Package: python2.6
Version: 2.6.6-8+b1
The Python interpreter arranges (during its startup) to trap SIGINT
and turn it into a KeyboardInterrupt exception. If this happens
before the execution of the part of the script (if any) which takes
countermeasures against this feature, the result is that the
interpreter prints the stack traceback to stderr and calls _exit(1)
(so that WIFEXITED(status) && WEXITSTATUS(status)==1).
This is a problem because it makes it impossible to write in Python
programs which behave like normal Unix programs - ie, which if they
receive a SIGINT always die with WIFSIGANLED(status) &&
WTERMSIG(status)==SIGINT.
For example, the "command-not-found" helper program is written in
Python and sometimes prints spurious stack traces. As another
example, any Python program which is expected to implement the
interface of (say) "diff" is buggy, because it may exit status 1
("comparison successful; differences found") when it should have died
with SIGINT ("comparison not completed due to interrupt signal"); this
could in principle cause data loss in some applications.
It is possible to reduce but not eliminate the time window during
which the bug will occur, by using Python's signal handling machinery
to put SIGINT back to SIG_DFL - ideally near the top of the script.
However, what is really needed is a way to fix the problem entirely.
Therefore we need a way to tell the Python interpreter not to trap
SIGINT. Not trapping SIGINT is better for programs because if they
want to trap it they can always do that themselves at any suitable
time during startup. (Any program which relies on trapping SIGINT at
_all_ times even before it gets control is buggy, of course, since
SIGINT may occur while eg the runtime linker is running.)
However, this would be an incompatible change. There therefore needs
to be a way for a script to declare to Python that normal SIGINT
behaviour is what it wants. I suggest that this should be a new
command-line option. The semantics of the option should simply be to
disable the setup of the default SIGINT trap.
This option can then be added to the #! line of all Python scripts.
(Or, at least, all Python scripts which might be invoked in
circumstances where it matters.)
I haven't done a comprehensive survey of affected Python scripts in
Debian, but almost anything in /usr/bin starting #!/usr/bin/python* is
probably affected.
Since I actually triggered this bug with command-not-found, I will
submit a bug about that specific instance and set this bug to block
it.
Please do not simply close this bug against Python unless either:
(a) you persuade the authors of command-not-found that the behaviour
exhibited by command-not-found (ie sometimes producing an
asynchronous stack backtrace to the user's stderr when they hit ^C)
is not a bug; or
(b) you explain some other reasonable way to fix the bug in
command-not-found.
Naturally this bug is in all versions of Python in Debian, ever.
Ian.
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
