Add Generic Enum.TryFormat() by heathbm · Pull Request #71590 · dotnet/runtime

heathbm · 2022-07-03T06:32:43Z

Partial implementation of #57881

Benchmarks

Code: https://gist.github.com/heathbm/59783b3d9e08d9556d7a86fa6ee2cd20

Before implementation:

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
FormatG	33.96 ns	0.722 ns	0.676 ns	33.80 ns	33.07 ns	35.16 ns	0.0019	24 B
FormatD	27.03 ns	0.149 ns	0.132 ns	27.00 ns	26.84 ns	27.29 ns	0.0019	24 B
FormatX	35.46 ns	0.737 ns	0.789 ns	35.42 ns	33.99 ns	36.74 ns	0.0050	64 B
FormatF	33.69 ns	0.270 ns	0.225 ns	33.73 ns	33.17 ns	33.95 ns	0.0018	24 B

After implementation:

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
FormatG	32.983 ns	0.2221 ns	0.1969 ns	32.980 ns	32.588 ns	33.40 ns	0.0018	24 B
TryFormatG	14.332 ns	0.1794 ns	0.1678 ns	14.412 ns	13.943 ns	14.51 ns	-	-
FormatD	26.751 ns	0.8298 ns	0.9556 ns	26.348 ns	25.166 ns	28.47 ns	0.0019	24 B
TryFormatD	9.906 ns	0.1853 ns	0.1734 ns	9.862 ns	9.747 ns	10.35 ns	-	-
FormatX	34.408 ns	0.2204 ns	0.1954 ns	34.430 ns	33.866 ns	34.63 ns	0.0050	64 B
TryFormatX	17.993 ns	0.0465 ns	0.0435 ns	17.998 ns	17.913 ns	18.08 ns	-	-
FormatF	34.604 ns	0.2278 ns	0.2131 ns	34.486 ns	34.385 ns	34.99 ns	0.0018	24 B
TryFormatF	17.251 ns	0.0471 ns	0.0368 ns	17.247 ns	17.172 ns	17.31 ns	-	-

No visible regression.

Profiling:

Profiling code: https://gist.github.com/heathbm/079c4bea1dc541ffc12c91f283892875
Tested with:

Visual Studio Performance Profiler (.NET Object Allocation Tracking)
DotMemory (Full, Entire process tree)

Only allocations appear in the first run from:

System.Enum.GetEnumInfo
System.RuntimeType.GetTypeCodeImpl
I believe there is nothing I can do to prevent these allocations on the first run as they generate cache.

ghost · 2022-07-03T06:32:52Z

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, to please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

ghost · 2022-07-03T06:32:53Z

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 · 2022-07-03T13:50:32Z

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

Issue Details

Partial implementation of #57881

Benchmarks

usingBenchmarkDotNet.Attributes;usingMicroBenchmarks;namespaceSystem.Tests{[MemoryDiagnoser][BenchmarkCategory(Categories.Libraries)]publicclassPerf_Enum{publicenumColors{Red=0x1,Orange=0x2,Yellow=0x4,Green=0x8,Blue=0x10}[Benchmark]publicvoidFormat(){vardestination=Enum.Format(typeof(Colors),Colors.Green,"F");}[Benchmark]publicvoidTryFormat(){Span<char>destination=stackallocchar[5];Enum.TryFormat(Colors.Green,destination,out_,"F");}}}

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
Format	31.90 ns	0.359 ns	0.318 ns	31.94 ns	31.04 ns	32.28 ns	0.0019	24 B
TryFormat	14.96 ns	0.053 ns	0.047 ns	14.95 ns	14.88 ns	15.05 ns	-	-

Profiling:

Profiling code: https://gist.github.com/heathbm/079c4bea1dc541ffc12c91f283892875
Tested with:

Visual Studio Performance Profiler (.NET Object Allocation Tracking)
DotMemory (Full, Entire process tree)

Only allocations appear in the first run from:

System.Enum.GetEnumInfo
System.RuntimeType.GetTypeCodeImpl
I believe there is nothing I can do to prevent these allocations on the first run as they generate cache.

Author:	heathbm
Assignees:	-
Labels:	`area-System.Runtime`, `new-api-needs-documentation`, `community-contribution`
Milestone:	-

deeprobin

Looks overall good to me

jkotas · 2022-07-09T05:03:28Z

The Enum tests are failing.

heathbm · 2022-07-13T15:58:03Z

The Enum tests are failing.

@jkotas I have removed a Debug.Assert that was unreliable since that assert can now deal with externally provided spans, that can be larger than what was originally expected. I did not see any way to elegantly satisfy the assert without additional operations just to make the assert pass.

danmoseley · 2022-09-06T02:19:39Z

@jkotas was your feedback addressed?

jkotas · 2022-09-06T02:53:08Z

It was addressed, thought there is still pretty significant code-bloat per instantiations that could be avoided by more significant refactoring.

Are there places in this repo where this method can be used?

stephentoub · 2022-09-27T15:01:59Z

Are there places in this repo where this method can be used?

My expectation is we'll use it at a minimum in our interpolated string handlers, to avoid the boxing and string allocations that come from falling back to using ToString today. We can start by calling this explicitly after doing some type testing (e.g. typeof(T).IsEnum when it's an intrinsic), though with the constraints on the public signature we'll probably need to move the logic out into an unconstrained helper. Then that special-casing can be removed if/when Enum implements ISpanFormattable and the JIT special-cases it, though at that point I expect this implementation would be used by that ISpanFormattable implementation.

dakersnar · 2022-10-10T18:39:47Z

@jkotas I added the "needs-author-action" label until your code-bloat comment is addressed.

ghost · 2022-10-24T21:02:11Z

This pull request has been automatically marked no-recent-activity because it has not had any activity for 14 days. It will be closed if no further activity occurs within 14 more days. Any new comment (by anyone, not necessarily the author) will remove no-recent-activity.

ghost · 2022-11-08T00:02:59Z

This pull request will now be closed since it had been marked no-recent-activity but received no further activity in the past 14 days. It is still possible to reopen or comment on the pull request, but please note that it will be locked if it remains inactive for another 30 days.

dakersnar · 2022-11-08T16:38:30Z

Discussed this with @jkotas, paraphrasing our conversation here to let others weigh in. Reopening for now.

To address his code bloat comment, we would need something like #76398. He does not have a strong opinion on whether or not we should merge this in its current state, but we will likely have some perf regressions in existing APIs if we do.

stephentoub · 2022-11-08T16:40:42Z

we will likely have some perf regressions in existing APIs if we do

What kind of regressions, and do you have an example of where we'd expect this?

jkotas · 2022-11-08T16:51:07Z

What kind of regressions, and do you have an example of where we'd expect this?

There seems to be some extra work done on paths through the existing APIs. We should run enum microbenchmarks to see whether it is going to show up.

stephentoub · 2022-11-08T17:20:34Z

There seems to be some extra work done on paths through the existing APIs.

I see. It seems like that was done in response to feedback that there was logical code duplication, and the changes are due to consolidating it?

It'd be really nice to get TryFormat added. Would you prefer it be entirely new code, without touching existing code paths, and then subsequently we can look at how to unify as much as possible without regression?

We should run enum microbenchmarks to see whether it is going to show up.

👍

jkotas · 2022-11-08T17:29:25Z

It'd be really nice to get TryFormat added. Would you prefer it be entirely new code, without touching existing code paths, and then subsequently we can look at how to unify as much as possible without regression?

Ok with me. I am not happy about the amount of code bloat generated for each use of the generic enums parse and format methods. It is a pre-existing problem.

stephentoub · 2022-11-12T16:58:13Z

I've assigned this to myself to land it.

- Add TryFormat XML docs - Make TryFormat's format parameter optional and allow an empty specifier - Make it an exception to specify invalid format specifier in TryFormat - Streamline TryFormat's default specifier case - Add missing cases where feasible for char/bool/nint/nuint underlying types - Reduce duplicate branches in hex formatting methods where e.g. int/uint cases can be shared - Refactor method implementations to better share return blocks - Add AggressiveInlining to places where call sites are finite / overhead is measurable - Remove checked multiplication from length calculation - Remove some branching from code to find name(s) in flags names - Revise multi-flag name writing to reduce bounds checking - Change all format specifier switches to use ASCII casing trick - Outline exception creation from generic methods to reduce specialized code bloat - Rename helper methods to make their functionality clearer - Tweak formatting for consistency within the file

- Use Enum.GetUnderlyingType instead of Type.GetTypeCode so that the intrinsic enables optimization at the call site - Add a generic cache with readonly statics that enables more generic specialization and branch elimination at call sites - Add more missing nint/nuint/char/etc. cases - Add a generic GetEnumName that can trim away half based on ValuesAreSequentialFromZero, that can get a hardcoded names address (once the frozen work is complete), and that can optimize out a branch in the inlined FindDefinedIndex - Remove a bounds check from GetEnumName on the find path - Expose an internal TryFormatUnconstrained without Enum/struct constraints for corelib interpolated string handlers to use

stephentoub · 2022-11-15T05:17:23Z

I've made a bunch of changes here, in additional commits (I squashed all the existing commits down to one). As part of validating it, I fixed #76157. This also contributes to #76398, the main missing piece there being different types for the values array. And I realize the GenericEnumInfo<TEnum> cache might be controversial, but while it adds a new generically-specialized class per enum, it also reduces the amount of code in various generic methods by eliminating unnecessary code paths... @jkotas?

I have a change to push up to dotnet/performance to add more tests. The existing non-generic methods effectively remain unchanged throughput-wise, subject to noise, or get a bit better. The tests:

[Benchmark][Arguments(Colors.Yellow)][Arguments(Colors.Yellow|Colors.Blue)][Arguments(Colors.Red|Colors.Orange|Colors.Yellow|Colors.Green|Colors.Blue)][Arguments(Colors.Yellow|(Colors)0x20)][Arguments(0x20)]publicstringEnumToString(Colorsvalue)=>value.ToString();[Benchmark][Arguments(SearchOption.TopDirectoryOnly)][Arguments(SearchOption.AllDirectories)][Arguments((SearchOption)(-1))]publicstringEnumToString_SmallNonFlagsEnum(SearchOptionvalue)=>value.ToString();[Benchmark][Arguments(UnicodeCategory.UppercaseLetter)][Arguments(UnicodeCategory.Control)][Arguments(UnicodeCategory.Format)][Arguments(UnicodeCategory.OtherNotAssigned)][Arguments((UnicodeCategory)42)]publicstringEnumToString_LargeNonFlagsEnum(UnicodeCategoryvalue)=>value.ToString();[Benchmark][Arguments(DayOfWeek.Sunday,"")][Arguments(DayOfWeek.Monday,"g")][Arguments(DayOfWeek.Tuesday,"d")][Arguments(DayOfWeek.Wednesday,"x")][Arguments(DayOfWeek.Thursday,"f")][Arguments(DayOfWeek.Friday,"X")][Arguments(DayOfWeek.Saturday,"D")][Arguments((DayOfWeek)7,"G")][Arguments((DayOfWeek)8,"F")]publicstringEnumToStringWithFormat(DayOfWeekvalue,stringformat)=>value.ToString(format);

on my machine result in:

Method	Job	Toolchain	value	format	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
EnumToString_SmallNonFlagsEnum	Job-EUILUD	\main\corerun.exe	-1	?	30.90 ns	0.842 ns	0.935 ns	30.85 ns	29.520 ns	32.73 ns	1.00	0.00	0.0089	56 B	1.00
EnumToString_SmallNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	-1	?	30.17 ns	0.354 ns	0.331 ns	30.10 ns	29.722 ns	30.83 ns	0.98	0.04	0.0089	56 B	1.00

EnumToString	Job-EUILUD	\main\corerun.exe	32	?	30.64 ns	0.626 ns	0.586 ns	30.52 ns	29.962 ns	31.66 ns	1.00	0.00	0.0089	56 B	1.00
EnumToString	Job-HGADFP	\pr\corerun.exe	32	?	29.51 ns	0.641 ns	0.569 ns	29.41 ns	28.576 ns	30.68 ns	0.96	0.02	0.0089	56 B	1.00

EnumToString	Job-EUILUD	\main\corerun.exe	36	?	31.45 ns	0.756 ns	0.841 ns	31.40 ns	30.093 ns	32.96 ns	1.00	0.00	0.0088	56 B	1.00
EnumToString	Job-HGADFP	\pr\corerun.exe	36	?	30.92 ns	0.697 ns	0.802 ns	30.94 ns	29.512 ns	32.16 ns	0.98	0.03	0.0089	56 B	1.00

EnumToString_LargeNonFlagsEnum	Job-EUILUD	\main\corerun.exe	42	?	23.27 ns	0.948 ns	1.054 ns	23.21 ns	21.876 ns	25.69 ns	1.00	0.00	0.0089	56 B	1.00
EnumToString_LargeNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	42	?	22.20 ns	0.886 ns	1.020 ns	21.97 ns	20.722 ns	24.58 ns	0.95	0.06	0.0088	56 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Sunday		12.65 ns	0.597 ns	0.688 ns	12.45 ns	11.782 ns	14.28 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Sunday		13.14 ns	0.573 ns	0.636 ns	13.20 ns	12.169 ns	14.19 ns	1.05	0.08	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Monday	g	13.78 ns	0.327 ns	0.321 ns	13.80 ns	13.160 ns	14.46 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Monday	g	14.35 ns	0.629 ns	0.725 ns	14.15 ns	13.465 ns	15.91 ns	1.05	0.07	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Tuesday	d	10.27 ns	0.324 ns	0.373 ns	10.25 ns	9.808 ns	11.09 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Tuesday	d	10.67 ns	0.500 ns	0.535 ns	10.67 ns	10.003 ns	11.55 ns	1.04	0.06	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Wednesday	x	24.40 ns	0.604 ns	0.696 ns	24.28 ns	23.485 ns	25.72 ns	1.00	0.00	0.0101	64 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Wednesday	x	23.63 ns	0.895 ns	0.994 ns	23.24 ns	22.779 ns	26.06 ns	0.97	0.05	0.0101	64 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Thursday	f	19.31 ns	0.663 ns	0.737 ns	19.25 ns	18.538 ns	20.99 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Thursday	f	18.21 ns	0.521 ns	0.600 ns	18.01 ns	17.386 ns	19.39 ns	0.94	0.04	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Friday	X	24.42 ns	0.529 ns	0.544 ns	24.30 ns	23.235 ns	25.43 ns	1.00	0.00	0.0101	64 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Friday	X	24.39 ns	1.094 ns	1.260 ns	24.11 ns	22.857 ns	27.07 ns	1.01	0.06	0.0102	64 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	Saturday	D	10.26 ns	0.349 ns	0.388 ns	10.17 ns	9.736 ns	10.99 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	Saturday	D	10.43 ns	0.236 ns	0.197 ns	10.43 ns	10.102 ns	10.80 ns	1.03	0.04	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	7	G	16.72 ns	0.866 ns	0.998 ns	16.46 ns	15.390 ns	19.04 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	7	G	16.91 ns	0.688 ns	0.792 ns	16.72 ns	15.934 ns	18.58 ns	1.02	0.09	0.0038	24 B	1.00

EnumToStringWithFormat	Job-EUILUD	\main\corerun.exe	8	F	27.46 ns	1.296 ns	1.492 ns	27.25 ns	25.967 ns	30.82 ns	1.00	0.00	0.0038	24 B	1.00
EnumToStringWithFormat	Job-HGADFP	\pr\corerun.exe	8	F	24.61 ns	0.537 ns	0.528 ns	24.78 ns	23.632 ns	25.43 ns	0.89	0.05	0.0038	24 B	1.00

EnumToString	Job-EUILUD	\main\corerun.exe	Red, Orange, Yellow, Green, Blue	?	56.09 ns	2.247 ns	2.497 ns	55.84 ns	52.697 ns	62.04 ns	1.00	0.00	0.0176	112 B	1.00
EnumToString	Job-HGADFP	\pr\corerun.exe	Red, Orange, Yellow, Green, Blue	?	50.89 ns	1.480 ns	1.704 ns	51.20 ns	48.498 ns	54.65 ns	0.91	0.05	0.0177	112 B	1.00

EnumToString_LargeNonFlagsEnum	Job-EUILUD	\main\corerun.exe	UppercaseLetter	?	12.61 ns	0.408 ns	0.470 ns	12.54 ns	11.753 ns	13.31 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_LargeNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	UppercaseLetter	?	12.07 ns	0.296 ns	0.304 ns	12.04 ns	11.664 ns	12.77 ns	0.96	0.05	0.0038	24 B	1.00

EnumToString_LargeNonFlagsEnum	Job-EUILUD	\main\corerun.exe	Control	?	11.62 ns	0.211 ns	0.176 ns	11.60 ns	11.412 ns	12.05 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_LargeNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	Control	?	11.09 ns	0.185 ns	0.164 ns	11.08 ns	10.898 ns	11.41 ns	0.95	0.02	0.0038	24 B	1.00

EnumToString_LargeNonFlagsEnum	Job-EUILUD	\main\corerun.exe	Format	?	12.32 ns	0.266 ns	0.236 ns	12.25 ns	12.059 ns	12.80 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_LargeNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	Format	?	12.32 ns	0.360 ns	0.400 ns	12.24 ns	11.825 ns	13.30 ns	1.00	0.04	0.0038	24 B	1.00

EnumToString_LargeNonFlagsEnum	Job-EUILUD	\main\corerun.exe	OtherNotAssigned	?	12.44 ns	0.429 ns	0.494 ns	12.38 ns	11.765 ns	13.54 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_LargeNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	OtherNotAssigned	?	12.19 ns	0.504 ns	0.580 ns	12.10 ns	11.369 ns	13.23 ns	0.98	0.06	0.0038	24 B	1.00

EnumToString_SmallNonFlagsEnum	Job-EUILUD	\main\corerun.exe	TopDirectoryOnly	?	12.71 ns	0.590 ns	0.679 ns	12.50 ns	11.860 ns	13.87 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_SmallNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	TopDirectoryOnly	?	11.58 ns	0.292 ns	0.312 ns	11.56 ns	11.258 ns	12.37 ns	0.92	0.05	0.0038	24 B	1.00

EnumToString_SmallNonFlagsEnum	Job-EUILUD	\main\corerun.exe	AllDirectories	?	12.41 ns	0.389 ns	0.448 ns	12.23 ns	11.899 ns	13.42 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString_SmallNonFlagsEnum	Job-HGADFP	\pr\corerun.exe	AllDirectories	?	11.72 ns	0.272 ns	0.212 ns	11.74 ns	11.248 ns	11.98 ns	0.95	0.03	0.0038	24 B	1.00

EnumToString	Job-EUILUD	\main\corerun.exe	Yellow	?	16.38 ns	0.561 ns	0.623 ns	16.10 ns	15.764 ns	17.78 ns	1.00	0.00	0.0038	24 B	1.00
EnumToString	Job-HGADFP	\pr\corerun.exe	Yellow	?	15.57 ns	0.349 ns	0.292 ns	15.61 ns	14.937 ns	16.00 ns	0.96	0.04	0.0038	24 B	1.00

EnumToString	Job-EUILUD	\main\corerun.exe	Yellow, Blue	?	36.05 ns	1.333 ns	1.535 ns	35.79 ns	33.932 ns	39.00 ns	1.00	0.00	0.0113	72 B	1.00
EnumToString	Job-HGADFP	\pr\corerun.exe	Yellow, Blue	?	35.32 ns	1.559 ns	1.795 ns	35.30 ns	32.570 ns	38.40 ns	0.98	0.04	0.0114	72 B	1.00

For TryFormat perf, the impact is directly visible via interpolated strings, which now use Enum.TryFormat:

[Benchmark][Arguments(Colors.Red|Colors.Green)][Arguments(0x20)]publicboolInterpolateEnum(Colorsvalue)=>MemoryExtensions.TryWrite(s_scratch,$"{value}{value:g}{value:d}{value:x}{value:f}",out_);

results in:

Method	Job	Toolchain	value	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
InterpolateEnum	Job-GEPFJR	\main\corerun.exe	32	158.75 ns	3.920 ns	4.357 ns	156.08 ns	155.11 ns	171.78 ns	1.00	0.00	0.0458	288 B	1.00
InterpolateEnum	Job-ABMNOH	\pr\corerun.exe	32	96.27 ns	1.065 ns	0.889 ns	96.10 ns	95.37 ns	97.89 ns	0.60	0.02	-	-	0.00

InterpolateEnum	Job-GEPFJR	\main\corerun.exe	Red, Green	173.81 ns	4.712 ns	5.238 ns	173.31 ns	164.44 ns	182.48 ns	1.00	0.00	0.0480	304 B	1.00
InterpolateEnum	Job-ABMNOH	\pr\corerun.exe	Red, Green	113.90 ns	3.000 ns	3.334 ns	113.61 ns	108.63 ns	120.92 ns	0.66	0.03	-	-	0.00

Other generic Enum methods also improve, e.g.

privateColors_colorValue=Colors.Blue;privateDayOfWeek_dayOfWeekValue=DayOfWeek.Saturday;[Benchmark]publicboolIsDefined()=>Enum.IsDefined(_colorValue);[Benchmark]publicboolIsDefined_NonFlags()=>Enum.IsDefined(_dayOfWeekValue);[Benchmark]publicstringGetName()=>Enum.GetName(_colorValue);[Benchmark]publicstringGetName_NonFlags()=>Enum.GetName(_dayOfWeekValue);[Benchmark]publicstring[]GetNames()=>Enum.GetNames<Colors>();

results in:

Method	Job	Toolchain	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
IsDefined	Job-UKYMDD	\main\corerun.exe	6.572 ns	0.1094 ns	0.0854 ns	6.559 ns	6.471 ns	6.715 ns	1.00	0.00	-	-	NA
IsDefined	Job-RHTFZD	\pr\corerun.exe	3.830 ns	0.0989 ns	0.1100 ns	3.773 ns	3.728 ns	4.091 ns	0.58	0.02	-	-	NA

IsDefined_NonFlags	Job-UKYMDD	\main\corerun.exe	3.751 ns	0.0670 ns	0.0627 ns	3.762 ns	3.652 ns	3.855 ns	1.00	0.00	-	-	NA
IsDefined_NonFlags	Job-RHTFZD	\pr\corerun.exe	1.407 ns	0.0512 ns	0.0479 ns	1.396 ns	1.357 ns	1.506 ns	0.38	0.01	-	-	NA

GetName	Job-UKYMDD	\main\corerun.exe	9.214 ns	0.5445 ns	0.6271 ns	9.119 ns	8.359 ns	10.771 ns	1.00	0.00	-	-	NA
GetName	Job-RHTFZD	\pr\corerun.exe	5.003 ns	0.1648 ns	0.1832 ns	5.017 ns	4.672 ns	5.262 ns	0.54	0.04	-	-	NA

GetName_NonFlags	Job-UKYMDD	\main\corerun.exe	4.945 ns	0.2827 ns	0.3255 ns	4.914 ns	4.603 ns	5.786 ns	1.00	0.00	-	-	NA
GetName_NonFlags	Job-RHTFZD	\pr\corerun.exe	2.499 ns	0.1435 ns	0.1653 ns	2.509 ns	2.311 ns	2.816 ns	0.51	0.05	-	-	NA

GetNames	Job-UKYMDD	\main\corerun.exe	15.713 ns	0.7786 ns	0.8966 ns	15.475 ns	14.753 ns	17.493 ns	1.00	0.00	0.0101	64 B	1.00
GetNames	Job-RHTFZD	\pr\corerun.exe	14.534 ns	0.3457 ns	0.3065 ns	14.432 ns	14.123 ns	15.052 ns	0.93	0.07	0.0101	64 B	1.00

drieseng · 2022-11-15T20:30:40Z

 [Serializable]
- [System.Runtime.CompilerServices.TypeForwardedFrom("mscorlib, Version=4.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089")]
+ [TypeForwardedFrom("mscorlib, Version=4.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089")]
 public abstract partial class Enum : ValueType, IComparable, IFormattable, IConvertible


I'm sure this is deliberate, but why not implement ISpanFormattable?

Every call to TryFormat would box the enum, just as every call to ToString() today boxes the enum. To avoid that would require non-trivial work in the JIT to rewrite calls from the instance method to something else. Unlike GetHashCode and HasFlags, which are special-cased, the implementation of formatting is complicated and not something we'd want to replicate in the JIT itself, so most likely we'd need a static generic TryFormat anyway the JIT could rewrite the calls to target. And we don't want to expose an ISpanFormattable implementation that's going to implicitly have such expense unless/until it can be eliminated.

@stephentoub thanks for taking the time to respond, I really appreciate it! Why can't you use constrained calls with enums to avoid boxing (like what's done here, for example)?
PS. Last question, I promise :p

No. Try just running:
for(inti=0;i<1000;i++)DayOfWeek.Monday.ToString();
under an allocation profiler. You'll see it allocates 1000 enum objects. The implementation of ToString itself is defined on the System.Enum reference type.

jkotas · 2022-11-16T03:50:11Z

it also reduces the amount of code in various generic methods by eliminating unnecessary code paths..

There is still a very non-trivial code duplication per enum specialization. Here are some numbers for a simple enum with 3 values. The code that I have used to get these numbers is here: https://gist.github.com/jkotas/62f1680635e230a44bad2027ace5effa

TryFormat specialization in this PR costs about 7.5kB of workingset footprint per enum. It feels like a lot. For comparison, non-generic Enum.ToString costs about 550 bytes per enum. A simple source-generated ToString method costs 680 bytes per enum (this is tier 0 cost, it is down to 439 bytes with tiered compilation disabled).

To get the per-enum instantiation overhead down, I think the implementation would need to look like this:

// This is the only method that is duplicated per enum instantiation[MethodImpl(MethodImplOptions.AggressiveInlining)]staticboolTryFormat<TEnum>(TEnumvalue,Span<char>destination,outintcharsWritten,[StringSyntax(StringSyntaxAttribute.EnumFormat)]ReadOnlySpan<char>format=default)whereTEnum:struct,Enum{EnumInfoenumInfo=GetEnumInfo(typeof(TEnum));// We may want to tweak the JIT so that it dead-code eliminates unreachable branches even in tier 0TypeunderlyingType=typeof(TEnum).GetEnumUnderlyingType();if(underlyingType==typeof(int))returnTryFormatPrimitive<int>(Unsafe.As<TEnum,int>(refvalue),destination,charsWritten,format,enumInfo);if(underlyingType==typeof(long))returnTryFormatPrimitive<long>(Unsafe.As<TEnum,long>(refvalue),destination,charsWritten,format,enumInfo);
...}// This method is only instantiated per finite set of primitive types.staticboolTryFormatPrimitive<TUnderlyingType>(TUnderlyingTypevalue,Span<char>destination,outintcharsWritten,ReadOnlySpan<char>format,EnumInfoenumInfo)whereTUnderlyingType:ISpanFormattable{
...}

stephentoub · 2022-11-16T15:31:23Z

There is still a very non-trivial code duplication per enum specialization

Which is why this doesn't close #76398.

But, fine. I'll close this and iterate on it some more locally rather than doing it subsequently. @heathbm, I'll preserve your original commit in any subsequent PR I put up.

stephentoub · 2022-11-16T15:38:02Z

// We may want to tweak the JIT so that it dead-code eliminates unreachable branches even in tier 0

@EgorBo

EgorBo · 2022-11-16T17:44:29Z

Reference in new

Thanks, will add it to the list of potential improvements for tier0

ghost added the new-api-needs-documentation label Jul 3, 2022

ghost added the community-contribution Indicates that the PR has been added by a community member label Jul 3, 2022

am11 added the area-System.Runtime label Jul 3, 2022

MichalPetryka mentioned this pull request Jul 3, 2022
Optimize Enum.GetUnderlyingType with a known type to a constant #70483
Closed

jkotas reviewed Jul 3, 2022
View reviewed changes

Comment threadsrc/libraries/System.Private.CoreLib/src/System/Enum.cs Outdated

deeprobin approved these changes Jul 3, 2022
View reviewed changes

deeprobin approved these changes Jul 8, 2022
View reviewed changes

danmoseley assigned jkotas Aug 19, 2022

stephentoub mentioned this pull request Sep 26, 2022
Special-case enums in interpolated string handlers #76157
Closed

dakersnar added the needs-author-action An issue or pull request that requires more info or actions from the author. label Oct 10, 2022

ghost added the no-recent-activity label Oct 24, 2022

ghost closed this Nov 8, 2022

dakersnar reopened this Nov 8, 2022

dakersnar added blocked Issue/PR is blocked on something - see comments and removed no-recent-activity labels Nov 8, 2022

build-analysisBot mentioned this pull request Nov 8, 2022
Tracking issue for CI build timeouts #76454
Closed

stephentoub self-assigned this Nov 11, 2022

Add generic Enum.TryFormat method
b5249a3

stephentoub removed blocked Issue/PR is blocked on something - see comments needs-author-action An issue or pull request that requires more info or actions from the author. labels Nov 15, 2022

stephentoub force-pushed the implement-enum-tryformat branch from bc25940 to 8299b28Compare November 15, 2022 04:25

stephentoub added 5 commits November 15, 2022 00:03

Use Enum.TryFormat in interpolated string handlers
89f84be

Use generic Enum.IsDefined in a few more places
7b05542

Fix Enum tests
f993036

stephentoub force-pushed the implement-enum-tryformat branch from 8299b28 to f993036Compare November 15, 2022 05:03

drieseng reviewed Nov 15, 2022
View reviewed changes

stephentoub closed this Nov 16, 2022

stephentoub mentioned this pull request Nov 18, 2022
Rewrite Enum and add {ISpanFormattable}.TryFormat #78580
Merged

ghost locked as resolved and limited conversation to collaborators Dec 17, 2022

Uh oh!

Conversation

heathbm commented Jul 3, 2022• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmarks

Profiling:

Uh oh!

ghost commented Jul 3, 2022

Uh oh!

ghost commented Jul 3, 2022

Uh oh!

ghost commented Jul 3, 2022

Benchmarks

Profiling:

Uh oh!

Uh oh!

deeprobin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jkotas commented Jul 9, 2022

Uh oh!

heathbm commented Jul 13, 2022• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

danmoseley commented Sep 6, 2022

Uh oh!

jkotas commented Sep 6, 2022

Uh oh!

stephentoub commented Sep 27, 2022

Uh oh!

dakersnar commented Oct 10, 2022

Uh oh!

ghost commented Oct 24, 2022

Uh oh!

ghost commented Nov 8, 2022

Uh oh!

dakersnar commented Nov 8, 2022

Uh oh!

stephentoub commented Nov 8, 2022

Uh oh!

jkotas commented Nov 8, 2022

Uh oh!

stephentoub commented Nov 8, 2022• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jkotas commented Nov 8, 2022

Uh oh!

stephentoub commented Nov 12, 2022

Uh oh!

stephentoub commented Nov 15, 2022

Uh oh!

driesengNov 15, 2022

Choose a reason for hiding this comment

Uh oh!

stephentoubNov 15, 2022• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

driesengNov 16, 2022• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

stephentoubNov 16, 2022

Choose a reason for hiding this comment

Uh oh!

jkotas commented Nov 16, 2022

Uh oh!

stephentoub commented Nov 16, 2022

Uh oh!

stephentoub commented Nov 16, 2022

Uh oh!

EgorBo commented Nov 16, 2022

Uh oh!

Reviewers

heathbm commented Jul 3, 2022•
edited
Loading

heathbm commented Jul 13, 2022•
edited
Loading

stephentoub commented Nov 8, 2022•
edited
Loading

stephentoubNov 15, 2022•
edited
Loading

driesengNov 16, 2022•
edited
Loading