Slashdot Mirror
Coder Accuses IBM of Patenting His Work
ttsiod writes "Back in 2001, I coded HeapCheck, a GPL library for Windows (inspired by ElectricFence) that detected invalid read/write accesses on any heap allocations at runtime — thus greatly helping my debugging sessions. I published it on my site, and got a few users who were kind enough to thank me — a Serbian programmer even sent me $250 as a thank you (I still have his mails). After a few years, Microsoft included very similar technology in the operating system itself, calling it PageHeap. I had more or less forgotten this stuff, since for the last 7 years I've been coding for UNIX/Linux, where valgrind superseded Efence/dmalloc/etc. Imagine my surprise when yesterday, Googling for references to my site, I found out that the technology I implemented, of runtime detection of invalid heap accesses, has been patented in the States, and to add insult to injury, even mentions my site (via a non-working link to an old version of my page) in the patent references! After the necessary 'WTFs' and 'bloody hells' I thought this merits (a) a Slashdotting, and (b) a set of honest questions: what should I do about this? I am not an American citizen, but the 'inventors' of this technology (see their names in the top of the patent) have apparently succeeded in passing this ludicrous patent in the States. If my code doesn't count as prior art, Bruce Perens's Efence (which I clearly state my code was inspired from) is at least 12 years prior! Suggestions/cursing patent trolls most welcome."
FWIW, your site wasn't cited by IBM's application, but by the USPTO examiner.
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=1&f=G&l=50&co1=AND&d=PTXT&s1=7552305.PN.&OS=PN/7552305&RS=PN/7552305
No kidding!!! What do you say at this point?
Unless you sue IBM and have very deep pockets once someone has stolen your idea and had the theft accredited by the rating agency (patent office) you are essentially screwed. You may file an appeal, but then the burden of proof is on you.
Tesla and Marconi had similar ideas around the same time but the patent office ruled in favor of Marconi, you my friend may be a Tesla at this point.
"The shortcomings of the prior art are overcome and additional advantages are provided through the provision of a method of facilitating management of dynamically allocated memory. The method includes, for instance, having a dynamically allocated memory buffer; and determining in real-time that an invalid access of the dynamically allocated memory buffer has occurred."
I'm no expert on HeapCheck but that doesn't sound far removed from its basic functionality.
No kidding!!! What do you say at this point?
As with all patents read Claim No 1. It should contain the novel element, you cannot leave it for later. The "novelty" in this one is that you do not need to recompile the program the way original efence worked.
However, times have changed since the days of the original Efence and things are linked dynamically at runtime on all OSes which in turn means that any LD_PRELOAD-ed Efence like debugging library which relies on OS R/W page management to control access is prior art. It satisfies literally the requirements of claim 1.
I do not recall off the top of my head what valgrind uses, but I would not be surprised if it fits the bill. The original Efence does not - it has to be linked in.
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
Back in 1990, I redid the Borland Pascal memory allocator so each block was given its own hardware-protected segment descriptor and length. Worked magnificently, as any reference outside a valid block would immediately fault. Only prob, you could only allocate about 4000 blocks as there were only 4,096 entries in the hardware segment table. So the next refinement was to allocate each block with short pre and postambles set to $12345678 and check thee for overwriting periodically. Worked almost as well, if not so immediately finding the errors.
And no, I did not try to patent this, as I knew the Burroughs machines, since 1961, allocated a fresh memory-protected segment for each array, and using pre and post safety zones sure sounded like an "obvious" thing to do..
Pretty sure the patent office is accepting applications for examiners right now, lol. They like people with Master's degrees.
I am busy this week and have not been able to look at the patent. However, if IBM referenced this person's work in the prior art section of the patent, they are admitting that his work IS prior art, and is not subject of the patent. The issue then becomes, do any of the PATENT CLAIMS (not the summary) claim art that is exercised in his program.
Electric Fence is listed as prior art in two ATT patents. When they saw it, they refrained from including two claims they otherwise would have, because I had precedence. The ATT patents should have expired by now.
I doubt there's anything to worry about in this specific case. However, software patents in general present a severe problem. [Oops - previously submitted this as AC. Sorry]
Bruce Perens.
Actually in another post by Sockatume pointed the USPTO examiner is the one mentioning his website and not IBM.
(for some reason I cannot link the article)
I only take an issue with your last paragraph - I certainly didn't leave the code in a graveyard, I released it with an open source license, so that anyone could use it and make his life a bit better. The fact that 10 years later, someone else has now patented the ideas in Electric Fence and HeapCheck, and can now sue me and everyone else using it, is what got me mad (hence the "trolls" comment).
Not at all. It is standard in the patent world to make the thing that you actually want to patent be the middle claim. The reasons are many and varied, but you can jump right to the thing that any company actually wants to get the patent upon by checking the center claim. There are 20 claims here, so it is likely this one: 10. The method of claim 1, further comprising: detecting that another allocation mode is to be used to allocate one or more buffers for the process; and dynamically deactivating the allocation mode that enables the determining in real-time, wherein the dynamically deactivating includes turning off an indicator in the process to deactivate the allocation mode, and wherein another dynamically allocated memory buffer for the process is allocated based on the another allocation mode, said another dynamically allocated memory buffer being allocated without additional memory as a guard.
Before spouting off about how the USPTO sucks so much ass solely on the basis of what the OP says, first try a little independent research. You can look up the entire prosecution history of the application in question on the USPTO's website at Public PAIR.
First off, the guy's website was cited by the examiner rather than by IBM. The examiner used the OP's website to reject the claims three times before IBM's attorney finally came around and amended the claims to include this:
wherein setting the allocation mode for the process to enable determining in real-time an invalid access is performed in real-time, and wherein the setting sets the allocation mode for an application executed by the process without requiring recompiling, linking or loading of the application to set, in real-time, the allocation mode for the application
This was cited by the examiner in the reasons for allowance. Does the OP's code do this?
Once again, this is an example of people who know very little about the patent system reading the abstract or looking at the figures, and not understanding that the really important part of the application is the claims. Nothing to see here, move along.
Nothing in your explanation describes why such simple coding (for someone with experience) should be patentable. The USPTO does "suck ass" because they allow patents that are pretty straightforward for any skilled practitioner.