feat(string): Implement UTF-8 string conversion and validation functions

[bobtista]

• Relates to refactor(string): Add functions for handling UTF8 encoded strings #2045

Adds portable UTF-8/wide-character conversion helpers to WWLib and uses them for engine string translation and the existing GameSpy conversion paths.

This picks up the work from #2045 with API and behavior changes based on review feedback.

WWLib/utf8.h / utf8.cpp

Adds four conversion functions:

• Wide_To_Utf8_Len
• Utf8_To_Wide_Len
• Wide_To_Utf8
• Utf8_To_Wide

The _Len functions return the required destination size without counting a null terminator. The conversion functions return the number of units written and write a null terminator when the destination has spare capacity.

Malformed UTF-8 is reported as UTF8_INVALID, keeping it distinct from a successful empty conversion, which returns 0.

The decoder validates UTF-8 according to RFC 3629, rejecting:

• malformed and truncated sequences;
• overlong encodings;
• surrogate codepoints;
• codepoints above U+10FFFF.

The encoder substitutes U+FFFD for unpaired UTF-16 surrogates and wide values that have no valid UTF-8 representation. Its output therefore always decodes successfully through Utf8_To_Wide.

The implementation uses no platform text APIs. It adapts to the platform width of wchar_t:

• 16-bit wchar_t is handled as UTF-16, including surrogate pairs;
• 32-bit wchar_t is handled as UTF-32 codepoints.

AsciiString::translate

Replaces the original 7-bit-only wide-to-narrow translation with UTF-8 encoding.

UnicodeString::translate

Decodes valid UTF-8 into the platform wide-character representation.

When the input is not valid UTF-8, translation falls back to the previous unsigned-byte-to-wide-character behavior. This preserves legacy byte values instead of rejecting or replacing the input. It is a compatibility fallback, not a complete code-page-specific decoder.

Because AsciiString does not carry encoding metadata, valid UTF-8 is preferred over the legacy fallback.

ThreadUtils.cpp

Replaces the GameSpy-specific Win32 conversion calls with the shared WWLib functions.

MultiByteToWideCharSingleLine uses the same invalid-UTF-8 fallback as UnicodeString::translate. Newline and carriage-return replacement remains unchanged.

Non-goals

This change does not make narrow filesystem APIs Unicode-aware. Paths are still passed to the existing narrow platform APIs; updating those APIs belongs in a separate change.

Verification

• Tested ASCII and 2-, 3-, and 4-byte UTF-8 round trips.
• Tested both 16-bit and 32-bit wchar_t paths.
• Tested rejection of malformed, truncated, overlong, surrogate, and out-of-range UTF-8 sequences.
• Tested replacement of unpaired wide surrogates and out-of-range wide values.
• All Generals and GeneralsMD VC6 and Win32 CI configurations pass.
• GeneralsMD replay checks pass.

[greptile-apps]

Greptile Summary

Adds a portable UTF-8/wide-character transcoder and integrates it into engine string and GameSpy conversion paths.

• Validates UTF-8 by rejecting malformed, overlong, surrogate, and out-of-range encodings.
• Replaces the previous ASCII-only AsciiString and UnicodeString translations with UTF-8 conversion and a legacy byte-cast fallback.
• Replaces GameSpy’s Win32 conversion calls with shared WWLib helpers.
• Registers the new implementation in the WWLib build.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
Core/Libraries/Source/WWVegas/WWLib/utf8.cpp	Implements width-aware UTF-8 transcoding with checks for malformed, overlong, surrogate, and out-of-range sequences.
Core/Libraries/Source/WWVegas/WWLib/utf8.h	Defines the portable conversion API, invalid-input sentinel, sizing contracts, and destination-buffer behavior.
Core/GameEngine/Source/Common/System/AsciiString.cpp	Replaces lossy wide-to-byte casting with correctly sized UTF-8 encoding.
Core/GameEngine/Source/Common/System/UnicodeString.cpp	Adds validated UTF-8 decoding while preserving legacy non-UTF-8 data through byte-wise fallback.
Core/GameEngine/Source/GameNetwork/GameSpy/Thread/ThreadUtils.cpp	Migrates GameSpy string conversion from Win32 APIs to WWLib and handles invalid UTF-8 through the documented fallback.
Core/Libraries/Source/WWVegas/WWLib/CMakeLists.txt	Adds the portable UTF-8 source and header to the WWLib target.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Engine or GameSpy text] --> B{Conversion direction}
    B -->|UTF-8 to wide| C[Utf8_To_Wide_Len]
    C --> D{Valid UTF-8?}
    D -->|Yes| E[Utf8_To_Wide]
    D -->|No| F[Legacy one-to-one byte fallback]
    B -->|Wide to UTF-8| G[Wide_To_Utf8_Len]
    G --> H[Wide_To_Utf8]
    E --> I[UnicodeString or std::wstring]
    F --> I
    H --> J[AsciiString or std::string]

Loading

_{Reviews (17): Last reviewed commit: "build: Use qualified include path for WW..." | Re-trigger Greptile}

[bobtista]

• Fixed the if formatting
• added RFC 3629 overlong and out-of-range checks
• RE the theoretical memory leak, can that even happen here? set() allocates via the engine's custom memory allocator which crashes on failure rather than throwing, so the leak path can't really be reached right?

[xezon]

Get_Utf8_Size should not include the null terminator in its size.

[xezon]

The diff now shows unrelated changes.

[bobtista]

The diff now shows unrelated changes.

Try again cleaned up the commits and force pushed

[Mauller]

So is this translating UTF16LE from windows into UTF8 that is then stored within AsciiString?

If so this may help with the paths issue with usernames and paths not using Latin characters, but file handling functions will need updating to use unicode variants instead of Ascii.

[bobtista]

Yes. It does not fix path handling on its own though, the file APIs still take the narrow string. Should that be a separate change?

[Mauller]

I wonder if we should also add a flag to state that the Ascii string is holding a UTF8 string?

I guess all normal ascii characters will display properly, it's just extended character sets that will look garbled.

[OmniBlade]

Slight nitpick on function naming, shouldn't it be Utf16 rather than Wchar as Wchar is only Utf16 on windows yet we will likely want the conversion functions on other platforms as well for at least csf parsing if nothing else.

[bobtista]

Slight nitpick on function naming, shouldn't it be Utf16 rather than Wchar as Wchar is only Utf16 on windows yet we will likely want the conversion functions on other platforms as well for at least csf parsing if nothing else.

Yeah, but it's consistent with the other naming as it is for now. How about we keep the naming and using a uint16_t/char16_t type internally rather than wchar_t when we make non windows paths? Or would you rather we rename to something like Utf16_To_Utf8?

[Mauller]

Slight nitpick on function naming, shouldn't it be Utf16 rather than Wchar as Wchar is only Utf16 on windows yet we will likely want the conversion functions on other platforms as well for at least csf parsing if nothing else.

Yeah, but it's consistent with the other naming as it is for now. How about we keep the naming and using a uint16_t/char16_t type internally rather than wchar_t when we make non windows paths? Or would you rather we rename to something like Utf16_To_Utf8?

If anything it would be Utf16Le_To_Utf8, windows uses the little endian utf16 format. Not sure if Utf16Be is used much anywhere but worth being concise with it.

[bobtista]

Pushed updates addressing the open comments:

• Portable transcoder, dropped the Win32 WideCharToMultiByte/MultiByteToWideChar wrappers and the #else #error for a hand-rolled RFC 3629 encode/decode in utf8.cpp. It compiles on all platforms now, and the wide side adapts to wchar_t width at compile time (UTF-16 with surrogate pairs on Windows, UTF-32 elsewhere), so it's reusable for CSF parsing off-Windows. @OmniBlade — I kept the Utf16Le_* naming for now per the earlier thread; happy to rename if you'd prefer.
• CP1252 fallback, UnicodeString::translate now validates first. Valid UTF-8 decodes as UTF-8; anything else (legacy CP1252) falls back to the old 1:1 byte cast instead of producing replacement characters. Bonus: data written by the old lossy path reads back byte-identical.
• Scope, reverted the ensureUniqueBufferOfSize null-terminator change, so the string files now carry only the translate rewrite.

Tested the codec in isolation: ASCII / 2- / 3- / 4-byte round-trips on both wide widths, plus RFC 3629 rejection of overlong, surrogate, out-of-range, and truncated sequences.

Ready for another pass

[bobtista]

Reworked the conversion interface:

Utf8_To_Wide_Len / Utf8_To_Wide return UTF8_INVALID on bad input. 0 now means empty only, so callers can tell the two apart.
The encoder substitutes U+FFFD for unpaired surrogates and out of range wide values. Before this, a lone surrogate encoded to ED A0 80, which the decoder then rejected.
Renamed to Wide_To_Utf8 / Utf8_To_Wide. The wide side is wchar_t, which is UTF-32 off Windows, so the Utf16Le name was wrong.
MultiByteToWideCharSingleLine falls back to a byte cast on non-UTF-8, matching UnicodeString::translate. It returned an empty string before.
Dropped Utf8_Validate, Utf8_To_Wide_Len covers it and nothing else called it.

[xezon]

I think it is a good idea to make this custom implementation and be independent of Windows API.

[CryoTheRenegade]

Have we thought about implementing ICU4C instead of a custom implementation?
Also for posix compatible utf-8, we could probably reference this

[OmniBlade]

Not a fan of having different encodings on different platforms personally and went with an ICU4C approach as suggested by CryoTheRenegade for OpenW3D which keeps the representation as utf-16 for Wide strings on all platforms. The refactor to support that is larger though and it adds a dependency.

[bobtista]

Have we thought about implementing ICU4C instead of a custom implementation? Also for posix compatible utf-8, we could probably reference this

Yes we've talked about ICU, so far the play has been hold off until VC6 is dropped

[xezon]

Can we use ICU for non VC6 and whatever else for VC6 ?

[bobtista]

Can we use ICU for non VC6 and whatever else for VC6 ?

Kind of, but we'd want to replace wchar_t with UTF-16 first. I think we merge this, then use OmniBlade’s OpenW3D approach to refactor WideChar/UnicodeString to fixed 16-bit storage. Then we can use ICU because both VC6 and non VC6 builds would share the same internal representation and public contract, only the backend would differ.

[xezon]

Looks good. A few things can be polished.

[xezon]

Any idea why this is removing \r\n ? Probably should have been a different function doing that.

[xezon]

While at it, maybe optimize these loops. It is searching for the character from the beginning of the string on every iteration. Or better in a follow up change.