[Tcl-bugs] [ tcl-Bugs-1527169 ] [scan] incompatibility

[SourceForge.net <no...@so...>]

Bugs item #1527169, was opened at 2006-07-23 00:38
Message generated for change (Settings changed) made by kennykb
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=1527169&group_id=10894

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: 23. [scan]
Group: development: 8.5a5
>Status: Pending
>Resolution: Wont Fix
Priority: 5
Submitted By: Don Porter (dgp)
Assigned to: Kevin B KENNY (kennykb)
Summary: [scan] incompatibility

Initial Comment:
On 64-bit system:

% info patch
8.4.14
% scan 0xfffff550 %x
-2736


% info patch
8.5a5
% scan 0xfffff550 %x
4294964560

TIP 237 doesn't appear
to call for this change.
A bug?

----------------------------------------------------------------------

>Comment By: Kevin B KENNY (kennykb)
Date: 2006-07-30 12:51

Message:
Logged In: YES 
user_id=99768

Alas, Tcl has had a great many inconsistencies over time
with the way that integers larger than the precision of the
machine are handled, and the way that differences among
machines are handled.

Moving forward, we've taken the approach in 8.5 that Tcl's
integers are of arbitrary precision, and provided %llx,
%llo, %llu and %lld format groups that work the same on all
platforms.  We've also adopted the idea that %x, %o, %u and
%d are the width of a native 'long int' or 'long unsigned
int' - whatever that is on the platform - and that %lx, %lo,
%lu, and %ld are the width of a 'long long' (or __int64 in
the Microsoft compilers) or 'unsigned long long' (unsigned
__int64).

This does indeed introduce a handful of incompatibilities,
but it also fixes a number of insidious bugs, of which the
ones mentioned in TIP #237 are representative.

To constrain an integer to 'signed 32-bit', irrespective of
the platform's integer width or the input format, the
easiest thing at present is to combine the bit weights yourself:

    expr {($x & 0x7fffffff) - ($x & 0x80000000)}

This expression will give the value of the least significant
32 bits of any integer, interpreted as a signed
(twos-complement) 32-bit quantity.

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2006-07-23 01:00

Message:
Logged In: YES 
user_id=80530


I know the revised docs
declare the new behavior
is correct, but don't trust
them; they were written by
the same fellow who wrote
the new code. :)




----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=1527169&group_id=10894