Interesting, Devel::Peek is a look into the internals of Perl that I never thought I'd need to see. Not as disturbing as watching sausage being made but some gory details about variable storage exposed with this method.

To see how nulls in the DB ('undef' as far as Perl is concerned) look when dumped, I added a fifth column (E) to the database table with a null value and added that to the query and dumped output. It seems that Dump sends its output to stderr and so I had to comingle the output streams when running this from a DOS command prompt (and then edit the result to collate the two streams). This is the result of the first case where the Single database columns containing values expressed in scientific notation (columns B & C) are returned with a "missing" value (a null byte--not quite the same as an empty string):

a=6.02518
SV = PV(0x277ae34) at 0x47d2b4
  REFCNT = 1
  FLAGS = (POK,IsCOW,pPOK)
  PV = 0x275664c "6.02518"\0
  CUR = 7
  LEN = 10
  COW_REFCNT = 1

b=
SV = PV(0x277ae5c) at 0x47da94
  REFCNT = 1
  FLAGS = (POK,pPOK)
  PV = 0x275672c "\0"\0
  CUR = 1
  LEN = 10

c=
SV = PV(0x277ae54) at 0x47da34
  REFCNT = 1
  FLAGS = (POK,pPOK)
  PV = 0x27567ac "\0"\0
  CUR = 1
  LEN = 10

d=8.99280607700348E-2
SV = PV(0x277ae4c) at 0x47d134
  REFCNT = 1
  FLAGS = (POK,IsCOW,pPOK)
  PV = 0x28ac1bc "8.99280607700348E-2"\0
  CUR = 19
  LEN = 21
  COW_REFCNT = 1

e=
SV = PV(0x277aeac) at 0x47d1f4
  REFCNT = 1
  FLAGS = ()
  PV = 0
[download]

As expected, in the second query where I added zeros to the columns (A+0, B+0, C+0, D+0, E+0) in the query, the problematic B and C columns are recast as double precision values and show up properly and look normal using Dump(). I'm not much on Perl internals and haven't needed to immerse myself in perlguts just yet so I'm hoping that something in the output above sheds some light on why Single values expressed in scientific notation cannot seem to be reliably returned with a fetch using DBI.

While unsure about what is going on, i must admit that was a totally unexpected result

i was wondering if you had run into something called a dualvar.

  use strict; use warnings; 
  $|=1; 

  use Scalar::Util qw(dualvar);


use Devel::Peek;
print "nv1\n"; my $nv1 = 1e-2                 ;Dump($nv1); print "nv1:
+$nv1 ".($nv1+0)."\n"; Dump($nv1); 
print "dv2\n"; my $dv2 = dualvar( 1e-2, "hi" );Dump($dv2); print "dv2:
+$dv2 ".($dv2+0)."\n"; Dump($dv2);
[download]

Result

nv1
SV = NV(0xa7c024) at 0xa3e614
  REFCNT = 1
  FLAGS = (PADMY,NOK,pNOK)
  NV = 0.01
nv1:0.01 0.01
SV = PVNV(0x3f7364) at 0xa3e614
  REFCNT = 1
  FLAGS = (PADMY,NOK,pIOK,pNOK)
  IV = 0
  NV = 0.01
  PV = 0xa959d4 "0.01"\0
  CUR = 4
  LEN = 36
dv2
SV = PVNV(0x3f73c4) at 0xa8facc
  REFCNT = 1
  FLAGS = (PADMY,NOK,POK,pNOK,pPOK)
  IV = 0
  NV = 0.01
  PV = 0xa45c5c "hi"\0
  CUR = 2
  LEN = 10
dv2:hi 0.01
SV = PVNV(0x3f73c4) at 0xa8facc
  REFCNT = 1
  FLAGS = (PADMY,NOK,POK,pIOK,pNOK,pPOK)
  IV = 0
  NV = 0.01
  PV = 0xa45c5c "hi"\0
  CUR = 2
  LEN = 10
[download]

Of particular interest here is the line "dv2:hi 0.01" where in string context you get hi, but in numeric you get 0.1. But it doesnt seem that dualvars are your problem.