[ci] Set __PWSH_LOGIN_CHECKED to avoid macOS pwsh 'procargs failed' startup crashes

[PureWeen]

Note

Are you waiting for the changes in this PR to be merged?
It would be very helpful if you could test the resulting artifacts from this PR and let us know in a comment if this change resolves your issue. Thank you!

Description

PowerShell 7.4.x intermittently crashes at startup on macOS with:

Call to 'procargs' failed with errno 5
Unhandled exception. Microsoft.PowerShell.ManagedPSEntry+StartupException
   at Microsoft.PowerShell.ManagedPSEntry.ThrowOnFailure(String call, Int32 code)
   at Microsoft.PowerShell.ManagedPSEntry.AttemptExecPwshLogin(String[] args)
   at Microsoft.PowerShell.ManagedPSEntry.Main(String[] args)

This is PowerShell/PowerShell#20802 — open and untriaged since November 2023, 0 comments, no linked PRs, no fix shipped. The crash happens in AttemptExecPwshLogin, which on macOS calls sysctl(KERN_PROCARGS2) to inspect its own argv[0] and decide whether pwsh was launched as a login shell. That sysctl call races intermittently and returns errno 5, aborting startup before pwsh reads any user script.

Repro / impact

Hit on the internal dotnet-maui pipeline (definition 1095), build 2990586, during release-prep for release/11.0.1xx-preview5:

##[section]Starting: Provision Android SDK - Common Packages → Call to 'procargs' failed with errno 5 → task fails with exit code null.

A rerun on a different agent passed, but every macOS pwsh: / PowerShell@2 step in the pipeline is a flake risk going forward. Inventory: ~57 pwsh tasks across 15 pipeline YAMLs, ~45 of them on macOS-eligible jobs.

Fix

PowerShell itself ships a short-circuit for this code path. From src/powershell/Program.cs:

private const string LOGIN_ENV_VAR_NAME = "__PWSH_LOGIN_CHECKED";

if (Environment.GetEnvironmentVariable(LOGIN_ENV_VAR_NAME) != null)
{
    Environment.SetEnvironmentVariable(LOGIN_ENV_VAR_NAME, null);
    return;   // ← skips the racy sysctl(KERN_PROCARGS2) entirely
}

Setting __PWSH_LOGIN_CHECKED=1 makes pwsh skip the racy syscall. This PR declares it as a pipeline-level variable in eng/pipelines/common/variables.yml (included by every macOS-touching pipeline — ci.yml, ci-official.yml, ci-device-tests.yml, ci-copilot.yml, ci-uitests.yml, device-tests.yml, ui-tests.yml, handlers.yml, etc.). Per AzDO docs, user-defined variables are injected as environment variables for every task — so every pwsh: / PowerShell@2 step in every job inherits it.

Safety analysis

This is safe because:

• CI never wants login-shell semantics. No -Login flag, no leading - in argv[0]. The skipped code's IsLogin() check would have returned false and the function would have returned without doing anything visible. We just avoid the broken syscall that precedes that decision.
• The env var is single-use per process. AttemptExecPwshLogin consumes it via SetEnvironmentVariable(LOGIN_ENV_VAR_NAME, null), so nested pwsh-launches-pwsh scenarios still work normally (each new task gets a fresh AzDO-injected value).
• Windows is unaffected. The read is inside a #if UNIX block; Windows pwsh ignores the var entirely. Setting it on Windows agents is a no-op.
• Safe even under -Login. If a future step explicitly adds -Login, the short-circuit still routes around the broken syscall — the same syscall the login-shell detection relies on. So there's no regression even in that (currently nonexistent) use case.

Empirical validation

Verified locally on pwsh 7.4.5 / macOS 26.5 (same 7.4.x minor as the CI agent that crashed):

Run	Command	Result
Control	pwsh ... -Command 'Write-Output "envvar=[$Env:__PWSH_LOGIN_CHECKED]"'	envvar=[] (var unset; nothing to clear)
Test	__PWSH_LOGIN_CHECKED=1 pwsh ... -Command 'Write-Output "envvar=[$Env:__PWSH_LOGIN_CHECKED]"'	envvar=[] ← proves the short-circuit ran (the only line that clears the var is inside the skip-path; if it cleared, the racy syscall was never reached)
Soak	100 × __PWSH_LOGIN_CHECKED=1 pwsh ... -Command 'exit 0'	0/100 failures; no behavioral change

Backport plan

• This PR: main (.NET 10 SR dev). Drives the fix into ongoing servicing builds.
• Will cherry-pick to net11.0 afterward (where build 2990586 crashed during preview5 prep). Diff is verified to apply cleanly to both branches.

Related macOS-agent flakes (out of scope)

The same dnceng macOS agent pool has produced other, unrelated flakes during this release-prep window (e.g. IDEDownloadableMetalToolchainCoordinator: Failed to remount and iOS 26.2 platform not installed on builds 2990581 and 2990384). Those are agent-image issues and need agent-side fixes — they're not addressed here. This PR fixes one specific class of macOS flake (pwsh startup race), not all of them.

[github-actions]

🚀 Dogfood this PR with:

⚠️ WARNING: Do not do this without first carefully reviewing the code of this PR to satisfy yourself it is safe.

curl -fsSL https://raw.githubusercontent.com/dotnet/maui/main/eng/scripts/get-maui-pr.sh | bash -s -- 35724

Or

• Run remotely in PowerShell:

iex "& { $(irm https://raw.githubusercontent.com/dotnet/maui/main/eng/scripts/get-maui-pr.ps1) } 35724"