Microsoft: Excel 2007 bug is skin deep

A bug in Excel 2007 reported earlier this week may be ugly, but it's only skin deep, Microsoft said late Tuesday.

With the bug, results that should have been shown as 65,535 showed instead as 100,000, and a related problem cropped up with 65,536, Microsoft's David Gainer said in a blog posting. But the problem affected how Excel showed the number, not in what the spreadsheet software actually "knows."

"This is an issue in a function that puts numbers in cells, so the values in Excel's memory are actually correct," Gainer said. "Any calculations based off that cell will be accurate too."

That's still a problem, of course, but it might be comforting to know that it's not as likely to creep outward from one result to the next.

A repair is in testing now. "We've come up with a fix for this issue and are in the final phases of a broad test pass in order to ensure that the fix works and doesn't introduce any additional issues--especially any other calculation issues," Gainer said. "This fix then needs to make its way through our official build lab and onto a download site--which we expect to happen very soon."

He didn't try to downplay the problem, as Intel once famously did with the FDIV calculation bug in some early Pentium processors, but he did indicate it's not widespread. "Of the 9.214*10^18 different floating-point numbers that Excel 2007 can store, there are 6 floating-point numbers (using binary representation) between 65534.99999999995 and 65535, and 6 between 65535.99999999995 and 65536 that cause this problem," Gainer said.

Floating-point numbers are those described with two components--the first between 1 and 10 and the second an exponent of the power of 10. So, for example, 2.6^10*2 is 2.6 * 100, or 260.

[news reader]

Bug is NOT only skin deep

The Microsoft blog was misleading. It's true that if you type "=850*77.1" into a cell, then the bug will only be in how it's displayed, but not how it's stored in memory. But if you type "=round(850*77.1)" into a cell, then the wrong value will actually, truly be in memory. It will propagate to other cells. It will affect sums and averages that you calculate. This bug is NOT only "skin deep".

[news reader]

Typo: =round(850*77.1,0)

One typo in that last comment: that should have been "=round(850*77.1,0)" to create a true incorrect value in memory.

[jds013]

Not problems

The bad value applies in most situations. Microsoft's disclaimer is ingenuous, and unfairly confuses a significant problem.

If you save your spreadsheet as a CSV, the wrong value of 100000 is written.

If you add a value to the incorrect number, the incorrect number is propagated.

The incorrect value is applied in SUMPRODUCT.

[MacHeads]

They Let balmer code these days?

Would explain some errors.

http://www.youtube.com/watch?v=BZDVKLESNXg

[MacHeads]

On second thought

Nahhh ... can it be the software that has been used to estimate
vista's actual users ? Can it ? 65535=100000... a few more
copies "displayed as sold".

A repair is in testing now. "We've come up with a fix for this
issue and are in the final phases of a broad test pass in order to
ensure that the fix works and doesn't introduce any additional
issues--especially any other calculation issues," Gainer said.
"This fix then needs to make its way through our official build
lab and onto a download site--which we expect to happen very
soon."

If this would not be a grave matter of concern it would be funny.
A spreadsheet program you can't trust to do operations properly
is not one worth considering using.

[Schpeyeder]

FYI: Only for * and /

This bug is present when using multiplication and division on the cell value. If you add or subtract the values, you will see the correct result. Strange bug though...

[ColdMast]

my old calculator can handle the number just fine?

I wonder what other problems will crop up with EXCEL 2007?

[R-Green]

About old calculators (did you check?)

Reminds me of the time in the late 60s when we were busy making chipsets for a small Japanese company whose name has since changed and who has become a major company. After shipping a lot, I accidently found that doing calculations like 66666*77777/66666 would give results such as 5345.533. Oops. Threw away a bunch of silicon and the calculator company was not so happy. If your calculator is really old, you might want to check.
Anonymously,
Bob

[lcminc]

It definitely more serious than formatting

Try the following vertical summation of these numbers.
2874
226345
153387
2285575
equals 2670188 instead of
2668181
I have no confidence in their products.

[gp2792]

uh no it doesn't

nice try, though.

[nick fortis]

Example at end of Excel Bug report

2.6^10*2 is **NOT** 260; I believe this is (2.6^10)*2 (which is
closer to 28233.419130676)

Perhaps one meant 2.6*(10^2) 2.6 times 10-squared = 2.6
times 100 = 260.

Furthermore, Floating point representations are possibly best
thought of as numbers expressed as a variant of Scientific
Notation.

E.g. The above: 2.823341913E+04

Nick F. 10.10.07

[lcminc]

Excel 2007 can not add correctly.

I have download the bug patch for the previously discussed problem with 850*77.1 =100000 and have found another bug that is more distrubing. Try adding the following numbers with a vertical sum 3277+ 240550 + 166937 + 2392356 = 2805127 instead of 2803120 . Any help would be appreciated

Logan McDonald

[artwolf]

Not for me

worked fine for me

[jlschuck]

Here's a bug that isn't skin deep:

"Charts on dialog sheets: Charts are not allowed on dialog sheets. It was not possible to implement this feature for the 2007 Office system."

Excel 2007 has less capability than the older version. My expensive EXCEL base software can't run in the newest, lastest, and not so greatest version of excel. What a rip off. Yes I did read the box and researched the web before I bought the software. They should have to tell people when the water down the software and then try to pass it off as an "upgrade". 2007 is all fluff and pretty graphics and nothing else. save your money.