[wasm][globalization][icu] Tracking issue for HybridGlobalization: Web API + ICU

[ilonatommy]

The task is to remove as much data from ICU files as possible and exchange ICU4C functions that are using this data with platform native functions - in the case of WASM with Web API. Because we are not able to get rid of ICU datafile completely (some functionalities are not easily replaceable) we will keep loading icudt.dat in a reduced form. This mode will be called HybridGlobalization and will be by default switched off. User can switch it on by setting MsBuild's <HybridGlobalization> to true. PoC branch is here: https://github.com/dotnet/runtime/compare/main...ilonatommy:runtime:icu-platform-native.

1) Removing collations for WASM

• [x] Prepare icudt_wasm.dat and corresponding sharded datafiles without collations/standard, enable setting HybridGlobalization and write WBT checking if the new file got loaded instead of the old one. https://github.com/dotnet/icu/pull/300 [ILONA]
• [x] Loading reduced ICU files to Blazor - this unified the startup for Blazor, so we could implement HG for it as well.
• [x] Finish implementing GlobalizationNative_ChangeCase + optimize memory usage - do not create a new string for returning the value but pass the address of buffer reserved on C# size that will hold the result. [ILONA] public API: TextInfo.ToLower, TextInfo.ToUpper, TextInfo.ToTitleCase`
• [x] Implement GlobalizationNative_IndexOf and GlobalizationNative_LastIndexOf - will not work for letters that consist of more than one grapheme, issue: https://github.com/tc39/ecma402/issues/506 [ILONA] public API: CompareInfo.IndexOf, String.IndexOf, MemoryExtensions.IndexOf, CompareInfo.LastIndexOf, String.LastIndexOf, MemoryExtensions.LastIndexOf.
• [x] Implement GlobalizationNative_StartsWith and GlobalizationNative_EndsWith [ILONA] public API: CompareInfo.IsSuffix, String.EndsWidth, MemoryExtensions.EndsWith, CompareInfo.IsPrefix, String.StartsWidth, MemoryExtensions.StartsWith.
• [x] Implement GlobalizationNative_CompareString (without Ordinal and OrdinalIgnoreCase, IgnoreKanaType, IgnoreWidth) [ILONA] public API: CompareInfo.Compare, String.Compare
• [x] Implement IgnoreKanaType and IgnoreWidth basing on pal_collation.c code [ILONA]
• [x] Investigate Ordinal and OrdinalIgnoreCase. [ILONA]
• [x] Throw PNSE on GlobalizationNative_GetSortKey
• [x] How much do we gain removing SortVersion? 32kB on uncompressed. Not worth it. If much, throw PNSE on GlobalizationNative_GetSortVersion.
• [x] Document how to use the flag and what to expect when switching it on.
• [x] Coordinate flow of HybridGlobalizationfrom Blazor. Changes in dotnet/sdk might be needed.

2) Removing normalization for WASM: Removed from planned Hybrid features. Savings from normalization removal on WASM are ~60kB. The removal breaks public APIs: string.Normalize, string.IsNormalized, IdnMapping.GetAsciii, IdnMapping.GetUnicode. Normalize/IsNormalized were succesfully replaced in https://github.com/dotnet/runtime/pull/85510. For GetAscii/GetUnicode replacement, Invariant implementation enhanced by normalization step was used, see branch https://github.com/ilonatommy/runtime/tree/idn-mapping. The mapping still lacked detection of disallowed/ignored/mapped characters and would need access to MappingTables of the current Unicode version to e.g. detect incorrect inputs to throw. One Unicode version mapping table in plain text weights ~900kB. Even if we compressed it, we still would need to maintain it with every Unicode version. Development time spent on correct implementation and chances of real size reduction, taking into cosideration the need to keep the mapping tables, are too small to remove normalization data from ICU.

• [ ] Update icudt_wasm.dat and corresponding sharded datafiles.
• [ ] Implement Punycode, ~~might be using this algorithm using InvariantGlobalization algorithm + normalization function.~~
• [x] Use normalization from the PoC branch.
• [ ] Update documentations.

3) Investigate implications of removing further data batches, e.g. check the effect of removing all collations, coll_ucadata, locales_tree etc.

• [ ] Fix no exception thrown for CultureInfoAll.LcidTest, CultureInfoAll.GetCultureTest, CultureInfoConstructor.Ctor_String (now we support wider range of locales so we should not expect some of them throw as it was with standard ICU)

4) (optional) Enhancement of collations by manual workarounds:

• [x] Assess risks connected with manual workarounds.
• [ ] Consider fixing some IgnoreSymbols by adding static data on JS side.
• [ ] Consider shifting katakana/hiragana and high/low symbols, based on pal_collation.c code
• [x] Check if we can implement invariant SortKey functionality for this scenario.
• [x] Refactoring according to suggestions: https://github.com/dotnet/runtime/pull/86895/files#r1222395670, https://github.com/dotnet/runtime/pull/86895/files#r1222406677 ...

5) (optional) Consider failing a build when HybridGlobalization function is not supported

• [ ] we are throwing PNSE when unsupported globalization functionality is used. Consider this idea: https://github.com/dotnet/runtime/pull/84249#discussion_r1158449084 to catch them already during the build time.

Tracking issues:

https://github.com/dotnet/runtime/issues/101912 https://github.com/dotnet/runtime/issues/102305 https://github.com/dotnet/runtime/issues/102373 https://github.com/dotnet/runtime/issues/95921 https://github.com/dotnet/runtime/issues/95795 https://github.com/dotnet/runtime/issues/95623

[ghost]

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

Issue Details

---

The task is to remove as much data from ICU files as possible and exchange ICU4C functions that are using this data with platform native functions - in the case of WASM with Web API. Because we are not able to get rid of ICU datafile completely (some functionalities are not easily replaceable) we will keep loading `icudt.dat` in a reduced form. This mode will be called `HybridGlobalization` and will be by default switched off. User can switch it on by setting MsBuild's `` to true. PoC branch is here: https://github.com/ilonatommy/runtime/tree/icu-platform-native. 1) Removing `collations/standard` for WASM - [ ] Prepare `icudt_wasm.dat` and corresponding sharded datafiles without `collations/standard`, enable setting `HybridGlobalization` and write WBT checking if the new file got loaded instead of the old one - [ ] 2) Removing `normalization` for WASM - [ ] Update `icudt_wasm.dat` and corresponding sharded datafiles - [ ] Implement Punycode, might be using [this algorithm](www.npmjs.com/package/punycode). - [ ] Use normalization from the PoC branch. ....

Author:	ilonatommy
Assignees:	ilonatommy, mkhamoyan
Labels:	`area-System.Globalization`
Milestone:	8.0.0

[ghost]

Tagging subscribers to 'arch-wasm': @lewing See info in area-owners.md if you want to be subscribed.

Issue Details

---

The task is to remove as much data from ICU files as possible and exchange ICU4C functions that are using this data with platform native functions - in the case of WASM with Web API. Because we are not able to get rid of ICU datafile completely (some functionalities are not easily replaceable) we will keep loading `icudt.dat` in a reduced form. This mode will be called `HybridGlobalization` and will be by default switched off. User can switch it on by setting MsBuild's `` to true. PoC branch is here: https://github.com/ilonatommy/runtime/tree/icu-platform-native. 1) Removing `collations/standard` for WASM - [ ] Prepare `icudt_wasm.dat` and corresponding sharded datafiles without `collations/standard`, enable setting `HybridGlobalization` and write WBT checking if the new file got loaded instead of the old one - [ ] 2) Removing `normalization` for WASM - [ ] Update `icudt_wasm.dat` and corresponding sharded datafiles - [ ] Implement Punycode, might be using [this algorithm](www.npmjs.com/package/punycode). - [ ] Use normalization from the PoC branch. ....

Author:	ilonatommy
Assignees:	ilonatommy, mkhamoyan
Labels:	`arch-wasm`, `area-System.Globalization`
Milestone:	8.0.0

[ilonatommy]

Closing, the planned work for HybridGlobalization was completed.