I used to agree with this logic - it annoyed me to no end that most of the FPEs out there were the result of sloppy FP programming.
Then this week, an application I needed to use started giving me FPEs. No number crunching here, just silly project management. I have rebuilt glib, libgnomecanvas, labart and a few other libraries this week with -mieee in order to get the application to work (as a side note why does apt-get insist on overwriting my versions each time it does an update?). The main platform that these applications are developed on permits sloppy FP programming and we need to deal with that. Lets face it, developers are not super concerned with Alpha compatibility. I have essentially changed my opinion on this issue - I want software to just work on my alpha when I get it from the Debian repositories. If there is performance critical code, that is written well, perhaps a pragma is needed to tell gcc to ignore IEEE compliance for that section and optimize the hell out of it. >> It seems that Richard Henderson (who I understand to be the >> Alpha tool chain >> guy from RedHat) has unofficially blocked the 'make -mieee >> the default on >> Alpha' patch. >> > >Personally, I don't like it because making it the default lets buggy code slip >through. 99.9% of the time that I have ever encountered, if something works after >adding "-mieee" (or the equivalent with Compaq compilers) that is broken without it, >it is usually a case of: > > - performing a calculation with an uninitialized floating point variable (either >stack or heap garbage left around) > - a division by 0 in a calculation that really should have a safety check around it >anyway > >Only in .1% of cases, such as in number crunching apps, have I ever seen any real >need to need/use -mieee. For run of the mill apps like browsers and office apps, a >crash of this nature is usually pointing to a bug that should be squashed. ...tom --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.515 / Virus Database: 313 - Release Date: 01-Sep-2003
