(pos, what);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file or directory, return true if the path is absolute.
|
|
// </summary>
|
|
public static bool IsAbsPath(this string instance)
|
|
{
|
|
return System.IO.Path.IsPathRooted(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path to a file or directory, return true if the path is relative.
|
|
// </summary>
|
|
public static bool IsRelPath(this string instance)
|
|
{
|
|
return !System.IO.Path.IsPathRooted(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether this string is a subsequence of the given string.
|
|
// </summary>
|
|
public static bool IsSubsequenceOf(this string instance, string text, bool case_insensitive)
|
|
{
|
|
int len = instance.Length;
|
|
|
|
if (len == 0)
|
|
return true; // Technically an empty string is subsequence of any string
|
|
|
|
if (len > text.Length)
|
|
return false;
|
|
|
|
int src = 0;
|
|
int tgt = 0;
|
|
|
|
while (instance[src] != 0 && text[tgt] != 0)
|
|
{
|
|
bool match = false;
|
|
|
|
if (case_insensitive)
|
|
{
|
|
char srcc = char.ToLower(instance[src]);
|
|
char tgtc = char.ToLower(text[tgt]);
|
|
match = srcc == tgtc;
|
|
}
|
|
else
|
|
{
|
|
match = instance[src] == text[tgt];
|
|
}
|
|
if (match)
|
|
{
|
|
src++;
|
|
if (instance[src] == 0)
|
|
return true;
|
|
}
|
|
|
|
tgt++;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether this string is a subsequence of the given string, considering case.
|
|
// </summary>
|
|
public static bool IsSubsequenceOf(this string instance, string text)
|
|
{
|
|
return instance.IsSubsequenceOf(text, false);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether this string is a subsequence of the given string, without considering case.
|
|
// </summary>
|
|
public static bool IsSubsequenceOfI(this string instance, string text)
|
|
{
|
|
return instance.IsSubsequenceOf(text, true);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether the string contains a valid float.
|
|
// </summary>
|
|
public static bool IsValidFloat(this string instance)
|
|
{
|
|
float f;
|
|
return float.TryParse(instance, out f);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether the string contains a valid color in HTML notation.
|
|
// </summary>
|
|
public static bool IsValidHtmlColor(this string instance)
|
|
{
|
|
return Color.HtmlIsValid(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether the string is a valid identifier. As is common in programming languages, a valid identifier may contain only letters, digits and underscores (_) and the first character may not be a digit.
|
|
// </summary>
|
|
public static bool IsValidIdentifier(this string instance)
|
|
{
|
|
int len = instance.Length;
|
|
|
|
if (len == 0)
|
|
return false;
|
|
|
|
for (int i = 0; i < len; i++)
|
|
{
|
|
if (i == 0)
|
|
{
|
|
if (instance[0] >= '0' && instance[0] <= '9')
|
|
return false; // Don't start with number plz
|
|
}
|
|
|
|
bool valid_char = (instance[i] >= '0' && instance[i] <= '9') || (instance[i] >= 'a' && instance[i] <= 'z') || (instance[i] >= 'A' && instance[i] <= 'Z') || instance[i] == '_';
|
|
|
|
if (!valid_char)
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether the string contains a valid integer.
|
|
// </summary>
|
|
public static bool IsValidInteger(this string instance)
|
|
{
|
|
int f;
|
|
return int.TryParse(instance, out f);
|
|
}
|
|
|
|
// <summary>
|
|
// Check whether the string contains a valid IP address.
|
|
// </summary>
|
|
public static bool IsValidIpAddress(this string instance)
|
|
{
|
|
string[] ip = instance.split(".");
|
|
|
|
if (ip.Length != 4)
|
|
return false;
|
|
|
|
for (int i = 0; i < ip.Length; i++)
|
|
{
|
|
string n = ip[i];
|
|
if (!n.IsValidInteger())
|
|
return false;
|
|
|
|
int val = n.to_int();
|
|
if (val < 0 || val > 255)
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
// <summary>
|
|
// Return a copy of the string with special characters escaped using the JSON standard.
|
|
// </summary>
|
|
public static string JsonEscape(this string instance)
|
|
{
|
|
StringBuilder sb = new StringBuilder(string.Copy(instance));
|
|
|
|
sb.Replace("\\", "\\\\");
|
|
sb.Replace("\b", "\\b");
|
|
sb.Replace("\f", "\\f");
|
|
sb.Replace("\n", "\\n");
|
|
sb.Replace("\r", "\\r");
|
|
sb.Replace("\t", "\\t");
|
|
sb.Replace("\v", "\\v");
|
|
sb.Replace("\"", "\\\"");
|
|
|
|
return sb.ToString();
|
|
}
|
|
|
|
// <summary>
|
|
// Return an amount of characters from the left of the string.
|
|
// </summary>
|
|
public static string Left(this string instance, int pos)
|
|
{
|
|
if (pos <= 0)
|
|
return string.Empty;
|
|
|
|
if (pos >= instance.Length)
|
|
return instance;
|
|
|
|
return instance.Substring(0, pos);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return the length of the string in characters.
|
|
/// </summary>
|
|
public static int Length(this string instance)
|
|
{
|
|
return instance.Length;
|
|
}
|
|
|
|
// <summary>
|
|
// Do a simple expression match, where '*' matches zero or more arbitrary characters and '?' matches any single character except '.'.
|
|
// </summary>
|
|
public static bool ExprMatch(this string instance, string expr, bool caseSensitive)
|
|
{
|
|
if (expr.Length == 0 || instance.Length == 0)
|
|
return false;
|
|
|
|
switch (expr[0])
|
|
{
|
|
case '\0':
|
|
return instance[0] == 0;
|
|
case '*':
|
|
return ExprMatch(expr + 1, instance, caseSensitive) || (instance[0] != 0 && ExprMatch(expr, instance + 1, caseSensitive));
|
|
case '?':
|
|
return instance[0] != 0 && instance[0] != '.' && ExprMatch(expr + 1, instance + 1, caseSensitive);
|
|
default:
|
|
return (caseSensitive ? instance[0] == expr[0] : char.ToUpper(instance[0]) == char.ToUpper(expr[0])) &&
|
|
ExprMatch(expr + 1, instance + 1, caseSensitive);
|
|
}
|
|
}
|
|
|
|
// <summary>
|
|
// Do a simple case sensitive expression match, using ? and * wildcards (see [method expr_match]).
|
|
// </summary>
|
|
public static bool Match(this string instance, string expr)
|
|
{
|
|
return instance.ExprMatch(expr, true);
|
|
}
|
|
|
|
// <summary>
|
|
// Do a simple case insensitive expression match, using ? and * wildcards (see [method expr_match]).
|
|
// </summary>
|
|
public static bool matchn(this string instance, string expr)
|
|
{
|
|
return instance.ExprMatch(expr, false);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the MD5 hash of the string as an array of bytes.
|
|
// </summary>
|
|
public static byte[] Md5Buffer(this string instance)
|
|
{
|
|
return NativeCalls.godot_icall_String_md5_buffer(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the MD5 hash of the string as a string.
|
|
// </summary>
|
|
public static string Md5Text(this string instance)
|
|
{
|
|
return NativeCalls.godot_icall_String_md5_text(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Perform a case-insensitive comparison to another string, return -1 if less, 0 if equal and +1 if greater.
|
|
// </summary>
|
|
public static int NocasecmpTo(this string instance, string to)
|
|
{
|
|
if (instance.Empty())
|
|
return to.Empty() ? 0 : -1;
|
|
|
|
if (to.Empty())
|
|
return 1;
|
|
|
|
int instance_idx = 0;
|
|
int to_idx = 0;
|
|
|
|
while (true)
|
|
{
|
|
if (to[to_idx] == 0 && instance[instance_idx] == 0)
|
|
return 0; // We're equal
|
|
else if (instance[instance_idx] == 0)
|
|
return -1; // If this is empty, and the other one is not, then we're less... I think?
|
|
else if (to[to_idx] == 0)
|
|
return 1; // Otherwise the other one is smaller..
|
|
else if (char.ToUpper(instance[instance_idx]) < char.ToUpper(to[to_idx])) // More than
|
|
return -1;
|
|
else if (char.ToUpper(instance[instance_idx]) > char.ToUpper(to[to_idx])) // Less than
|
|
return 1;
|
|
|
|
instance_idx++;
|
|
to_idx++;
|
|
}
|
|
}
|
|
|
|
// <summary>
|
|
// Return the character code at position [code]at[/code].
|
|
// </summary>
|
|
public static int OrdAt(this string instance, int at)
|
|
{
|
|
return instance[at];
|
|
}
|
|
|
|
// <summary>
|
|
// Format a number to have an exact number of [code]digits[/code] after the decimal point.
|
|
// </summary>
|
|
public static string PadDecimals(this string instance, int digits)
|
|
{
|
|
int c = instance.Find(".");
|
|
|
|
if (c == -1)
|
|
{
|
|
if (digits <= 0)
|
|
return instance;
|
|
|
|
instance += ".";
|
|
c = instance.Length - 1;
|
|
}
|
|
else
|
|
{
|
|
if (digits <= 0)
|
|
return instance.Substring(0, c);
|
|
}
|
|
|
|
if (instance.Length - (c + 1) > digits)
|
|
{
|
|
instance = instance.Substring(0, c + digits + 1);
|
|
}
|
|
else
|
|
{
|
|
while (instance.Length - (c + 1) < digits)
|
|
{
|
|
instance += "0";
|
|
}
|
|
}
|
|
|
|
return instance;
|
|
}
|
|
|
|
// <summary>
|
|
// Format a number to have an exact number of [code]digits[/code] before the decimal point.
|
|
// </summary>
|
|
public static string PadZeros(this string instance, int digits)
|
|
{
|
|
string s = instance;
|
|
int end = s.Find(".");
|
|
|
|
if (end == -1)
|
|
end = s.Length;
|
|
|
|
if (end == 0)
|
|
return s;
|
|
|
|
int begin = 0;
|
|
|
|
while (begin < end && (s[begin] < '0' || s[begin] > '9'))
|
|
{
|
|
begin++;
|
|
}
|
|
|
|
if (begin >= end)
|
|
return s;
|
|
|
|
while (end - begin < digits)
|
|
{
|
|
s = s.Insert(begin, "0");
|
|
end++;
|
|
}
|
|
|
|
return s;
|
|
}
|
|
|
|
// <summary>
|
|
// Decode a percent-encoded string. See [method percent_encode].
|
|
// </summary>
|
|
public static string PercentDecode(this string instance)
|
|
{
|
|
return Uri.UnescapeDataString(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Percent-encode a string. This is meant to encode parameters in a URL when sending a HTTP GET request and bodies of form-urlencoded POST request.
|
|
// </summary>
|
|
public static string PercentEncode(this string instance)
|
|
{
|
|
return Uri.EscapeDataString(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// If the string is a path, this concatenates [code]file[/code] at the end of the string as a subpath. E.g. [code]"this/is".plus_file("path") == "this/is/path"[/code].
|
|
// </summary>
|
|
public static string PlusFile(this string instance, string file)
|
|
{
|
|
if (instance.Length > 0 && instance[instance.Length - 1] == '/')
|
|
return instance + file;
|
|
else
|
|
return instance + "/" + file;
|
|
}
|
|
|
|
// <summary>
|
|
// Replace occurrences of a substring for different ones inside the string.
|
|
// </summary>
|
|
public static string Replace(this string instance, string what, string forwhat)
|
|
{
|
|
return instance.Replace(what, forwhat);
|
|
}
|
|
|
|
// <summary>
|
|
// Replace occurrences of a substring for different ones inside the string, but search case-insensitive.
|
|
// </summary>
|
|
public static string Replacen(this string instance, string what, string forwhat)
|
|
{
|
|
return Regex.Replace(instance, what, forwhat, RegexOptions.IgnoreCase);
|
|
}
|
|
|
|
// <summary>
|
|
// Perform a search for a substring, but start from the end of the string instead of the beginning.
|
|
// </summary>
|
|
public static int Rfind(this string instance, string what, int from = -1)
|
|
{
|
|
return NativeCalls.godot_icall_String_rfind(instance, what, from);
|
|
}
|
|
|
|
// <summary>
|
|
// Perform a search for a substring, but start from the end of the string instead of the beginning. Also search case-insensitive.
|
|
// </summary>
|
|
public static int Rfindn(this string instance, string what, int from = -1)
|
|
{
|
|
return NativeCalls.godot_icall_String_rfindn(instance, what, from);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the right side of the string from a given position.
|
|
// </summary>
|
|
public static string Right(this string instance, int pos)
|
|
{
|
|
if (pos >= instance.Length)
|
|
return instance;
|
|
|
|
if (pos < 0)
|
|
return string.Empty;
|
|
|
|
return instance.Substring(pos, (instance.Length - pos));
|
|
}
|
|
|
|
public static byte[] Sha256Buffer(this string instance)
|
|
{
|
|
return NativeCalls.godot_icall_String_sha256_buffer(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the SHA-256 hash of the string as a string.
|
|
// </summary>
|
|
public static string Sha256Text(this string instance)
|
|
{
|
|
return NativeCalls.godot_icall_String_sha256_text(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the similarity index of the text compared to this string. 1 means totally similar and 0 means totally dissimilar.
|
|
// </summary>
|
|
public static float Similarity(this string instance, string text)
|
|
{
|
|
if (instance == text)
|
|
{
|
|
// Equal strings are totally similar
|
|
return 1.0f;
|
|
}
|
|
if (instance.Length < 2 || text.Length < 2)
|
|
{
|
|
// No way to calculate similarity without a single bigram
|
|
return 0.0f;
|
|
}
|
|
|
|
string[] srcBigrams = instance.Bigrams();
|
|
string[] tgtBigrams = text.Bigrams();
|
|
|
|
int src_size = srcBigrams.Length;
|
|
int tgt_size = tgtBigrams.Length;
|
|
|
|
float sum = src_size + tgt_size;
|
|
float inter = 0;
|
|
|
|
for (int i = 0; i < src_size; i++)
|
|
{
|
|
for (int j = 0; j < tgt_size; j++)
|
|
{
|
|
if (srcBigrams[i] == tgtBigrams[j])
|
|
{
|
|
inter++;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
return (2.0f * inter) / sum;
|
|
}
|
|
|
|
// <summary>
|
|
// Split the string by a divisor string, return an array of the substrings. Example "One,Two,Three" will return ["One","Two","Three"] if split by ",".
|
|
// </summary>
|
|
public static string[] split(this string instance, string divisor, bool allow_empty = true)
|
|
{
|
|
return instance.Split(new string[] { divisor }, StringSplitOptions.RemoveEmptyEntries);
|
|
}
|
|
|
|
// <summary>
|
|
// Split the string in floats by using a divisor string, return an array of the substrings. Example "1,2.5,3" will return [1,2.5,3] if split by ",".
|
|
// </summary>
|
|
public static float[] split_floats(this string instance, string divisor, bool allow_empty = true)
|
|
{
|
|
List<float> ret = new List<float>();
|
|
int from = 0;
|
|
int len = instance.Length;
|
|
|
|
while (true)
|
|
{
|
|
int end = instance.Find(divisor, from);
|
|
if (end < 0)
|
|
end = len;
|
|
if (allow_empty || (end > from))
|
|
ret.Add(float.Parse(instance.Substring(from)));
|
|
if (end == len)
|
|
break;
|
|
|
|
from = end + divisor.Length;
|
|
}
|
|
|
|
return ret.ToArray();
|
|
}
|
|
|
|
private static readonly char[] non_printable = {
|
|
(char)00, (char)01, (char)02, (char)03, (char)04, (char)05,
|
|
(char)06, (char)07, (char)08, (char)09, (char)10, (char)11,
|
|
(char)12, (char)13, (char)14, (char)15, (char)16, (char)17,
|
|
(char)18, (char)19, (char)20, (char)21, (char)22, (char)23,
|
|
(char)24, (char)25, (char)26, (char)27, (char)28, (char)29,
|
|
(char)30, (char)31, (char)32
|
|
};
|
|
|
|
// <summary>
|
|
// Return a copy of the string stripped of any non-printable character at the beginning and the end. The optional arguments are used to toggle stripping on the left and right edges respectively.
|
|
// </summary>
|
|
public static string strip_edges(this string instance, bool left = true, bool right = true)
|
|
{
|
|
if (left)
|
|
{
|
|
if (right)
|
|
return instance.Trim(non_printable);
|
|
else
|
|
return instance.TrimStart(non_printable);
|
|
}
|
|
else
|
|
{
|
|
return instance.TrimEnd(non_printable);
|
|
}
|
|
}
|
|
|
|
// <summary>
|
|
// Return part of the string from the position [code]from[/code], with length [code]len[/code].
|
|
// </summary>
|
|
public static string substr(this string instance, int from, int len)
|
|
{
|
|
return instance.Substring(from, len);
|
|
}
|
|
|
|
// <summary>
|
|
// Convert the String (which is a character array) to PoolByteArray (which is an array of bytes). The conversion is speeded up in comparison to to_utf8() with the assumption that all the characters the String contains are only ASCII characters.
|
|
// </summary>
|
|
public static byte[] to_ascii(this string instance)
|
|
{
|
|
return Encoding.ASCII.GetBytes(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Convert a string, containing a decimal number, into a [code]float[/code].
|
|
// </summary>
|
|
public static float to_float(this string instance)
|
|
{
|
|
return float.Parse(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Convert a string, containing an integer number, into an [code]int[/code].
|
|
// </summary>
|
|
public static int to_int(this string instance)
|
|
{
|
|
return int.Parse(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return the string converted to lowercase.
|
|
// </summary>
|
|
public static string to_lower(this string instance)
|
|
{
|
|
return instance.ToLower();
|
|
}
|
|
|
|
// <summary>
|
|
// Return the string converted to uppercase.
|
|
// </summary>
|
|
public static string to_upper(this string instance)
|
|
{
|
|
return instance.ToUpper();
|
|
}
|
|
|
|
// <summary>
|
|
// Convert the String (which is an array of characters) to PoolByteArray (which is an array of bytes). The conversion is a bit slower than to_ascii(), but supports all UTF-8 characters. Therefore, you should prefer this function over to_ascii().
|
|
// </summary>
|
|
public static byte[] to_utf8(this string instance)
|
|
{
|
|
return Encoding.UTF8.GetBytes(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return a copy of the string with special characters escaped using the XML standard.
|
|
// </summary>
|
|
public static string xml_escape(this string instance)
|
|
{
|
|
return SecurityElement.Escape(instance);
|
|
}
|
|
|
|
// <summary>
|
|
// Return a copy of the string with escaped characters replaced by their meanings according to the XML standard.
|
|
// </summary>
|
|
public static string xml_unescape(this string instance)
|
|
{
|
|
return SecurityElement.FromString(instance).Text;
|
|
}
|
|
}
|
|
}
|