Re: [Jack-Devel] ladish + jack autoconnect

On 01/07/2012 09:43 PM, Fons Adriaensen wrote:

Hi!

>> Stephane, Fons et al may provide some [constructive] criticism once the
>> detailed proposal is out.
>
> Constructive or not, this is what I think about the proposal.
>
> I'm _not at all_ in favour of adding functionality to Jack
> that only serves to resolve problems created by application
> authors who either by ignorance or by design do things they
> shouldn't do, such as autoconnecting without providing an
> option to configure or disable it.
>
> Doing this only encourages the creation or continued acceptance
> of apps that don't behave as required. The correct course of
> action is to discourage the same, and force app authors to
> either do their homework or face being excluded.

While I think you clearly have a point here, the discussion reminds me
about the glibc memcpy() bug on newer Intel processors: copying
backward turned out to be faster on some architectures, and for the
sake of efficiency, glibc changed from copying forward to copying
backward on these machines.

Though memcpy's manpage is absolutely clear that source and target
memory areas must not overlap, some application developers did not worry
about it. Overlap was no problem as long as memcpy() was traversing from
low to high addresses. Once changed, many applications broke, including
flash.

There was a big discussion and blaming of application developers, Adobe
in particular, that memmove() should have been used as mentioned all
over the place.

Linus' position was rather clear: the user doesn't care who broke it,
the user demands a working system. As simple as that, end of the story.


It's exactly what we're facing here. We could probably continue to argue
purity vs. pragmatism for weeks.

How about the following: let's do both - work around the bug to please
the user *AND* blame the application developers.

For this to work, I suggest to emit a (rude) warning every time an
application tries to self-connect (its output ports) if jackd is running
in one of the "compatibility modes" (a.k.a. ladish mode).

Something like:

    "Application foo attempted to self-connect its ports. See
     http://some.url.org/faq/... why this is bad and update your
     application to a fixed version."

If there are really many broken applications out there, users would
hopefully bug the upstream authors until they either do something about
it or prove Nedko wrong. ;)


Just my €0.02

1325980880.27984_0.ltw:2,a <4F08DCC2.2080202 at drcomp dot erfurt dot thur dot de>