commit d623c44351e3f0de0f8c237df73b3cad441ca65c
author Raphael Kubo da Costa <kubo@igalia.com> Tue Mar 11 15:13:10 2025
committer Chromium LUCI CQ <chromium-scoped@luci-project-accounts.iam.gserviceaccount.com> Tue Mar 11 15:13:10 2025
tree 7fc02ed3723acde058ef0afd5972514131c681d9
parent e63915025934d03d66e48b232016ac01d20143e3

iossim: Invoke `xcrun simctl launch` with --console-pty, not --console

When running content_shell and hitting certain (D)CHECKs with a long
stack trace, the entire process was blocking and freezing indefinitely
on the write() call to stderr when the length of the string was larger
than 8192 bytes (half of vm.pagesize on a 2023 Mac Studio).

It is hard to know what exactly is happening in `xcrun simctl` given
its (extremely) concise public documentation, but any write to stderr (I
have not checked stdout) larger than 8kb would cause the same hang on
the iOS and tvOS simulators.

Using `--console-pty` instead of `--console` fixes the issue, possibly
due to some different kind of buffering inside xcrun.

blinkpy's IOSSimulatorServerProcess had to be adjusted accordingly, as
with the change stderr was being lost and all web test runs were timing
out.

The solution is to make the code more similar to FuchsiaServerProcess's,
which is what it was originally based upon anyway: since iOS uses the
same trick as Fuchsia and redirects stdin and stdout to a Unix domain
socket, we can merge stderr and stdout in the pipe() call and use that
Popen object's stdout as its stderr, as the real stdout was actually
sent to the socket.

Bug: 391990604, 40254930, 40259174
Change-Id: Ifb9f6f6d99df928ed19913c66da7bfd9d87b7650
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6339850
Reviewed-by: Jonathan Lee <jonathanjlee@google.com>
Reviewed-by: Rohit Rao <rohitrao@chromium.org>
Commit-Queue: Raphael Kubo da Costa <kubo@igalia.com>
Cr-Commit-Position: refs/heads/main@{#1430864}