[cl-plus-ssl-devel] Possible bug in test-https-server (2008-11-04)

[John McAleely]

[repeat for the list. apologies David for the duplicate]

On 21 Jan 2009, at 13:25, David Lichteblau wrote:

> Quoting John McAleely (john at mcaleely.com):
>> Ah. When I tried bundling CL+SSL into the webserver I'm using
>> (portable aserve), I see the same error symptom. What would be the
>> best way to remove the BIO entirely? I've tried a couple of things,
>> and both failed to remove the BIO - I was seeing the code try to call
>> BIO_xxx functions and then failing.
>
> If your code is really using the Lisp BIO (not the FD BIO), i.e. if it
> calls make-bio-lisp-method, it would mean that :unwrap-stream-p is  
> false

Currently all code paths appear to call make-bio-lisp method via  
ensure-initialised (in two possible places, in fact!). Commenting  
these out results in a cl+ssl that fails (in the same unwrap-strategy  
nil tests, and my portableaserve installation) with an error:

aborting test thread echo server for strategy test: value NIL is not  
of the expected type MACPTR.

I assume something is still trying to use the Lisp BIO. Hence my  
question about how to remove it entirely.


>
> or that CL+SSL didn't know how to get an FD from the stream.
>
> (One reason could be that the stream argument is actually a gray  
> stream.
> Since portable-aserve has a hack for bivalent streams, it would be
> important to set up the CL+SSL socket stream from the real-socket,  
> then
> wrap it in a CLOS stream, not the other way around.  But this may be
> obvious to you.  Hard to say without seeing the code.)


I had assumed that once I had a working cl+ssl installation I would  
need to study how to construct the streams in portable aserve  - this  
description has saved me some effort. Thanks!

However, while the cl+ssl test suite, and portable aserve, fail *in  
the same way* I think I've not reached this point yet.

Anywhere else I should be commenting out attempts to use the BIO?

J