| Commit message (Collapse) | Author | Files | Lines |
|---|
|
hardcoded "Logger", so derived classes (eg, StampedLogger) identify
themselves correctly.
.__init__(): Clarified function docstring a bit.
|
|
expressed, and log failures to smtp-failures.
|
|
wrapper scripts. Apparently the Roxen server requires this.
|
|
potential confusion pointed out by jeremy.)
|
|
or else failures, like bad local recipients, would remain on the queue
forever, and cause repeat deliveries. (I believe this is part of the
problem behind the repeats we're seeing - but i'm not convinced it's
all of it.)
To fix, i made it so only partcicular exceptions - currently
socket.error, for absent SMTP listener - cause items to be left on
queue. Otherwise it falls through to a blanket except which discards
the item, leaving a note in the error log to that effect.
*** I think where the info should go back to the maillist structure,
to, e.g., disable the recipient, or whatever. However, this is
happening in a forked process, so we cannot use an exception, and the
routine itself doesn't, and shouldn't know which the list is.
|
|
others.
|
|
|
|
|
|
seemed necessary for the mmsitepass, and I figured other scripts
couldn't hurt.
|
|
don't accidently get the built-in list function. Second, import and
run main() of the passed in `which' script, which seems currently to
be only options.py.
Q: How did this ever work???
|
|
uncaught exceptions through to the top level.
Uncommented out the open() definition. I think this is very helpful
during the beta debugging phase. We'll comment this out for 1.0
release (it also won't be necessary in Python 1.5.2, but nobody has
that yet!)
Exception printing has been changed so that the MultiLogger is no
longer necessary. Removes one more potential crash, but also
rationalizes print_exception() and print_environment(). There's
always exactly two places information goes:
1. To a plain text log file. This could be sys.__stderr__ if no
explicit log file is given -- e.g. the Web server's log file.
2. To an HTML sink. This will always be sys.stdout (since we use
the `print' statement), and will generally always be connected
to the input of the Web browser.
Log file printing happens first, just in case (it's better that the
server knows what's happening), but we always try to print both.
`Main' has been rewritten also. We now do not catch any problems with
import sys. If sys can't be imported, then the Python installation is
royally screwed -- even `print' won't work! Better to just let the
catastrophy percolate up to the Web browser. Even so this should
almost never happen.
Poke __stderr__ and __stdout__ into the sys module if they aren't
already there. GvR says this is a 1.5.1 feature, but we can make it
work well enough for older versions of Python, and it's convenient.
Give one final shot at printing the traceback and environment if the
exception percolates all the way up out of run_main(). This time,
just print them to stderr (usually meaning the Web server's error
log). Should this fail too, do whatever we can not to generate an
end-user visible server error.
|
|
which is passed through to the Logger.__init__().
|
|
XXX: This class is no longer used by the driver script.
|
|
Logger.__init__(): New default argument `immediate' says to open the
log file immediately upon creation (so we learn of errors early).
Otherwise the log file is opened when first needed.
Logger.__get_f(): Re-raise the exception using Python 1.5 semantics;
can't guarantee everyone has Python 1.5.1. Also if nofail, then set
the file pointer to be sys.__stderr__ so at least the exceptions will
get logged to the Web server log file.
|
|
doesn't get name mangled.
|
