document that we do machine arithmetic

[StefanKarpinski]

https://groups.google.com/d/topic/julia-users/esqGRbQdodM/discussion
https://groups.google.com/d/topic/julia-dev/1jrQtu-i8Tk/discussion

In particular this:

https://groups.google.com/d/msg/julia-users/esqGRbQdodM/QZ9cAe5YVecJ

[JeffBezanson]

Fixed in 016823a

[waldyrious]

Wouldn't it be relevant to also add some notes on why machine arithmetic is used? Specifically something along these lines (with some polishing):

Julia is able to produce code that runs faster in part because Julia represents problems using units of computation like integer and floating point arithmetic that are hardcoded by modern chips. Lots of other dynamic languages are less efficient because they invest a lot of computational resources in hiding the underlying machine from the user. Julia's approach is very different from what other dynamic languages take, and is much closer to C. C, Assembly, etc. don't check memory registers for overflow after every operation. Julia promotes manual checking after certain operations that the programmer, not the language, deems "suspect". This is exactly how languages such as Assembly and C have done things for decades.

[johnmyleswhite]

+1 for this text

-- John

On Apr 18, 2013, at 7:46 AM, Waldir Pimenta [email protected] wrote:

Wouldn't it be relevant to also add some notes on why machine arithmetic is used? Specifically something along these lines (with some polishing):

Julia is able to produce code that runs faster in part because Julia represents problems using units of computation like integer and floating point arithmetic that are hardcoded by modern chips. Lots of other dynamic languages are less efficient because they invest a lot of computational resources in hiding the underlying machine from the user. Julia's approach is very different from what other dynamic languages take, and is much closer to C. C, Assembly, etc. don't check memory registers for overflow after every operation. Julia promotes manual checking after certain operations that the programmer, not the language, deems "suspect". This is exactly how languages such as Assembly and C have done things for decades.

—
Reply to this email directly or view it on GitHub.

[jiahao]

Can users check for overflow explicitly in Julia? I can understand if the default behavior is to not check explicitly, but it would be very important to check for some applications. Is there an overflow flag that is set that can be tested for, without some clunky test like checking if a positive integer suddenly turned negative? I see some people mentioning this but didn't see a clue of whether such a thing existed.

[StefanKarpinski]

Open issue for arithmetic overflow checking: #855.

[StefanKarpinski]

I like @Waldir's text above. What's the source of the text?

[JeffBezanson]

Performance is really the only reason. It's not that we think tediously manually checking everything is good. In fact we will provide options to do more automatic checks.

[jiahao]

@StefanKarpinski The first part appears to come from @johnmyleswhite in the first discussion thread you posted.

I've added a paragraph to similar effect to the beginning to the numbers section, but am wondering now if the specific sentence about "hiding the underlying machine from the user" should have been put in as well.

[waldyrious]

What's the source of the text?

It's a shortened version of a selection of quotes from this thread, which I originally posted here. These are the same links as those you used in the opening comment of this issue.

[andrioni]

If a overflow flag is to be developed, it'd be good if it were extensible enough to also work with the MPFR flags. I'm trying to think of a good way to expose them in BigFloat, but have yet to find it.