Re: FW: [vhdl-200x] 1076 & Entity balloting

Before getting swamped in the details, it's important not to lose sight
of the big picture. The FVTC took 4 years of technical activity to
create the assertion language that Accellera wanted, before Accellera
decided, for what appear to be political reasons, that the donation of
one of its sponsors (which had already essentially been rejected by the
FVTC) was more important than the work of its own technical committee.

I'm not now, and I have never, said that this is either a good or a bad
thing. It's simply how vendor-driven standards are created. My point is
that we, in the IEEE, should not be going down this road. If it's
possible to carry on using a process in which individuals have a
technical input and a vote, then that's what we should be doing. If, on
the other hand, we end up in a situation where those individuals donate
their time and expertise and end up without a vote, then they will very
quickly come to the conclusion that they are being used, and that they
are wasting their time.

Bailey, Stephen wrote:

> There were many people that thought that a single assertion/property
> language would be best and sufficient. The alignment process that
> the Accellera Board "dictated" was directed to get SVA and PSL closer
> to the ideal of a single language. For various reasons (some
> technical and some related to syntactic sugaring), one language was
> not the result.

I diasgree with the reasons you give: I think the primary reason was
simply political. Why didn't the sv-ac just go with PSL? Why did they
instead vote to accept the OVA donation, when the FVTC had clearly done
much more work than the sv-ac, and when the FVTC had already rejected
ForSpec?

> So, if organization entity membership is the evil that this example was meant to demonstrate, why:

I'm not saying it's evil; I'm saying it's different.

> 1. Isn't PSL simply SVA? (Or vice versa)

In fact, PSL1.1 and SVA are nearly the same. Complete alignment wasn't
possible because PSL and SVA have two different goals: PSL is
stand-alone, SVA is embedded within SystemVerilog. The final report of
the alignment committee states that 'It is our opinion that, as of
January 2004, we have achieved as much alignment as is possible given
the differing objectives of the two languages'.

> 2. Why did the Accellera board approve PSL 1.01 and 1.1?

If memory serves me correctly, they delayed approval of 1.01 as long as
possible, until it was clear that it was already obsolete because it no
longer had the support of the board, and because it was to be replaced
by 1.1.

> 3. Why didn't the Accellera board dissolve the FVTC or instruct the
> SV working group to narrow its scope to not include assertions?

They couldn't dump the FVTC after 4 years of work because they would
never have lived it down. Instead, they just told it what conclusions
were necessary.

> 4. Why has Accellera donated both SV and PSL to the IEEE? It could
> have easily squashed one or both.

SVA is embedded, PSL is stand-alone so, at the end of the day, they
couldn't dump one. On the other hand, PSL has a SystemVerilog flavour,
so perhaps they could have dumped one.

> If there's any lessons to be learned by the Accellera/FVTC experience
> it is that all organizations with a vested interest need to
> participate and have their voices heard. If 4 companies failed in
> one domain and succeeded in the other, their success was only
> possible because the organizations responsible for their failure in
> the one did not participate in the other.

I would suggest that any company which participated in the FVTC would
have had a reasonable expectation that the outcome of the FVTC's
deliberations would be the 'standard' Accellera assertion language. Why
would these companies have had an interest in also participating in the
SystemVerilog effort? Surely they would have expected their work to
appear in SystemVerilog anyway? Do you think the goalposts might have
been moved, following the OVA donation?

> If you think individual membership eliminates this problem, you are
> very naïve. As I have said before, it is currently easier for an
> organization to dominate the outcome of a WG by purchasing the $40
> membership of enough of their employees. And, individual membership
> makes it easier to camouflage what is happening as the assumption is
> that individuals are making their own decisions and not just carrying
> water for their company.

To solve the problem you need transparency and a well-defined agenda. As
it happens, I don't think that the current system is the best way to
achieve this. However, I do think that a move to entity voting, with
individual technical input, is not an advance. I've also brought up the
issue of entities flooding the DASC with their own employees, but there
was little interest in this on the DASC list.

> Worse, individual membership can result in individuals having agendas
> with no market relevance or negative market relevance. Well meaning
> engineers can be notoriously stubborn in holding out for the holy
> grail of technical solutions no matter the cost.

Agreed. And, even worse, individuals with a weak technical background
can make the process meaningless.

> In conclusion, your example does not prove that organizational entity
> membership is inherently inferior to individual membership.

I agree entirely; but my point was that they're *different*, and are
driven by different agendas, not that one is better or worse. This is a
point I've made several times in the other thread (on the DASC list?).
The vendors already have their own standards organisation; let's not let
them muscle in on ours. The membership issues have to be fixed, but we
can't do this simply by moving to entity participation.

In fact,
> the results from Accellera prove that organizational entity
> membership can result in coordination of standards and different
> standards that overlap in scope. It also shows that while a single
> entity (or even 4 like-minded entities) can influence, they cannot
> dictate the outcome.

I have to disagree with that. The whole history of PSL, SVA, and
SystemVerilog shows, if nothing else, that individual vendors can
completely control the outcome of a standardisation process.

Evan Lavelle
Received on Thu Jun 24 02:50:17 2004