[Libreoffice] minutes of tech. steering call ...

[Stephan Bergmann]

On 10/21/2011 11:08 AM, Kevin Hunter wrote:
> At 4:11am -0400 Fri, 21 Oct 2011, Stephan Bergmann wrote:
>> On 10/21/2011 02:06 AM, Kevin Hunter wrote:
>>> 5. That API definition will be a *lot* of work, but hopefully somewhat
>>> thought out already through only a mild reengineering of the current
>>> binary API.
>>
>> The UNO API is already there. Or what do you mean?
>
> I was talking about an API that is not dependent on an ABI. But I freely
> admit I know very little about ABIs, so I may have just conflated that
> term. See below.
>
>>> The upside is that if we're talking a major version change, /now/ would
>>> be the time to do this.
>>
>> A downside is that you would still need to maintain (and build!) the UNO
>> runtime for the MSVC ABI.
>
> This may be the crux of what I'm not getting, but why? Why can't a
> protocol be, say, text-based via (local, or other) socket? In my mind, I
> see two independent programs, from two different compilers, using the OS
> and something akin to pipes to communicate. I admit it might a smidgen
> slower to do it that way, but do people actually use LO in HPC
> scenarios? (And I fully accept that they might, I just haven't seen it
> yet in my various interactions.)

That's all already there with UNO.  Only, for any code to make use of 
that, it needs to talk with "bridge" code that handles the (intra- or 
inter-process) communication.  That bridge code (which is necessarily 
ABI-specific) is also already there.

The only thing is that, if you wanted to give up building LibO with MSVC 
and switch to MinGW, but wanted to retain the MSVC-specific bridge code 
(so that old extensions can continue to run, in- or out-of-processes), 
you could not give up building LibO with MSVC completely, because you 
would still need to build that bridge code with MSVC.

Designing a new communication protocol would not buy you anything.

-Stephan