[prev in list] [next in list] [prev in thread] [next in thread] 

List:       fossil-users
Subject:    Re: [fossil-users] db errors after Ctrl-C
From:       rosscanning () fastmail ! com
Date:       2016-12-14 5:28:10
Message-ID: 1481693290.449327.818390537.1B9FFA83 () webmail ! messagingengine ! com
[Download RAW message or body]


Unfortunately, the bug is a proper Heisenbug, and refuses to manifest at
all under the debugger.

I read up on the behaviour of Ctrl-C under Windows, here:
https://msdn.microsoft.com/en-us/library/windows/desktop/ms682541%28v=vs.85%29.aspx
and it's supposed to work like this:
1) If no handler is installed (as in Fossil), Windows spawns a new
thread within the process.
2) That thread calls ExitProcess()
3) ExitProcess() stops every thread except itself, and then does a bunch
more shutdown stuff.

Meanwhile, the code in Fossil that prints the "abandoning commit"
message does the following:
1) Calls fossil_fatal()
2)  fossil_fatal() prints the message, calls db_force_rollback(), and
then fossil_exit().
3) fossil_exit() calls db_close(), then exit()

So my wild-ass guess is that there is a race condition between the
shutdown sequence being executed by fossil_fatal(), and the shutdown
sequence being executed by the spawned thread. Maybe  the spawned thread
is closing file handles just before fossil_fatal() tries to access them?
That would explain the "invalid file handle" message that I sometimes
see.

This SO answer:
http://stackoverflow.com/questions/915312/different-ways-of-exiting-a-process-in-c
indicates that returning from main() is the preferred method of shutting
down in Windows, so I might see if I can hack Fossil to do that, and see
what happens.





On Wed, Dec 14, 2016, at 08:22 AM, Richard Hipp wrote:
> On 12/13/16, rosscanning@fastmail.com <rosscanning@fastmail.com> wrote:
> > SNIPPET 1
> > C:\Users\Ross\src>fossil commit -m "created"
> > ./math/src/transforms.rs contains mixed line endings. Use --no-warnings
> > or the "crnl-glob" setting to disable this warning.
> > Commit anyhow (a=all/c=convert/y/N)?
> >     *** I hit Ctrl-C here ***
> > Abandoning commit due to mixed line endings in ./math/src/transforms.rs
> > SQLITE_MISUSE: API call with invalid database connection pointer
> > SQLITE_MISUSE: misuse at line 138842 of [707875582f]
> > ^C
> 
> Can you run fossil in a debugger - set a breakpoint on the
> fossil_sqlite_log() function, then provoke the behavior above.  When
> you hit the breakpoint, capture a stack trace and post it to the list.
> That will probably make the problem obvious.
> -- 
> D. Richard Hipp
> drh@sqlite.org
_______________________________________________
fossil-users mailing list
fossil-users@lists.fossil-scm.org
http://lists.fossil-scm.org:8080/cgi-bin/mailman/listinfo/fossil-users

[prev in list] [next in list] [prev in thread] [next in thread]