Bug on Solaris

Tue Jan 8 10:20:00 GMT 2002

Jeff Sturm writes:
 > On Tue, 8 Jan 2002, Andrew Haley wrote:
 > > There is no way, unless the compiler that you sed to build gcj is
 > > broken, that it should be possible to get different results like this.
 > 
 > Hi Andrew,
 > 
 > I tried Martin's test program on a sparc v9 build.
 > 
 > gcj emits the constants MIN_DOUBLE and MAX_DOUBLE as follows:
 > 
 > .section ".rodata"
 > .align 8
 > .LLC0:
 > .long 0
 > .long 0
 > .align 8
 > .LLC1:
 > .long 0
 > .long 1
Thanks for confirming this.
 > This compiler is from unmodified sources, and built by bootstrap. A
 > similar C program doesn't break. So my hunch is a frontend bug.
Yeah, that's more or less what I said to Martin offline.
It is, however, a front end bug that manifests itself only when
running on SPARC. Endian, perhaps?
Andrew.