OpenEuphoria

1. Bug in 64 bit Phix [fixed, thanks]

• Posted by petelomax May 29, 2021
• 834 views

This one has apparently lain dormant for a while...

?{999999999,9999999994} 
for i=1 to 2 do 
    ?{999999999,i,9999999994} 
end for 

Output:

{999999999,9999999994} 
{999999999,1,1410065402} 
{999999999,2,1410065402} 

Clearly something wrong in pmain.e/DoSequence() or the pEmit2.e side of that,
otherwise I'm not quite sure how I'm going to fix it but I figured y'all should know,
and may as well ask if anyone has any bright ideas...

3. Re: Bug in 64 bit Phix

• Posted by petelomax May 29, 2021
• 818 views
• Last edited May 30, 2021

Yep, that was precisely it.

9999999994 === #2540BE3FA, and pilx86.e/emitHexDword()  
   was emitting #540BE3FA, dropping the #2 high-dword. 

It now checks for v<-#80000000 or v>#FFFFFFFF, and there is a new emitHexQuadword() for cases that need it.

Thanks. I had a bit of brain-freeze due to the rather long-winded wild goose chase it took to get to the above example.

(Of course I've run slap into the next issue... then again me finding them has to be better than shipping a release w/o detecting them)

Update, just for completeness: Now fixed, that turned out to be another long-dormant shocker... cmp [var],reg was (always) emitting an out-by-4 address! (again, on 64-bit only)