Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[release/6.0-preview7] Eliminate backtracking in the interpreter for patterns with .* #55960

Merged
merged 17 commits into from
Jul 20, 2021
Merged

[release/6.0-preview7] Eliminate backtracking in the interpreter for patterns with .* #55960

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
daa2813
First cut of look up table for speeding up Go()
Apr 13, 2021
4a917c2
More efficient .* in RegexInterpreter
Apr 13, 2021
4cf9ce9
sq
Apr 19, 2021
11d00c1
Get more debug info
Apr 20, 2021
a015470
Remove assert and add unit test
Apr 26, 2021
d01e32c
Potential unit test
May 26, 2021
7b15bac
temp
Jun 21, 2021
c086dc8
Fix a bug
Jun 22, 2021
c9621dd
sq
Jun 22, 2021
485aac5
Add extra protection to the backtracking optimization
Jun 22, 2021
dba0279
Add unit test
Jun 22, 2021
2059451
Revert
Jun 30, 2021
2766f30
RegexCompiler changes
Apr 23, 2021
f136bde
sq
Jul 14, 2021
1db6349
Remove debug unit tests
Jul 14, 2021
848aff0
Add a length to the AsSpan call
Jul 14, 2021
69a9c1d
Address RegexCompiler comments and add unit tests
Jul 16, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
140 changes: 123 additions & 17 deletions ...raries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexCompiler.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,7 @@ internal abstract class RegexCompiler
private static readonly MethodInfo s_spanSliceIntIntMethod = typeof(ReadOnlySpan<char>).GetMethod("Slice", new Type[] { typeof(int), typeof(int) })!;
private static readonly MethodInfo s_spanStartsWith = typeof(MemoryExtensions).GetMethod("StartsWith", new Type[] { typeof(ReadOnlySpan<>).MakeGenericType(Type.MakeGenericMethodParameter(0)), typeof(ReadOnlySpan<>).MakeGenericType(Type.MakeGenericMethodParameter(0)) })!.MakeGenericMethod(typeof(char));
private static readonly MethodInfo s_stringAsSpanMethod = typeof(MemoryExtensions).GetMethod("AsSpan", new Type[] { typeof(string) })!;
private static readonly MethodInfo s_spanLastIndexOfMethod = typeof(MemoryExtensions).GetMethod("LastIndexOf", new Type[] { typeof(ReadOnlySpan<>).MakeGenericType(Type.MakeGenericMethodParameter(0)), typeof(ReadOnlySpan<>).MakeGenericType(Type.MakeGenericMethodParameter(0)) })!.MakeGenericMethod(typeof(char));
private static readonly MethodInfo s_stringAsSpanIntIntMethod = typeof(MemoryExtensions).GetMethod("AsSpan", new Type[] { typeof(string), typeof(int), typeof(int) })!;
private static readonly MethodInfo s_stringGetCharsMethod = typeof(string).GetMethod("get_Chars", new Type[] { typeof(int) })!;
private static readonly MethodInfo s_stringIndexOfCharInt = typeof(string).GetMethod("IndexOf", new Type[] { typeof(char), typeof(int) })!;
Expand All @@ -90,6 +91,7 @@ internal abstract class RegexCompiler
private LocalBuilder? _runstackLocal;
private LocalBuilder? _textInfoLocal; // cached to avoid extraneous TLS hits from CurrentCulture and virtual calls to TextInfo
private LocalBuilder? _loopTimeoutCounterLocal; // timeout counter for setrep and setloop
private LocalBuilder? _maxBacktrackPositionLocal;

protected RegexOptions _options; // options
protected RegexCode? _code; // the RegexCode object
Expand Down Expand Up @@ -891,6 +893,8 @@ private void GenerateForwardSection()
Mvfldloc(s_runtrackposField, _runtrackposLocal!);
Mvfldloc(s_runstackField, _runstackLocal!);
Mvfldloc(s_runstackposField, _runstackposLocal!);
Ldc(-1);
Stloc(_maxBacktrackPositionLocal!);

_backpos = -1;

Expand Down Expand Up @@ -1705,7 +1709,7 @@ protected void GenerateFindFirstChar()
// if (!CharInClass(textSpan[i + 2], prefix[2], "...")) goto returnFalse;
// ...
Debug.Assert(charClassIndex == 0 || charClassIndex == 1);
for ( ; charClassIndex < _leadingCharClasses.Length; charClassIndex++)
for (; charClassIndex < _leadingCharClasses.Length; charClassIndex++)
{
Debug.Assert(needLoop);
Ldloca(textSpanLocal);
Expand Down Expand Up @@ -3310,6 +3314,7 @@ protected void GenerateGo()
}
_runtextbegLocal = DeclareInt32();
_runtextendLocal = DeclareInt32();
_maxBacktrackPositionLocal = DeclareInt32();

InitializeCultureForGoIfNecessary();

Expand Down Expand Up @@ -4258,7 +4263,61 @@ private void GenerateOneCode()
//: break Backward;
{
string str = _strings![Operand(0)];
Label multiCode = DefineLabel();
if (!IsRightToLeft())
{
// if (runtextend - runtextpos < c)
Ldloc(_runtextendLocal!);
Ldloc(_runtextposLocal!);
Sub();
Ldc(str.Length);
BgeFar(multiCode);
// if (!caseInsensitive && _maxBacktrackPosition != -1 && runtextpos > _maxBacktrackPosition)
if (!IsCaseInsensitive())
{
Ldloc(_maxBacktrackPositionLocal!);
Ldc(-1);
BeqFar(_backtrack);
Ldloc(_runtextposLocal!);
Ldloc(_maxBacktrackPositionLocal!);
BleFar(_backtrack);
// runtextpos = _maxBacktrackPosition;
Ldloc(_maxBacktrackPositionLocal!);
Stloc(_runtextposLocal!);
// ReadOnlySpan<char> runtextSpan = runtext.AsSpan(_maxBacktrackPosition, runtextend - _maxBacktractPosition);
Ldloc(_runtextLocal!);
Ldloc(_maxBacktrackPositionLocal!);
Ldloc(_runtextendLocal!);
Ldloc(_maxBacktrackPositionLocal!);
Sub();
using (RentedLocalBuilder runtextSpanLocal = RentReadOnlySpanCharLocal())
{
Call(s_stringAsSpanIntIntMethod);
Stloc(runtextSpanLocal);
using (RentedLocalBuilder lastIndexOfLocal = RentInt32Local())
{
// int lastIndexOf = runtextSpan.LastIndexOf(str.AsSpan());
Ldloc(runtextSpanLocal);
Ldstr(str);
Call(s_stringAsSpanMethod);
Call(s_spanLastIndexOfMethod);
Stloc(lastIndexOfLocal);
// if (lastIndexOf > -1)
Ldloc(lastIndexOfLocal);
Ldc(-1);
BleFar(_backtrack);
// runtextpos = lastIndexOf + _maxBacktrackPosition;
Ldloc(lastIndexOfLocal);
Ldloc(_maxBacktrackPositionLocal!);
Add();
Stloc(_runtextposLocal!);
BrFar(_backtrack);
}
}
}
}

MarkLabel(multiCode);
Ldc(str.Length);
Ldloc(_runtextendLocal!);
Ldloc(_runtextposLocal!);
Expand Down Expand Up @@ -4598,6 +4657,9 @@ private void GenerateOneCode()

using RentedLocalBuilder lenLocal = RentInt32Local();
using RentedLocalBuilder iLocal = RentInt32Local();
using RentedLocalBuilder tempMaxBacktrackPositionLocal = RentInt32Local();
Ldloc(_runtextposLocal!);
Stloc(tempMaxBacktrackPositionLocal);

if (!IsRightToLeft())
{
Expand Down Expand Up @@ -4847,6 +4909,12 @@ private void GenerateOneCode()
DoPush();

Track();
// if (_operator == RegexCode.Notoneloop) maxBacktrackPosition = tempMaxBacktrackPosition
if (_regexopcode == RegexCode.Notoneloop)
{
Ldloc(tempMaxBacktrackPositionLocal);
Stloc(_maxBacktrackPositionLocal!);
}
}
break;
}
Expand All @@ -4870,28 +4938,66 @@ private void GenerateOneCode()
//: if (i > 0)
//: Track(i - 1, pos - 1);
//: Advance(2);
PopTrack();
Stloc(_runtextposLocal!);
Label noBacktrackPositionBranch = DefineLabel();
PopTrack();
using (RentedLocalBuilder posLocal = RentInt32Local())
{
Stloc(posLocal);
Ldloc(posLocal);
Ldc(0);
BleFar(AdvanceLabel());
PopTrack();
using (RentedLocalBuilder iBacktrackLocal = RentInt32Local())
{
Stloc(iBacktrackLocal);
// if (!caseInsensitive && maxBacktrackPosition != -1 && pos > maxBacktrackPosition && runtextpos < pos && _operator == (RegexCode.Notoneloop | RegexCode.Back) && !_rightToLeft)
if (!IsCaseInsensitive() && _regexopcode == (RegexCode.Notoneloop | RegexCode.Back) && !IsRightToLeft())
{
Ldloc(_maxBacktrackPositionLocal!);
Ldc(-1);
Beq(noBacktrackPositionBranch);
Ldloc(posLocal);
Ldloc(_maxBacktrackPositionLocal!);
Ble(noBacktrackPositionBranch);
Ldloc(_runtextposLocal!);
Ldloc(posLocal);
Bge(noBacktrackPositionBranch);
/*
int difference = pos - maxBacktrackPosition;
pos = runtextpos;
i -= difference;
maxBacktrackPosition = -1;
*/
// int difference = pos - maxBacktrackPosition;
Ldloc(iBacktrackLocal);
Ldloc(posLocal);
Ldloc(_maxBacktrackPositionLocal!);
Sub();
Sub();
Stloc(iBacktrackLocal);
Ldloc(_runtextposLocal!);
Stloc(posLocal);
Ldc(-1);
Stloc(_maxBacktrackPositionLocal!);
}

MarkLabel(noBacktrackPositionBranch);
Ldloc(posLocal);
Stloc(_runtextposLocal!);
Ldloc(iBacktrackLocal);
Ldc(0);
BleFar(AdvanceLabel());
ReadyPushTrack();
Ldloc(iBacktrackLocal);
}
Ldc(1);
Sub();
DoPush();
ReadyPushTrack();
Ldloc(posLocal);
Ldloc(_runtextposLocal!);
Ldc(1);
Sub(IsRightToLeft());
DoPush();
Trackagain();
Advance();
}
Ldc(1);
Sub();
DoPush();
ReadyPushTrack();
Ldloc(_runtextposLocal!);
Ldc(1);
Sub(IsRightToLeft());
DoPush();
Trackagain();
Advance();
break;

case RegexCode.Onelazy:
Expand Down
27 changes: 27 additions & 0 deletions ...ies/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexInterpreter.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ internal sealed class RegexInterpreter : RegexRunner
private int _codepos;
private bool _rightToLeft;
private bool _caseInsensitive;
private int _maxBacktrackPosition = -1;

public RegexInterpreter(RegexCode code, CultureInfo culture)
{
Expand Down Expand Up @@ -223,6 +224,20 @@ private bool MatchString(string str)
{
if (runtextend - runtextpos < c)
{
// If MatchString was called after a greedy op such as a .*, we would have zipped runtextpos to the end without really examining any characters. Reset to maxBacktrackPos here as an optimization
if (!_caseInsensitive && _maxBacktrackPosition != -1 && runtextpos > _maxBacktrackPosition)
{
// If lastIndexOf is -1, we backtrack to the max extent possible.
runtextpos = _maxBacktrackPosition;
ReadOnlySpan<char> runtextSpan = runtext.AsSpan(_maxBacktrackPosition, runtextend - _maxBacktrackPosition);
int lastIndexOf = runtextSpan.LastIndexOf(str);
if (lastIndexOf > -1)
{
// Found the next position to match. Move runtextpos here
runtextpos = _maxBacktrackPosition + lastIndexOf;
}
}

return false;
}

Expand Down Expand Up @@ -1185,6 +1200,7 @@ protected override void Go()
int len = Math.Min(Operand(1), Forwardchars());
char ch = (char)Operand(0);
int i;
int tempMaxBacktrackPosition = runtextpos;

if (!_rightToLeft && !_caseInsensitive)
{
Expand Down Expand Up @@ -1217,6 +1233,7 @@ protected override void Go()
if (len > i && _operator == RegexCode.Notoneloop)
{
TrackPush(len - i - 1, runtextpos - Bump());
_maxBacktrackPosition = tempMaxBacktrackPosition;
}
}
advance = 2;
Expand Down Expand Up @@ -1261,6 +1278,16 @@ protected override void Go()
{
int i = TrackPeek();
int pos = TrackPeek(1);
if (!_caseInsensitive && _maxBacktrackPosition != -1 && pos > _maxBacktrackPosition && runtextpos < pos && _operator == (RegexCode.Notoneloop | RegexCode.Back) && !_rightToLeft)
{
// The Multi node has bumped us along already
int difference = pos - _maxBacktrackPosition;
Debug.Assert(difference > 0);
pos = runtextpos;
i -= difference;
// We shouldn't be backtracking anymore.
_maxBacktrackPosition = -1;
}
runtextpos = pos;
if (i > 0)
{
Expand Down
Loading