Streamline bool.TryParse/Format

[stephentoub]

    private char[] _true = new char[] { 'T', 'r', 'u', 'e' };
    private char[] _false = new char[] { 'F', 'a', 'l', 's', 'e' };
    private char[] _somethingElse = new char[] { 'O', 't', 'h', 'e', 'r' };

    [Benchmark] public bool ParseTrue() => bool.TryParse(_true, out _);
    [Benchmark] public bool ParseFalse() => bool.TryParse(_false, out _);
    [Benchmark] public bool ParseSomethingElse() => bool.TryParse(_somethingElse, out _);

    [Benchmark] public bool FormatTrue() => true.TryFormat(_true, out _);
    [Benchmark] public bool FormatFalse() => false.TryFormat(_false, out _);

Method	Toolchain	Mean	Error	StdDev	Ratio
ParseTrue	\main\corerun.exe	7.311 ns	0.0171 ns	0.0142 ns	1.00
ParseTrue	\pr\corerun.exe	1.149 ns	0.0243 ns	0.0203 ns	0.16
ParseFalse	\main\corerun.exe	8.749 ns	0.0569 ns	0.0504 ns	1.00
ParseFalse	\pr\corerun.exe	2.309 ns	0.0572 ns	0.0477 ns	0.26
ParseSomethingElse	\main\corerun.exe	17.638 ns	0.1478 ns	0.1382 ns	1.00
ParseSomethingElse	\pr\corerun.exe	8.219 ns	0.0511 ns	0.0453 ns	0.47
FormatTrue	\main\corerun.exe	3.117 ns	0.0245 ns	0.0204 ns	1.00
FormatTrue	\pr\corerun.exe	2.000 ns	0.0287 ns	0.0269 ns	0.64
FormatFalse	\main\corerun.exe	2.101 ns	0.0363 ns	0.0322 ns	1.00
FormatFalse	\pr\corerun.exe	2.065 ns	0.0416 ns	0.0389 ns	0.98

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

    private char[] _true = new char[] { 'T', 'r', 'u', 'e' };
    private char[] _false = new char[] { 'F', 'a', 'l', 's', 'e' };
    private char[] _somethingElse = new char[] { 'O', 't', 'h', 'e', 'r' };

    [Benchmark] public bool ParseTrue() => bool.TryParse(_true, out _);
    [Benchmark] public bool ParseFalse() => bool.TryParse(_false, out _);
    [Benchmark] public bool ParseSomethingElse() => bool.TryParse(_somethingElse, out _);

    [Benchmark] public bool FormatTrue() => true.TryFormat(_true, out _);
    [Benchmark] public bool FormatFalse() => false.TryFormat(_false, out _);

Method	Toolchain	Mean	Error	StdDev	Ratio
ParseTrue	\main\corerun.exe	7.311 ns	0.0171 ns	0.0142 ns	1.00
ParseTrue	\pr\corerun.exe	1.149 ns	0.0243 ns	0.0203 ns	0.16
ParseFalse	\main\corerun.exe	8.749 ns	0.0569 ns	0.0504 ns	1.00
ParseFalse	\pr\corerun.exe	2.309 ns	0.0572 ns	0.0477 ns	0.26
ParseSomethingElse	\main\corerun.exe	17.638 ns	0.1478 ns	0.1382 ns	1.00
ParseSomethingElse	\pr\corerun.exe	8.219 ns	0.0511 ns	0.0453 ns	0.47
FormatTrue	\main\corerun.exe	3.117 ns	0.0245 ns	0.0204 ns	1.00
FormatTrue	\pr\corerun.exe	2.000 ns	0.0287 ns	0.0269 ns	0.64
FormatFalse	\main\corerun.exe	2.101 ns	0.0363 ns	0.0322 ns	1.00
FormatFalse	\pr\corerun.exe	2.065 ns	0.0416 ns	0.0389 ns	0.98

Author:	stephentoub
Assignees:	stephentoub
Labels:	area-System.Runtime
Milestone:	-

[danmoseley]

Do you think that TrimWhiteSpaceAndNull can be improved also? It's on the "not a boolean" path.

[stephentoub]

Do you think that TrimWhiteSpaceAndNull can be improved also?

I don't see anything about the implementation that can be meaningfully improved without restoring to unsafe code, at least not until the JIT recognizes some sort of backward iteration pattern for bounds-check elimination.

We can, however, push that whole trimming path off into a separate function, making it a bit cheaper to invoke the main parsing routine, along with being able to add a few more checks to more quickly weed out bad inputs.

Method	Toolchain	value	Mean	Error	Ratio
TryParse	\main\corerun.exe	True	18.4991 ns	0.0338 ns	1.00
TryParse	\pr\corerun.exe	True	9.2014 ns	0.0485 ns	0.50
TryParse	\main\corerun.exe	0	15.6529 ns	0.0662 ns	1.00
TryParse	\pr\corerun.exe	0	1.2378 ns	0.0064 ns	0.08
TryParse	\main\corerun.exe	Bogus	17.7503 ns	0.0832 ns	1.00
TryParse	\pr\corerun.exe	Bogus	6.8170 ns	0.0342 ns	0.38
TryParse	\main\corerun.exe	False	7.8471 ns	0.0274 ns	1.00
TryParse	\pr\corerun.exe	False	1.4798 ns	0.0066 ns	0.19
TryParse	\main\corerun.exe	TRUE	8.1593 ns	0.0226 ns	1.00
TryParse	\pr\corerun.exe	TRUE	0.8097 ns	0.0060 ns	0.10
TryParse	\main\corerun.exe	false	7.9468 ns	0.0555 ns	1.00
TryParse	\pr\corerun.exe	false	1.4835 ns	0.0090 ns	0.19
TryParse	\main\corerun.exe	true	7.1882 ns	0.0645 ns	1.00
TryParse	\pr\corerun.exe	true	0.8110 ns	0.0068 ns	0.11

[danmoseley]

Assuming that once we get to trimming the result will usually be a failure, I wonder whether IndexOfAny(char[] { 'e', 'E' } might allow bailing out quickly if the input is large.

[stephentoub]

Assuming that once we get to trimming the result will usually be a failure, I wonder whether IndexOfAny(char[] { 'e', 'E' } might allow bailing out quickly if the input is large.

Have you seen any real-world cases of large inputs filled with whitespace and nulls being passed to bool.{Try}Parse?

[uweigand]

This commit introduced serious regressions on s390x, about 40 library test cases now fail.

The problem seems to be that this (and similar lines):

BinaryPrimitives.WriteUInt64LittleEndian(MemoryMarshal.AsBytes(destination), 0x65007500720054); // "True"

is incorrect on big-endian systems. The destination memory will be set up as the following bytes:

 54 00 72 00 75 00 65 00

while on a big-endian system, the string should actually be:

 00 54 00 72 00 75 00 65

I'm currently testing a fix.

[uweigand]

I'm currently testing a fix.

This is now #65078

[EgorBo]

Improvements: dotnet/perf-autofiling-issues#3529

[stephentoub]

Improvements

Always nice when what you saw locally is mirrored by the lab subsequently :)