sapling/hgext3rd/errorredirect.py
Jun Wu 6c51d725bd errorredirect: log and fallback to old error handler when appropriate
Summary:
Previously errorredirect swallows errors in some cases. This diff makes it
more robust.

  - Always log commandexception. `blackbox.log` will probably catch it.
  - Print the trace if the handler script cannot be executed.
  - Print the trace if the handler script exits with non-zero.
  - Add exception for the Ctrl+C case, to not pollute user's terminal.

Test Plan: Added some test cases

Reviewers: #mercurial, rmcelroy

Reviewed By: rmcelroy

Subscribers: rmcelroy, mjpieters

Differential Revision: https://phabricator.intern.facebook.com/D4824767

Signature: t1:4824767:1491299807:3fd863cb99ce71a6c2b59643b0c89eece7985d4b
2017-04-04 10:27:30 -07:00

92 lines
2.9 KiB
Python

# errorredirect.py
#
# Copyright 2015 Facebook, Inc.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""redirect error message
Redirect error message, the stack trace, of an uncaught exception to
a custom shell script. This is useful for further handling the error,
for example posting it to a support group and logging it somewhere.
The config option errorredirect.script is the shell script to execute.
If it's empty, the extension will do nothing and fallback to the old
behavior.
Two environment variables are set: TRACE is the stack trace, which
is the same as piped content. WARNING is the warning message, which
usually contains contact message and software versions, etc.
Examples::
[errorredirect]
script = tee hgerr.log && echo 'Error written to hgerr.log'
[errorredirect]
script = echo "$WARNING$TRACE" >&2
[errorredirect]
script = (echo "$WARNING"; cat) | cat >&2
"""
import signal
import subprocess
import traceback
from mercurial import (
dispatch,
encoding,
extensions,
)
def _printtrace(ui, warning):
# Like dispatch.handlecommandexception, but avoids an unnecessary ui.log
ui.warn(warning)
return False # return value for "handlecommandexception", re-raises
def _handlecommandexception(orig, ui):
warning = dispatch._exceptionwarning(ui)
trace = traceback.format_exc()
# let blackbox log it (if it is configured to do so)
ui.log("commandexception", "%s\n%s\n", warning, trace)
script = ui.config('errorredirect', 'script')
if not script:
return _printtrace(ui, warning)
# run the external script
env = encoding.environ.copy()
env['WARNING'] = warning
env['TRACE'] = trace
# decide whether to use shell smartly, see 9335dc6b2a9c in hg
shell = any(c in script for c in "|&;<>()$`\\\"' \t\n*?[#~=%")
try:
p = subprocess.Popen(script, shell=shell, stdin=subprocess.PIPE,
env=env)
p.communicate(trace)
except StandardError:
# The binary cannot be executed, or some other issues. For example,
# "script" is not in PATH, and shell is False; or the peer closes the
# pipe early. Fallback to the plain error reporting.
return _printtrace(ui, warning)
else:
ret = p.returncode
# Python returns negative exit code for signal-terminated process. The
# shell converts singal-terminated process to a positive exit code by
# +128. Ctrl+C generates SIGTERM. Re-report the error unless the
# process exits cleanly or is terminated by SIGTERM (Ctrl+C).
ctrlc = (ret == signal.SIGTERM + 128) or (ret == -signal.SIGTERM)
if ret != 0 and not ctrlc:
return _printtrace(ui, warning)
return True # do not re-raise
def uisetup(ui):
extensions.wrapfunction(dispatch, 'handlecommandexception',
_handlecommandexception)