Re: guile threading deadlock

[Linas Vepstas]

Hi,

2008/11/9 Ludovic Courtès <address@hidden>:
> "Linas Vepstas" <address@hidden> writes:
>
>> However, there is a simple solution: wrap every call
>> to scm_c_eval_string() with a call to scm_with_guile().
>> Doesn't seem very optimal or pretty, but it works.
>
> Cool.

Well, not quite. I also discovered that I have to re-define
smob-based functions in every new thread. That is,
I have to call scm_c_define_gsubr() in each thread
that I plan to use; simply calling these once  in some
parent thread results in

ERROR: Unbound variable: whatever-xxx
ABORT: unbound-variable

in the other threads.  I find this behaviour "surprising";
is there another spot of documentation that I should have
seen, that explained this, or is this a bug, or is this
working as designed?

> Again, it *is* documented, but maybe not sufficiently prominently, or
> maybe it could be repeated in other places of the manual.  Can you say
> specifically how you'd like to change that?

Haven't I answered this question three times now?

I feel that this conversation is not moving forward, I am
not sure why, or what i can do to be more helpful.

> Please send patches to <address@hidden>.

I'll try to do so in the next few says.

> If they are non-trivial,
> you'll need to assign copyright to the FSF, but we can work this out
> off-line in due time.

I have a generic assignment form with the FSF.

--linas