Created attachment 4633 [details]
Junit generated excel file

this one works in Excel v.X...

This does not work in any version of Excel, unless you run the testcase against 
the head. In that case, the testcase produces a spreadsheet that loads 
properly, but the testcase fails.

The reason is that the testcase calls HSSCell.setErrorValue( (byte) 1), which 
is an invalid argument. The valid arguments are

0   #NULL!
7   #DIV/0!
15  #VALUE!
23  #REF!
29  #NAME?
36  #NUM!
42  #N/A

These are the 7 error types allowed in Excel. There are no others. If you set 
the erro value of a BOOLERR record to any other value, the spreadsheet will not 
load in Excel.

Why these magic numbers? Put the error codewords together in a string and 
seperate each with a \0, like you would if you were programming in C. The magic 
numbers are the offsets to the beginning of each codeword.

#NULL! #DIV/0! #VALUE! #REF! #NAME? #NUM! #N/A
0123456789012345678901234567890123456789012345
0         1         2         3         4

I would work on a patch for this but... the HEAD is very strange right now. The 
performance branch merge seems to have changed *everything*.

My instinct is that BoolErrRecord should throw an exception if the client tries 
to set an unacceptable value. Add public static final int into BoolErrRecord to 
represent the 7 allowed values. Change the testcase that produces 
testBoolErr.xls. And add a new testcase to verify that invalid arguments to 
HSSFCell.setErrorValue throw an exception.

Fixed in Branch. Added check to BoolErrorRecord, to throw RuntimeException, and
updated Testcase. Committed the same changes to HEAD, but the tescase was
already failing in HEAD.