I have a simple test file, test01.py, with this output:
$ cat test01.py
#!/usr/bin/env python

print("Line 1")
print("Line 2")
print("Line 3")
$
$ # Now try debugging it.
$ python3.0 -mpdb test01.py
--Return--
> /home/ericp/opt/Python-3.0.1/lib/python3.0/io.py(762)closed()->False
-> return self.raw.closed
(Pdb) b test01.py:4
Breakpoint 1 at /home/ericp/lab/Python-3.0.1/test01.py:5
(Pdb) r
--Return--
> /home/ericp/opt/Python-3.0.1/lib/python3.0/io.py(1471)closed()->False
-> return self.buffer.closed
(Pdb) c
line 1
line 2
line 3
The program finished and will be restarted

Two main problems:

1. I shouldn't see the code for io.py

2. The program doesn't stop at the breakpoint, because that 
stack frame is exposed to bdb as (file:"<string>", line:4),
not (file:"/home/.../test01.py", line:4).

Similar problem with 3.1rc1:

C:\...>c:\Python31rc1\python.exe -m  pdb hello01.py
--Return--
> c:\python31rc1\lib\encodings\cp437.py(19)encode()->b'Hello'
-> return codecs.charmap_encode(input,self.errors,encoding_map)[0]
(Pdb) b 2
*** Blank or comment
(Pdb) b hello01.py:2
Breakpoint 1 at c:\...\hello01.py:2

(Pdb) b
Num Type         Disp Enb   Where
1   breakpoint   keep yes   at c:\...\hello01.py:2
(Pdb) c
Hello
here
bye
The program finished and will be restarted
--Return--
> c:\python31rc1\lib\encodings\cp437.py(19)encode()->b'Hello'
-> return codecs.charmap_encode(input,self.errors,encoding_map)[0]
(Pdb)

Why has this not been resolved yet? Not having a working debugger is a
severe hindrance to the acceptance of Py3k.

Attached patch confirmedly fixes this.  Will commit as soon as svn is
back up.

Reading this patch, is it normal that the file is opened in text mode
without an encoding specified? What if the source file uses a different
encoding than the default system one?

That's a good question :)

A quick test shows that exec() does honor coding cookies in code given
as a bytestring, so simply changing to open(..., "rb") should be fine.

Fixed in r74366, and merged back to 3.1 branch in r74367.