[akacross]

Fixes the dedicated-server smoke timeouts and most of the renderer startup-matrix failures that every PR and main currently show in CI. Both are one regression from f15a41f.

Cause

idCmdArgs::argv points into the object's own tokenized buffer, and idCmdArgs::operator= re-points it. f15a41f made idList::Append copy its argument into a temporary before growing the list (const type value = obj;), but idCmdArgs has no copy constructor. The implicit member-wise copy leaves value.argv aimed at the caller's buffer, and list[num] = value then computes each offset as value.argv[i] - value.tokenized, from two unrelated buffers.

Startup + commands are queued in idCmdSystemLocal::tokenizedCmds, an idList<idCmdArgs> that grows every 16 entries. So the 17th, 33rd, ... startup command reaches ExecuteTokenizedString with an empty argv[0] and is skipped silently. The same commit stopped reporting empty commands as unknown.

What it broke in CI

• Dedicated-server smokes (Linux, macOS, Windows): +quit is the 17th command, so the server runs until the 45 s timeout ("missing marker: Game Shutdown"). On main the step only looks green because earlier failures skip it.
• Renderer startup matrix: the self-tests that land on a growth never run ("RendererRenderGraphResource / RendererModernVisibility self-test passed", the Vulkan render-target/MRT/resolve set).
• Outside CI, any client command line with more than 16 + commands, +set included, loses one, e.g. a +map after many +sets.

Fix

idCmdArgs( const idCmdArgs &args ) { argc = 0; *this = args; }, so every copy re-points argv.

Verification (Linux x64, GCC 16)

Check	main	this branch
tools/tests/linux_dedicated_server_smoke.py	fail: 45 s timeout, no Game Shutdown	pass, 1.1 s
renderer_validation_matrix.py --cases renderer-foundation-selftests (CI arguments)	fail: RenderGraphResource and ModernVisibility missing	pass
renderer_validation_matrix.py --cases renderer-vk-clear-startup	fail: 9 Vulkan self-test markers missing	those 9 present*
client with +quit as the 17th + command	still running at 60 s	exits in 4 s

*Locally it still misses only "Vulkan: validation enabled", because this machine has no Khronos validation layer installed. CI did not list that marker as missing.

New tools/tests/cmd_args_copy.py (wired into both workflows and openq4_validate.py) compiles the production idCmdArgs header, operator= and AppendArg against the real idList, copy-constructs a command and queues 40 through several growths. It fails on main and passes here.

Remaining CI failures this doesn't address

• Windows ARM64: openq4-imported-game-regressions fails in mp_match_authentication_contract, mp_match_pause_contract and mp_match_teams_contract. They pass on Linux and Windows x64. Their logs (game-regression-logs/) aren't uploaded, so I couldn't see why. The contracts' compiler probe asks vswhere for the x64 MSVC toolset, which is a plausible ARM64 difference.
• Linux x64 ASan+UBSan: the sanitizer build alone takes about 82 of the 90 minutes, so tests are cut off. It needs a longer timeout or a compiler cache.
• macOS Intel and Windows ARM64 timeouts: the profile's roughly 300 sequential Python checks run past 90 minutes. Run independent validation checks in parallel #191 runs them in parallel.

🤖 Generated with Claude Code

[themuffinator]

The copy constructor correctly reuses the existing rebasing assignment operator. Independently compiled the production header/operator=/AppendArg with the real idList on Windows: cbe967f loses queued and drained commands 17 and 33; this head preserves all 40 commands and passes the direct-copy case. cmd_args_copy.py also passes locally.

The successful Windows x64, Linux Clang/ARM64 and macOS ARM64 checks support the fix. Remaining timeouts are not passes. The cross-build failure was in Ubuntu package installation before compilation (liblzma5:arm64 dependency resolution); I have retried that job once. I am preparing joint validation with #191 and CI diagnostics so the command-queue fix and reduced validation time can be qualified together before integration.

[themuffinator]

Joint validation with #191 is now running in draft integration PR #200. It preserves both original commit histories and adds Windows ARM64 native failure diagnostics plus time for sanitizer tests after the build. The independent Windows native negative control reproduces the lost 17th/33rd commands on main and passes on this head. The standalone ARM64 cross-build retry has also progressed past the earlier package-install failure.

[themuffinator]

The targeted Linux ARM64 cross-build retry completed successfully, including ELF verification and staged-package validation: run 39460306900. The original apt dependency failure did not recur. The integrated #200 cross build also passed. Other platform/CI gates remain outstanding; this does not turn the previous timeouts or Windows ARM64 native failures into passes.