2094265047
Remove metadata-only source caching so mapped edits and transient read failures cannot leave stale rebuild decisions. Keep directory pruning, lazy npm lookup, renderer manifests and the COM scheduler query. Separate stdout and stderr logging, decode bounded logical lines before redaction, and route desktop records to the GUI log. Add real-child, mapped-write and temporary-read-lock regressions. Verified by independent review, targeted Windows/Linux suites, renderer tests and packaged Windows manual checks. Existing Windows logging failures reproduce on the parent; no end-to-end startup percentage is claimed for this corrected version.
71 lines
2.6 KiB
Python
71 lines
2.6 KiB
Python
"""Keep Windows console backpressure off the packaged desktop startup path."""
|
|
|
|
import logging
|
|
import os
|
|
import sys
|
|
import threading
|
|
import time
|
|
from contextlib import contextmanager
|
|
|
|
logger = logging.getLogger("hermes_cli.desktop")
|
|
|
|
|
|
def desktop_launch_notice(message: str, *, source_mode: bool = False) -> None:
|
|
if sys.platform == "win32" and not source_mode:
|
|
# setup_logging() uses an async queue and rotating, redacted files.
|
|
logger.info(message)
|
|
else:
|
|
print(message)
|
|
|
|
|
|
@contextmanager
|
|
def desktop_console_output(*, source_mode: bool):
|
|
"""Drain packaged Windows stdout/stderr into the existing rotating logs.
|
|
|
|
Inheriting the console lets an unresponsive console host/profiler block
|
|
Electron's synchronous startup messages before it can create a window.
|
|
Keep source launches interactive and preserve subprocess.run's lifecycle
|
|
and exit-code handling. Read bounded lines so output cannot grow a buffer
|
|
without limit, and never wait indefinitely on a descendant's open pipe.
|
|
"""
|
|
if sys.platform != "win32" or source_mode:
|
|
yield {}
|
|
return
|
|
|
|
def drain(read_fd, level):
|
|
with os.fdopen(read_fd, "rb") as stream:
|
|
# Decode/redact complete records, not arbitrary pipe-read fragments.
|
|
# Discard an oversized line in full: even its prefix may be a secret.
|
|
limit = 128 * 1024
|
|
while data := stream.readline(limit + 1):
|
|
if len(data) > limit:
|
|
while data and not data.endswith(b"\n"):
|
|
data = stream.readline(8192)
|
|
logger.log(level, "[desktop] [oversized line omitted]")
|
|
else:
|
|
logger.log(level, "[desktop] %s", data.decode("utf-8", errors="replace").rstrip())
|
|
|
|
streams = {}
|
|
readers = []
|
|
try:
|
|
# Keep diagnostics visible even when ordinary INFO output is disabled.
|
|
for name, level in (("stdout", logging.INFO), ("stderr", logging.ERROR)):
|
|
read_fd, write_fd = os.pipe()
|
|
streams[name] = write_fd
|
|
reader = threading.Thread(
|
|
target=drain, args=(read_fd, level), name=f"desktop-console-{name}", daemon=True,
|
|
)
|
|
try:
|
|
reader.start()
|
|
except BaseException:
|
|
os.close(read_fd)
|
|
raise
|
|
readers.append(reader)
|
|
yield streams
|
|
finally:
|
|
for write_fd in streams.values():
|
|
os.close(write_fd)
|
|
deadline = time.monotonic() + 1
|
|
for reader in readers:
|
|
reader.join(timeout=max(0, deadline - time.monotonic()))
|