Reduce overheads of Regex.Replace with a text replacement string #85564
Conversation
|
Tagging subscribers to this area: @dotnet/area-system-text-regularexpressions Issue DetailsIf the replacement string doesn't contain any backreferences, we can reduce the overheads involved in processing the replacement. Rather than storing a list of private static readonly Regex s_vowels = new Regex("[aeiou]", RegexOptions.Compiled);
private static readonly string s_input = new HttpClient().GetStringAsync(@"https://www.gutenberg.org/cache/epub/3200/pg3200.txt").Result;
[Benchmark]
public string RemoveVowels() => s_vowels.Replace(s_input, "");
|
src/libraries/System.Text.RegularExpressions/src/System/Text/PositionStringBuilder.cs
Outdated
Show resolved
Hide resolved
...raries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexReplacement.cs
Outdated
Show resolved
Hide resolved
If the replacement string doesn't contain any backreferences, we can reduce the overheads involved in processing the replacement. Rather than storing a list of `ReadOnlyMemory<char>` segments for every portion of the original string or replacement, we can just store a list of `(int offset, int count)` pairs; if the offset is non-negative it refers to the original string, and if it's negative, it means to use the whole replacement. We can also avoid evaluating the rules each time, and since we're not storing string references into the arrays, we don't need to clear the arrays before returning them to the pool.
1a4311d to
83189b1
Compare
|
Nice! LGTM |
If the replacement string doesn't contain any backreferences, we can reduce the overheads involved in processing the replacement. Rather than storing a list of
ReadOnlyMemory<char>segments for every portion of the original string or replacement, we can just store a list of(int offset, int count)pairs; if the offset is non-negative it refers to the original string, and if it's negative, it means to use the whole replacement. We can also avoid evaluating the rules each time, and since we're not storing string references into the arrays, we don't need to clear the arrays before returning them to the pool. This is all primarily helpful when there are lots of matches found in the input.