Yet another bugfix committed to https://github.com/ChrisMoller/edif.git
Mostly, what it does is detect multiple ⎕fx invocations of libedif2.so
and short-circuits them, leaving the inotify() process intact and the
message queue opened only once. It also detects )clear and
closes/unlinks the queue, kills the inotify() process, and closes any
open editor processes.
Chris
On 20/08/18 10:38, Hans-Peter Sorge wrote:
Hello Chris,
The session stays blocked - what ever I try fiddling around.
During this fiddling I came across on other situation that makes the
edif2 case even more interesting:
(EDIF2 is a work space having edif2 as function only)
The first session (eg. after reboot) has a double free corruption in
either case.
1) 'libedif.so' ⎕fx 'edif2' .. 'libedif.so' ⎕fx 'edif2' .. )off
or
2) )load a WS having function edif2 then 'libedif.so' ⎕fx 'edif2' ..
)off
[joy@joyw520 ~]$ apl
'libedif.so' ⎕fx 'edif2'
edif2
)clear
CLEAR WS
)load EDIF2
SAVED 2018-08-20 16:03:15 (GMT+2)
'libedif.so' ⎕fx 'edif2'
edif2
)off
double free or corruption (fasttop)
Abgebrochen (Speicherabzug geschrieben)
The following case is not expected. Yet one could argue )save does
some cleaning.
[joy@joyw520 ~]$ apl
'libedif.so' ⎕fx 'edif2'
edif2
)save EDIF2
NOT SAVED: THIS WS IS CLEAR WS
)drop EDIF2
2018-08-20 16:10:32 (GMT+2)
)save EDIF2
2018-08-20 16:10:36 (GMT+2)
'libedif.so' ⎕fx 'edif2'
edif2
)off
Goodbye.
Session duration: 133.863 seconds
But now, on a naked WS, the corruption is gone .... sofar I could not
repeat a subsequent corruption until after a reboot!
[joy@joyw520 ~]$ apl
'libedif.so' ⎕fx 'edif2'
edif2
'libedif.so' ⎕fx 'edif2'
edif2
)off
Goodbye.
Session duration: 6.48602 seconds
Best regards,
Hans-Peter
Am 20.08.2018 um 05:26 schrieb Chris Moller:
Hi, Hans-Peter,
That quirk where you quad-fx edif2 twice...
When edif2 starts, it fork()s a process to watch for changes in the
working-file copies of the functions passed to emacs, and it also
opens a POSIX message queue so the monitor process can communicate
changes to APL. When you quad-fx-ed edif2 twice, all that was
duplicated--two monitor processes, another sender on the queue, and
more signals being emitted. I've put a patch in (but haven't
committed it yet, and it should fix the segfault) that prevents all
that, but my question is whether the "session blocked" bug happens
only after the second quad-fx. I just can't reproduce the blocking
bug here, so I'm looking desperately for some non-obvious cause.
Thanks,
Chris
On 08/18/18 15:29, Hans-Peter Sorge wrote:
Hello Chris,
I have just one process running.
I recompiled and reinstalled edif to make sure it's the most recent
version.
During editing with emacs:
session is blocked, then in APL session
Ctrl C
fg
'some data' ENTER - no response
in emacs: save
in APL session:
'some data' replayed
'some data' response
---------------------------------
There is on other quirk too:
....
'libedif.so' ⎕fx 'edif2'
edif2
'libedif.so' ⎕fx 'edif2'
edif2
)off
double free or corruption (fasttop)
Abgebrochen (Speicherabzug geschrieben) [segmentation fault]
Best regards,
Hans-Peter
Am 18.08.2018 um 02:39 schrieb Chris Moller:
I can't make it fail, but I'll look closer in the morning.
Is there any chance you have multiple APL processes running
concurrently? Apparently, that causes problems I hadn't caught before.
Chris
On 17/08/18 17:14, Hans-Peter Sorge wrote:
Hello Chris,
edif2 blocks the apl session again.
edif: latest git.
apl: 1069M
Best Regards,
Hans-Peter
Am 16.08.2018 um 21:31 schrieb Chris Moller:
Not that I expected otherwise, but edif and edif2 still work as
expected.
cm
On 16/08/18 15:14, Juergen Sauermann wrote:
Hi Hans-Peter,
thanks, hopefully fixed in *SVN 1069*.
Best Regards,
/// Jürgen
On 08/16/2018 07:29 PM, Hans-Peter Sorge wrote:
Hello,
The ⎕IO bug is back.
(note: ⎕IO ←→ 0 seems to be set to 0 using edif2)
∇qio_test[⎕]∇
∇
[0] qio_test
[1] ⎕IO
∇
⎕IO←1
⎕IO
1
qio_test
1
⍎¨ 'qio_test' 'qio_test'
1
1
⎕IO←0
⎕IO
0
qio_test
0
⍎¨ 'qio_test' 'qio_test'
1
1
Greetings
Hans-Peter
Am 14.08.2018 um 18:17 schrieb Chris Moller:
I just committed a minor patch to edif, fixing a relatively
low probability bug involving a possible message queue name
collision if you have two or more APL sessions open
simultaneously.
https://github.com/ChrisMoller/edif
--cm
