pty.spawn handles errors improperly

[niemeyer]

BPO	17824
Nosy	@vadmium

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2013-04-24.02:59:30.099>
labels = ['type-bug', 'library']
title = 'pty.spawn handles errors improperly'
updated_at = <Date 2017-01-12.21:01:10.088>
user = 'https://bugs.python.org/niemeyer'

bugs.python.org fields:

activity = <Date 2017-01-12.21:01:10.088>
actor = 'BreamoreBoy'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2013-04-24.02:59:30.099>
creator = 'niemeyer'
dependencies = []
files = []
hgrepos = []
issue_num = 17824
keywords = []
message_count = 4.0
messages = ['187681', '228264', '252445', '285300']
nosy_count = 2.0
nosy_names = ['niemeyer', 'martin.panter']
pr_nums = []
priority = 'normal'
resolution = None
stage = 'needs patch'
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue17824'
versions = ['Python 2.7', 'Python 3.4', 'Python 3.5']

[niemeyer]

This simple script will spawn N Python interpreters that aren't properly collected due to the improper error handling:

import pty
for i in range(N):
    try: pty.spawn(["/non-existent"])
    except: pass

[BreamoreBoy]

Can a linux guru comment on this please.

[vadmium]

The spawn() function has this code outside of any error handler:

pid, master_fd = fork()
if pid == CHILD:
os.execlp(argv[0], *argv)

If fork() succeeds, there will actually be a parent Python process and a child Python process. If exec() then fails, an exception will escape the spawn() function in the child process, while the parent process will carry on as if all is well. Maybe it would be worthwhile studying how the “subprocess” module handles exec() failure in the child process.

[vadmium]

The patch for bpo-26228 proposes an improvement to the situation, although it is not perfect and does not include a test. I wonder if it is possible to replace fork() and execlp() with a subprocess.Popen invokation, at least in a new Python release?