Sure, the numbers are rationals, but the *arithmetic* is lossy: every full-precision floating point addition throws away one (ok, a half) bit of the result and every multiply throws away *half* of the result bits. Sure, you can find some specific examples where those bits were zero anyway, but that's not the way to a well-rounded understanding of what's going on. Better to understand the implications of rounding and precision, and use algorithms that operate appropriately in the presence of that noise injection.
A couple years ago I wrote an extensive fixed point library and one of the nice things was it at least threw away the LSbits in a multiply. I spent a lot of my misspent youth in central america. The ancient Mayan math was done with rationals most of it using 5 digits of base 20 numbers.
They had some interesting rational algorithms that I ran into from time to time especially related to planet orbital calculations.
The difference is, Chris, you have a position as a representative on a c standards (sub)committee. Your admission about participating on one extreme end of the spectrum is at least granted to be a frank one, if dishonorable to that public role. You don't have to be unbiased, but your public position should be or it undermines the rest. Just as a judge must at least appear to be if the system is to hold the respect it must have. The loss in confidence among some of us in the application developer space has done the organization little good.
I represent myself on standards committees. No one else. I do not represent any committee when posting here.
I am not extreme. You should here the people who don't post here. :-)
It is not a "public role" and you clearly have no understanding of what you are talking about. However you do seem to resort to personal abuse quite readily instead.
Undermines who?
The standards bodies are made up of businesses with a commercial/ business view to support. There are no truly independent and unbiased members of any standard that I know of. This is the same for the vast majority of standards. (I would say "all but there is bound to be an exception)
I am not sure anyone actually cares. The standards are not done for you.
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
Yes I did post that.. I was not sure given the complete lack of response to saying that the commercial compiler vendors have a shared code base in their compilers. That is across many vendors.
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
No, no - everyone else is at least partly mistaken. When people write opinion pieces, they are bound to differ occasionally from the facts I write.
There is also a place for open development, and closed development, and open testing, and closed testing.
You kid. Making the sources of their avionics software available to anyone will not make it unsafe - unless you are suggesting that avionics software typically contains such serious security flaws that "bad people" could disrupt its operation if they saw the code. If that were likely, then the developers should be *forced* to reveal their code - and forced to suffer the consequences of losing their contracts, validations, certifications, reputations, and whatever else they may have of value. You write security-critical code on the assumption that the "bad people" have read access to the source code.
Fairy 'nuff.
That's certainly true - the voting machines are only a minor point in the gap between US elections and actual democracy (here's a hint for the uninitiated - a fair democratic election requires the same rules for everyone, not different machines and different laws for different people. It also requires that everyone's vote counts equally - that means that everyone votes at the same time, and *no* votes are counted or even predicted until all voting is closed).
*Anyone* with physical access to the machine (such as election workers, and independent observers - preferably from other countries) should be able to verify the binary. In fact, I believe that any voter should be able to do this too. It could be done in a simple process - the person checking the software would enter a phrase of their choosing. The software would use that to initialise the shift register of a very wide CRC check (or other hash algorithm), and run its binary through the CRC, and print out the checksum. The voter could download the source code from the manufacturers, compile it using the compiler version used by the manufacturers (which would be a specified version of gcc, letting people check the code there too if they want), generate an identical binary himself, and do the same checksum at home. Of course, that would only verify that the expected binary is on the machine, not necessarily that it is the code that is running - somewhere along the line, you have to trust the election officials.
Actually uploading new software to the machines should only be possible for independent observers, preferably from another country.
Alternatively, they could just apply the same rules and rigour that the state of Nevada does when certifying gambling machines.
Perhaps people are trying to limit the number of times they run round the same circles without making any progress? However, some of us never learn...
Yes, I think most of us know that closed source vendors regularly use code licensed from other places (just as open source developers use other open source code). That's also one of the reasons why I (and, I think, you) think it is unusual that ByteCraft are willing to give customers access to the source code of the compilers (in appropriate circumstances) - most tool vendors (and any other closed source vendor) could not do that even if they wanted to, since they don't have the relevant rights to code they use. They probably don't even have access themselves to all the code used in their tools (as a trivial example, they won't have access to the code for all the windows dll's that are linked in).
Not sure on that. I think they could if they wanted to.
That is not what I was thinking. I was actually referring to the compiler itself not the surrounding IDE and DLL code.
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
Yes there is. To start with FP values are normally stored as values from 1.0 to less than 2.0, times a multiplier in the form of a power of 2. These have no exact representations for any fractional decimal values with a divisor that is not a power of 2. Secondly, the input expression is normally in the form of a value and a decimal exponent. That cannot represent any fractions with denominators not containing the factors 2 and 5.
The FP values can represent any value WITHIN A SPECIFIED ACCURACY. In C, examine float.h for the values of FLOAT_EPSILON and DOUBLE_EPSILON. These specify the accuracy of that number between
1.0 and 2.0.
VARIATION: Some FP systems do not use powers of 2 as the multiplier. This only changes the details.
--
[mail]: Chuck F (cbfalconer at maineline dot net)
[page]:
Try the download section.
Disagree. As long as he is biased, he should not hide it. That way we KNOW what to expect from him on standards organizations, and can accept or reject those positions.
I have also revised the subject line on this message.
--
[mail]: Chuck F (cbfalconer at maineline dot net)
[page]:
Try the download section.
I agree here, but you have some difficulties in your proposal. There is no way to guarantee the software compiles to the same binary, without duplicating everything, including the machine, compilers, linkers, operators, optimization, libraries, etc.
Again, the subject has been changed to separate the threads.
--
[mail]: Chuck F (cbfalconer at maineline dot net)
[page]:
Try the download section.
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
--
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
\/\/\/\/\ Chris Hills Staffs England /\/\/\/\/
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
No, you don't. Of course, not. I don't imagine you do. But you not surprisingly missed my point, then. Trust is very important and people in important roles of trust must operate so as to maintain that appearance, if not in actual fact. Trust is hard to build, easy to destroy. What you do represent here as your own opinion would at least give me significant pause were I to consider the idea of paying money to represent some open source group or compiler tool. You undermine that sense of good faith and trust.
But you wrote, "That was my point. They are no more neutral than I am.." You are referring to FreeRTOS's point about extreme views and basically admitting your relative position. Now you want to claim you are a moderate on the subject of open source and GCC?
I will allow others to decide that. I have come to my conclusion.
Well, I suppose it's good in this particular instance that he is so out in the open. It may explain some things to me that had remained obscure before and I'm glad to have a more realistic context. But he is a very visible antagonist and the fact that he remains despite it does injury to the faith and credit of the organization itself.
Well, I guess I had a higher level of respect before, that that. Sincerely. It's changed and I suppose having a more realistic picture is better than being overly idealistic. I'll grant you that.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.