[sane-devel] Using C++ in sane backends

[Olaf Meeuwissen]

Hi Povilas, list,

Povilas Kanapickas writes:

> On 2019-05-12 07:32, Olaf Meeuwissen wrote>> On 2019-05-09 11:56,
> Povilas Kanapickas wrote:
>>>> I would propose we allow C++, but that its use must not affect API or
>>>> ABI in any way. There should be a hard requirement that there must be no
>>>> visible changes outside a specific backend, except the need to link the
>>>> backend with libstdc++.
>>
>> Technologically, there is no reason why one cannot use another language
>> than C.  All the SANE API specification requires is that SANE backends
>> provide a certain C API.  What a backend does "under the covers" is that
>> backends business.
>>
>> I have implemented a third party SANE backend in C++ myself.  See
>>
>>   https://gitlab.com/utsushi/utsushi/blob/master/sane/backend.cpp
>>
>> Whether we want to allow C++ in the SANE Project's sane-backends (and to
>> what extent) is an issue that we need some consensus on, I think.
>
> Agreed. The reason why I thought that it would be possible to achieve
> some consensus at all was that there's no consistent design of the
> backends included within the sane-backends repository so they could be
> considered separate projects within a monorepo. In that case using C++
> in one backend is not a problem as most SANE developers will not need to
> interact with that code in any way.

ACK.

>>> Two benefits that would yield almost immediate improvements are the
>>> following:
>>>
>>>  - ability to use exceptions instead of manual error handling which
>>> currently often explodes the amount of code by 4 times.
>>
>> I was "afraid" of this ;-)
>>
>> Please be aware that your backend will have to catch *all* exceptions,
>> even the ones you don't throw yourself, e.g. an `std::bad_alloc`.  Any
>> uncaught exception will crash SANE frontends right then and there with
>> no chance of saving state.
>
> Agreed. My plan was to catch all exceptions, print something to logs and
> then return some error status code so that everything works just like if
> the C implementation got an error it couldn't recover from. The caller
> code should be completely isolated from what happens within the backend.

Good plan but might be harder than you think.  I had the same plan way
back when I wrote that utsushi SANE backend and learnt the hard way.
Especially the exception that was thrown *before* its sane_init() was
even called had me bang my head against the wall for a while.

Hope this helps,
--
Olaf Meeuwissen, LPIC-2            FSF Associate Member since 2004-01-27
 GnuPG key: F84A2DD9/B3C0 2F47 EA19 64F4 9F13  F43E B8A4 A88A F84A 2DD9
 Support Free Software                        https://my.fsf.org/donate
 Join the Free Software Foundation              https://my.fsf.org/join