Use span for LengthConverter

[ThomasGoulet73]

Description

Use span for LengthConverter and other similar places. I also introduced the internal struct PixelUnit to share some of the logic.

I found that LengthConverter.FromString can be called a lot on startup (~3000 times in a sample app).

Benchmark

Method	s	Mean	Error	StdDev	Ratio	Gen0	Code Size	Allocated	Alloc Ratio
Old	12	37.447 ns	0.0442 ns	0.0369 ns	1.00	-	1,146 B	-	NA
New	12	33.244 ns	0.0190 ns	0.0148 ns	0.89	-	670 B	-	NA
Old	12cm	39.023 ns	0.0914 ns	0.0810 ns	1.00	0.0019	1,176 B	32 B	1.00
New	12cm	33.253 ns	0.0478 ns	0.0424 ns	0.85	-	714 B	-	0.00
Old	12CM	43.755 ns	0.1675 ns	0.1567 ns	1.00	0.0038	1,158 B	64 B	1.00
New	12CM	33.792 ns	0.0985 ns	0.0921 ns	0.77	-	714 B	-	0.00
Old	12in	36.953 ns	0.0672 ns	0.0561 ns	1.00	0.0019	1,176 B	32 B	1.00
New	12in	32.398 ns	0.0756 ns	0.0707 ns	0.88	-	693 B	-	0.00
Old	12IN	43.497 ns	0.1695 ns	0.1585 ns	1.00	0.0038	1,176 B	64 B	1.00
New	12IN	31.147 ns	0.0807 ns	0.0755 ns	0.72	-	664 B	-	0.00
Old	12pt	40.756 ns	0.0865 ns	0.0722 ns	1.00	0.0019	1,176 B	32 B	1.00
New	12pt	33.522 ns	0.0525 ns	0.0410 ns	0.82	-	679 B	-	0.00
Old	12PT	46.214 ns	0.1658 ns	0.1551 ns	1.00	0.0038	1,176 B	64 B	1.00
New	12PT	34.190 ns	0.0717 ns	0.0636 ns	0.74	-	676 B	-	0.00
Old	12px	35.508 ns	0.1054 ns	0.0985 ns	1.00	0.0019	1,176 B	32 B	1.00
New	12px	30.360 ns	0.0408 ns	0.0341 ns	0.85	-	630 B	-	0.00
Old	12PX	40.378 ns	0.1564 ns	0.1463 ns	1.00	0.0038	1,176 B	64 B	1.00
New	12PX	29.259 ns	0.0727 ns	0.0644 ns	0.72	-	630 B	-	0.00
Old	auto	5.010 ns	0.0224 ns	0.0210 ns	1.00	-	656 B	-	NA
New	auto	2.149 ns	0.0287 ns	0.0268 ns	0.43	-	509 B	-	NA

Customer Impact

Better perf and lower memory usage.

Regression

No.

Testing

Local testing.

Risk

Low.

Microsoft Reviewers: Open in CodeFlow

[pchaurasia14]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[lindexi]

Awesome

[ThomasGoulet73]

I updated my branch to main to fix a conflict and pushed a commit with the PR feedback.

[harshit7962]

Overall the changes looks good. Checked performance gain on multiple applications. There is about 4% gain on performance at xaml reading and it reflects out to be around 2% of gain at LoadComponent level when averaged out over multiple runs.

[h3xds1nz]

Sounds amazing, I can't wait to push a PR benefiting from those changes! 😄

[harshit7962]

@ThomasGoulet73 we have encountered a use case where the following changes are causing a regression. Please find the minimal repro here.

A bit of context, and to summarize the failure, the Height specified to Figure tag here does not take Content as a unit with these changes.

Can you please look if we can mitigate this failure. I believe we might need changes in XamlFigureLengthSerializer here and might have to possibly check for similar use case in XamlGridLengthSerializer.

[ThomasGoulet73]

Hi @harshit7962, I think I found the problem. Here it should've been StringComparison.OrdinalIgnoreCase instead of StringComparison.Ordinal: https://github.com/dotnet/wpf/pull/8622/files#diff-e8d1ce9b0355662f53afd3f319bfcb23d2d2c8b1baa37572c7daacfed58bbf83R213

I looked at my other changes and they all use StringComparison.OrdinalIgnoreCase correctly, I'm not sure how I missed this one.

I'll test to make sure it fixes your repro later today.

[ThomasGoulet73]

@harshit7962: I was able to reproduce the issue using your minimal repro and I have confirmed that the suspected fix in my previous comment did fix the issue.

Thank you for the simple repro!

[harshit7962]

@ThomasGoulet73 verified that the edge case is fixed with the change. Thank you for your contributions.

[ThomasGoulet73]

Thanks @harshit7962