[Info-vax] "bad select 38" (OpenSSL on VMS)

[Stephen Hoffman]

On 2016-09-14 12:56:03 +0000, John E. Malmberg said:

> On 9/13/2016 10:32 PM, Neil Rieck wrote:
>> 
>> I discovered a piece of code (term_sock.c) ...
> 
> That method uses ASTs to have terminal input feed a TCP/IP socket so 
> that the actual select() or poll() can be used.  In this case, the 
> terminal input has already been read into an additional buffer.
> 
> ...


The select() call implementation has been broken — sorry, 
"feature-constrained" — for ~decades.   How long does a common C call 
need to remain broken before it can be fixed?

The dependencies are increasing in length and complexities, the work 
arounds more complex, the effort higher, the resulting end-user 
application code more convoluted and tricky, the more glue code, and 
the risks of bugs and security exposures increase, as does the 
development and maintenance costs.   From the vendor side, the scale 
and scope and speed of the product requirements and the features and 
enhancements is increasing, too.   Massively.

Fun times.

But I'm really getting tired of jackets and workarounds and everybody 
repeatedly porting over or reimplementing their own version of 
available open source.   Jackets all the way down.   it all makes for 
more costs and more work and more application source code and more 
complexity and more instability and more insecurity and more 
maintenance.

As for this "bad select 38" error, that's been hacked around (again!) 
in SSL1 1.0.1h, and s_client is now working.   Things are working 
again, at least until some other application smashes into select() or 
some other gap.



-- 
Pure Personal Opinion | HoffmanLabs LLC