Re: patenting compiler technology

On Wed, 6 Jan 2010 06:30:37 -0800 (PST), rcmetzger
<rcmetzger@grandecom.net> wrote:


>http://www.google.com/patents
>search for "optimizing compiler"
>382 patents identified
>
>the first 200 are all relevant, and with a couple of exceptions,
>all were granted between 1990 and 1999
>[My, that's a lot. isn't it. It doesn't tell us how useful they are,
>but since any patent granted in the US before 1993 has how run its
>course and expired, it does tell us that they didn't destroy the
>software industry. -John]


IANAL


I think it's telling that many of the holders of general algorithm
patents have apparently chosen not to enforce costly licensing
agreements for their technology ... I have family members who practice
IP law and routinely refer me to court cases and PTO directives
regarding software, yet I know of few cases involving any general
algorithm patents. Though there certainly have been some algorithm
centric court cases, most cases tend not to be about the uses of the
algorithms themselves, but rather are about some higher level device
or process that happens to incorporate them.


In the US, patent examiners are required to grant applications unless
they can find conflicting prior art. For quite a long time the PTO
examiners had no good databases for researching software art (IEEE,
ACM and others have worked very hard to remedy this). A number of
early software patents have been overturned on reexamination because
prior art was shown to exist that the examiners were unaware of at the
time. Of course, the applicant is supposed to do due diligence[1] and
refer to relevant prior art in the application. The examiner's search
is cursory at best ... if unmentioned prior art can't be found in an
hour or so, the patent is likely to be granted (unless the examiner
can't understand it).


One of the basic precepts of patent law is that the granted monopoly
should be of material value to the holder. An undefended patent is
little more than very expensive wallpaper. Also, US law technically
forbids pure algorithm patents although the USPTO unilaterally ruled
that data transformation counts as a "physical process" under the law
and grants algorithm patents under that decision. (I don't happen to
agree but what can you do? Congress collectively has the intellect of
a philodendron - I don't expect any rational revisions to patent law
any time soon.)


While there may be some altruism at work - the patent does give credit
to the inventor and, in some cases, maybe that's all that is wanted -
it does make one wonder whether large companies with vast portfolios
of undefended algorithm patents do not enforce them because know they
scammed the PTO the first time around and that their patents would not
stand against a court challenge or a PTO reexamination.


IANAL


George
[1] "Doesn't matter how slow you say it, Will. It's still doo-doo,
isn't it?" -- Harlin Polk (Will and Grace).

Return to the comp.compilers page.
Search the comp.compilers archives again.