develooper Front page | perl.perl5.porters | Postings from July 2001

Re: [ Re: Redef of MAX_UV fixes half of UTSproblems]

Thread Previous | Thread Next
Dominic Dunlop
July 4, 2001 00:17
Re: [ Re: Redef of MAX_UV fixes half of UTSproblems]
Message ID:
At 11:32 -0500 2001-07-02, Jarkko Hietaniemi wrote:
>----- Forwarded message from Hal Morris <> -----
>>  > ===================== op/sprintf ====================
>>  > not ok 8 >%E< >123456.789< >1.234568E+05< >1.234567E+05< 
>>=> >1.234567E+5< # Like %e, but using upper-case "E"
>>  > not ok 162 >%-13g< >1234567.89< >1.23457e+06  < >1.23456e+06  < 
>>=> >1.23456e+6   <
>>  I hope Hugo or Jarkko know more about this test, as I've not looked at it
>>  [yet]
>Well, this is the way our sprintf works, apparently, by truncation.
>I would appreciate any reference to a POSIX, ANSI, or other standards
>document that says it should NOT work that way.  If they exist, I'll
>see that UTS conforms (or at least can be made to conform with a
>POSIX flag set or some such -- we don't want to break our customer's
>applications if they've been compensating for it.
>Thanks,   Hal Morris,  UTS Global, LLC
>----- End forwarded message -----

Sorry for delay in reply: I'm somewhat out of the loop being a boring 
banker at the moment. Anyway, sorry Hal. but I'd say your 
implementation doesn't follow the standard, which says

e,E    The double argument is converted in the style [-]d.ddde+/-dd 
... The value is rounded to the appropriate number of digits.  ...

(ANSI X3.159-1989, section, page 134, lines 25-32. I have not 
forked over $225 for the latest version of the standard -- ISO/IEC 
9899:1999, but a late draft I have has the same wording as the 1989 

Note that, although the standard specifies that rounding should 
occur, it doesn't specify a rounding algorithm. Having worked on 
standards committees, I'd say that this was intentional. When first I 
cooked up the expanded sprintf.t, I included tests which assumed IEEE 
754-style rounding, but it quickly became clear that this was not 

While I'm here, the late draft adds a clause to the description of 
the e,E format: "The exponent always contains at least two digits, 
and only as many digits as are necessary to represent the exponent." 
That second clause is new, and effectively says that Microsoft's (and 
maybe others') C is wrong always to produce three exponent digits -- 
and enclosing thread. I wonder if the Redmond faction was alert 
enough to prevent that from making it to the final document.
Dominic Dunlop

Get Your Private, Free Internet Access from FREE INTERNET at

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About