Re: Pike 8.0: Sql.pgsql objects don't get garbage collected?

Jeff Hungerford wrote:
>Debug builds tend to abort with a core dump after a few minutes
>of testing; It looks like the "exception in destruct()" check 
>which is only enabled in debug builds gets triggered. (object.c:816)

This is one of the worst misfeatures of Pike; what is the reason
exceptions in destroy() handlers don't evoke standard exception
processing?  I have to revert to silly things as sprinkling in catch
statements (which only serves to hide problems).

Nonetheless, this bug should be fixed now.

>/home/hungerf3/git/pike/src/threads.c:2571: Fatal error:
>Mutex unlock from wrong key 0xf7 != 0x11dd618!
>Pike was in GC stage 400 when this fatal occurred.

I'm not even sure I understand what happened here.  Smells like a bug in
Pike.  The current patches might circumvent this bug so it doesn't happen
anymore.

>It looks like there is an issue with multiple threads trying to open a
>connection at the same time; I've seen the following often when
>starting up a bunch of test workers at the same time:
>-=-=-=-=-
>Attempting to clone an unfinished program
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/Sql.pike:342:
>    Sql.Sql()->create("localhost:5432","hungerf3","hungerf3","CENSORED",UNDEFINED)

Again, I'm not quite sure what happened here.  Anyone?
It's unlikely that the last patch will fix this one, though one could hope.

>Lookup in destructed object.
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql.pike:604:

Fixed.

>Unknown program: destructed object->function()
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql.pike:627:

Fixed.

>Indexing the NULL value with "i".
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql.pike:635:
>    pgsql(hungerf3@localhost:5432/hungerf3,0,24878)->procmessage()

Fixed.

>Indexing the NULL value with "trylock".
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql_util.pmod:175:
>    Sql.pgsql_util.bufcon(0 bytes, read=[..-1] data=[0..-1]
>    free=[0..0] string
>    )->sendcmd(1,UNDEFINED)
>/usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql.pike:1165:
>    unknown function()

Probably fixed.

>Indexing the NULL value with "trylock".
>     /usr/local/pike/8.0.175/lib/modules/Sql.pmod/pgsql_util.pmod:257:
>     conxion  fd: -1 input queue: 0/224 queued portals: 1
>     output queue: 0/224
>     ->start(UNDEFINED)

Probably fixed.


>And one random error with no useful info as to from whence it came:

>Attempt to call the NULL-value
>Unknown program:
>    0(conxion  fd: -1 input queue: 0/224 queued portals: 1  output

Most likely fixed as well.

>Let me know if there is any more info I can provide.

Best would probably be to grab the latest version and see if you can still
break it.
I changed the destruction of objects to synchronise with the natural extinction
of the threads.  This might/should fix all the accessing-member-in-destructed-
object occurrences.
-- 
Stephen.