Fix fdsan violation in StdoutLogger::Stop() on Android API 29+

[CedricGuillemet]

Problem

StdoutLogger::Start() calls fdopen(fd[0], "r") to hand each pipe read-end to a reader thread. fdopen() transfers ownership of the file descriptor to the FILE*; it must be released with fclose(), not close().

StdoutLogger::Stop() was calling close(fd_stdout[0]) and close(fd_stderr[0]) directly. On Android API 29+, fdsan tracks descriptor ownership and aborts with SIGABRT when a raw close() is called on a descriptor owned by a FILE*:

Fatal signal 6 (SIGABRT)
Abort message: 'fdsan: attempted to close file descriptor 94, expected to be unowned, actually owned by FILE* 0x6e357db180'
  #03 android::StdoutLogger::Stop()+184

Fix

In Stop(), only close the write ends (fd[1]). Closing fd[1] sends EOF through the pipe; getline() in the reader thread returns -1, exits the loop, and calls fclose(stream) — which correctly releases the read end. The explicit close(fd[0]) calls are removed entirely.

[Copilot]

Pull request overview

This PR aims to prevent Android API 29+ fdsan aborts by avoiding close() on pipe read-ends that have been transferred to FILE* ownership via fdopen() in StdoutLogger.

Changes:

• Updates StdoutLogger::Stop() to close only the pipe write-ends and stop directly closing pipe read-ends.
• Adds explanatory comments about fdopen() ownership and fdsan behavior on Android API 29+.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bghgary]

[Approved by Copilot on behalf of @bghgary]

LGTM.

A few pre-existing issues in this file, not introduced here and not blocking, but worth tracking as follow-ups:

• Detached reader threads are never joined — Stop() returns while they may still be draining.
• Original stdout/stderr fds aren't saved, so writes after Stop() hit a closed pipe (EPIPE).
• pthread_create error check compares against -1, but it returns an errno code on failure; the error path also leaks the pipe fds. pipe()'s return value is ignored too.