I have begun the long slow journey to redemption.

honey the codewitch

My Visual FA project was a bust, with broken benchmarks and too slow code. I know in theory a DFA regex can beat one that backtracks, but Microsoft has done a damned good job of optimizing their engine in recent years. My issue is they're using ReadOnlySpan over the input string so if I want to beat it I have to use that as well. That won't stream, but I can generate separate code for that. Here's my result so far Microsoft Regex compiled "Lexer": Found 170000 matches in 13ms FAStringRunner: Found 170000 matches in 10ms I can probably make it a little faster still. The FAStringRunner was created by hand, by me, not generated by a regular expression. It looks like this

q1:
if (((((ch >= 9)
&& (ch <= 10))
|| (ch == 13))
|| (ch == 32)))
{
++len;
if (position < span.Length - 1)
{
ch = span[unchecked((int)++position)];
}
else
{
ch = '\0';
}
goto q1;
}
return FAMatch.Create(2, span.Slice(unchecked((int)p), len).ToString(), p, l, c);
q2:
if (((ch >= 49)
&& (ch <= 57)))
{
++len;
if (position < span.Length - 1)
{
ch = span[unchecked((int)++position)];
}
else
{
ch = '\0';
}
goto q3;
}
goto errorout;

I wrote this by hand because I needed to test to see if I could match faster than Microsoft before I wrote any generator code to do so. I've got a lot of code to write. Edit: Okay they're really challenging me here:

Microsoft Regex compiled "Lexer": Found 1440000 matches in 91ms
FAStringRunner: Found 1430000 matches in 81ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 89ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 80ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 70ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 71ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 71ms
FAStringRunner: Found 1430000 matches in 81ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 71ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 73ms
FAStringRunner: Found 1430000 matches in 79ms
Microsoft Regex compiled "Lexer": Found 1440000 matches in 72ms
FAStringRunner: Found 1430000 matches in 80ms
Microsoft Regex compiled "Lexer": Found 1440000 m

englebart

is the difference between 1440000 and 1430000 significant? Or is that part of the reason for early prototyping? Almost good enough? “Perfect is the enemy of good enough”

honey the codewitch

no. i noticed it after the fact. it's a one off error in my benchmark code, iterated 10,000 times. :laugh:

Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix

Lost User

Doesn’t this piece of code have one set of brackets too many? Admittedly C type languages are not my forte so I’m probably wrong if (((ch >= 49) && (ch <= 57)))

honey the codewitch

The jump tables were copied from code I generated using an AST. It is valid code, even if there are spurious parentheses. They have no effect on the compilation or execution.

Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix

BernardIE5317

Greetings and Kind Regards Am I mistaken or is the 1st if logically identical to if (ch == 9 || ch == 10 || ch == 13 || ch == 32) ? Thank You Kindly

honey the codewitch

yes. The if conditions were generated by a tool, and it doesn't bother breaking up the range 9-10 into two single equality comparisons.

Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix

BernardIE5317

Greetings and Kind Regards It can also be if (c >= 49 && ch <= 57) as relational operators have higher precedence than logical.