Client-side mode: accented characters in test class corrupted during load

[GendronAC]

Summary

In client-side editing mode on Windows, test classes containing accented characters (é, è, à, etc.) fail when run via "Run Test" but pass via "Debug Test".

Related: intersystems-community/vscode-objectscript#692

Minimal reproduction

Two .cls files in the workspace (UTF-8 without BOM):

CSampleData.cls — already compiled on server, not reloaded by the test run:

Class Tests.Unit.CSampleData Extends %RegisteredObject
{
ClassMethod returnsData() As %String
{
   return "Différents"
}
}

CSample.cls — test class, reloaded by $System.OBJ.Load before execution:

Class Tests.Unit.CSample Extends %UnitTest.TestCase
{
Method TestMe()
{
   do $$$AssertEquals(##class(Tests.Unit.CSampleData).returnsData(), "Différents")
}
}

Run Test output (FAILS)

Loading file d:\...\CSample.cls as udl
AssertEquals:##class(CSampleData).returnsData()== "Diff?ents" was 'Différents' (failed)

The string "Différents" in CSample.cls is corrupted during load. The é (UTF-8 bytes C3 A9) becomes ?.

Debug Test output (PASSES)

AssertEquals: (passed)

Debug works because the preload uses /load/debug, then the actual run uses /noload — so the already-compiled (correct) class is used.

Byte-level evidence

• Source .cls file on disk: UTF-8, no BOM. é = 0xC3 0xA9
• CSampleData.returnsData() internal bytes on server: $ASCII = 233 (correct, U+00E9)
• After $System.OBJ.Load of CSample.cls: the literal "Différents" is corrupted

Environment

• Windows 11, French locale
• IRIS for Windows (x86-64) 2025.1.3 (Build 481_1U) Tue Feb 3 2026 15:26:39 EST
• Default I/O translate table: UTF8 ($zutil(96,18,0) returns UTF8)
• Client-side editing mode
• ^%SYS("CSP","DefaultFileCharset")="UTF-8" was tried — no effect (expected, since this affects CSP file writes, not $System.OBJ.Load) csp-Files encoding issue vscode-objectscript#692

[GendronAC]

REST traffic analysis

With "objectscript.outputRESTTraffic": true, here is the actual exchange:

POST (request) — accent is correct:

POST /api/atelier/v8/REG05_ENSEMBLE/work
Content-Type: application/json

{
  "request": "unittest",
  "debug": false,
  "load": [{
    "file": "d:\\...\\CSample.cls",
    "content": ["...", "do $$$AssertEquals(##class(CSampleData).returnsData(), \"Différents\")", "..."]
  }]
}

"Différents" is correctly UTF-8 encoded in the request body. ✓

GET (response) — accent is corrupted:

GET /api/atelier/v8/REG05_ENSEMBLE/work/82626954
Content-Type: application/json; charset=utf-8

{
  "console": [
    "AssertEquals:##class(CSampleData).returnsData()== \"Diff?ents\" was 'Différents' (failed)"
  ]
}

The é sent in the POST (Différents) becomes ? in the compiled class, but the runtime value from CSampleData.returnsData() returns the correct Différents. The corruption happens server-side between receiving the load[].content and compiling the class.

Debug mode does not use this /work endpoint — it uses the older debugger-based flow, which is why it passes.