mirror of
https://https.git.savannah.gnu.org/git/bash.git
synced 2026-06-30 08:59:56 +02:00
Chet Ramey
85 lines
2.6 KiB
Plaintext
85 lines
2.6 KiB
Plaintext
CRA = my diffs.
|
|
BDE = Bruce diffs.
|
|
|
|
BDE version 19980212:
|
|
|
|
Test 2: It should send SIGQUIT to cat, it doesn't.
|
|
|
|
Test 9: can't break wait()
|
|
|
|
%% older notes:
|
|
|
|
test2:
|
|
trap '' 2
|
|
while : ; do cat ; echo -n $? ; done
|
|
test6:
|
|
trap '' 3
|
|
while : ; do cat ; echo -n $? ; done
|
|
|
|
The trapped signal should be ignored. If SIGINT is ignored (test 2),
|
|
SIGQUIT exits `cat`, but not the shell (means the loop will continue
|
|
and `cat` will be called again).
|
|
|
|
If SIGQUIT is ignored, SIGINT both exit the whole script.
|
|
|
|
I'm not sure this asymmetric behaviour is intended, but currently CRA
|
|
and BDE versions has is.
|
|
|
|
I think SIGTERM should exit the script in all cases immedeatly. CRA
|
|
version does this, BDE version ignores SIGTERM.
|
|
|
|
bash1 falls over completely on these tests.
|
|
|
|
bash2 works like CRA version.
|
|
|
|
Other tests:
|
|
------------
|
|
|
|
BDE version fails on 5, the shell terminated while the forground child
|
|
is running (like the unpatched FreeBSD sh did).
|
|
|
|
BDE version handles SIGQUIT seperately from SIGINT in test 1, and 3,
|
|
but not in test 4 and 5.
|
|
|
|
bash2 fails on 5, it gets a SIGQUIT that was sent to a
|
|
SIGQUIT-catching process after that process exits. Reliably, it
|
|
doesn't matter how much time was passed between SIGQUIT to process and
|
|
exit or process.
|
|
|
|
As Bruce noted, bash version 1 fails on test 2 like my version of sh
|
|
does. bash1 doesn't have the bash2 problem on test 5 (getting signals
|
|
that were caught by the child after the child exits).
|
|
|
|
zsh fails on test 1 and 5 in a similar way bash2 failed on 5: After a
|
|
child exits that received SIGINT, it terminates immedeatly.
|
|
|
|
I wonder how bash 2 manages to fail on test 5, but to pass test 1...
|
|
|
|
csh fails on test 1 in two ways:
|
|
- when the SIGINT/SIGQUIT-catching child receives both SIGQUIT and
|
|
SIGINT, csh exits immedeatly, while the child is still running.
|
|
- when only one or more SIGQUIT are delivered to the child, csh gets
|
|
it after the child exits and terminates with coredump.
|
|
|
|
csh also ignores SIGQUIT where SIGINT is caught.
|
|
|
|
test2.csh is different from test1.sh in that all signals should be
|
|
caught. csh fails in that it sometimes gets SIGQUIT and exits with
|
|
coredump (the shell, not the child).
|
|
|
|
|
|
Other stuff not directly relevant to these fixes:
|
|
=================================================
|
|
|
|
csh oddness:
|
|
------------
|
|
|
|
When the "catcher" program resets the terminal, there's a time-window
|
|
where C-g is mapped to SIGQUIT, but "catcher" already exited. In other
|
|
word, if you type C-g while in "catcher", then immedeatly C-d, csh
|
|
will get SIGQUIT. I thought that remapping the terminal keys should
|
|
have been done before the shell is re-entered. The same happens to
|
|
tcsh.
|
|
|
|
With sh variants I didn't manage to deliver SIGQUIT by C-g.
|