#2807 format problem with wide ints

[Brian Griffin]

The format command errors out on %d of a wide int, but
than converts the wide int to int so that subsequent
format works without error.

Place the following in a file "format_bug.tcl":

set x [file size format_bug.tcl]

puts stdout "First format:"
if {[catch {format "%d bytes" $x} res]} {
puts stdout "Error: $res\nTrace back: $errorInfo"
} else {
puts stdout $res
}

puts stdout "Second format:"
if {[catch {format "%d bytes" $x} res]} {
puts stdout "Error: $res\nTrace back: $errorInfo"
} else {
puts stdout $res
}

Running the above code produces:

First format:
Error:
Trace back:
while executing
"format "%d bytes" $x"
Second format:
337 bytes

[Brian Griffin]

Logged In: YES
user_id=22949

I forgot to mention: This fails on 64-bit Solaris.

[Jeffrey Hobbs]

Logged In: YES
user_id=72656

This appears to be related to bug 699060 fixed in the 8.5
head. It needs to be backported as it clearly breaks on 8.4
(not all 64-bit plats, but something is wrong).

[Don Porter]

Logged In: YES
user_id=80530

Report 699060 claims the fix
went into Tcl 8.4.3 .

[Jeffrey Hobbs]

Logged In: YES
user_id=72656

It may, and 702622 appears to be in 8.4 as well, but that
doesn't discount the fact that %[duxX] have different code
between 8.4 and the head - there are no ChangeLog docs to
really indicate why they should be so separate.

[Don Porter]

Logged In: YES
user_id=80530

Is this bug present in Tcl 8.4.2 ?

[Jeffrey Hobbs]

Logged In: YES
user_id=72656

I found that this:

- if (wideValue>ULONG_MAX || wideValue<LONG_MIN) {
+ if (wideValue>LONG_MAX || wideValue<LONG_MIN) {

at line 2209 of tclCmdAH.c:Tcl_FormatObjCmd made it work on
Solaris-64, which indicates a sign extension bug in the
compiler.
Note that the above fix isn't actually correct code for this
case (although I wonder whether we should be mingling
unsigned checks for %d as well as the %[uoxX] this case covers).

Also, a shorter test case is:

et x [expr {wide(1)}]
set code [catch {format "%d bytes" $x} res]
puts "1st format ($code): $res"
set code [catch {format "%d bytes" $x} res]
puts "2nd format ($code): $res"

[Brian Griffin]

Logged In: YES
user_id=22949

Compiler bugs aside, I question why isn't %d automatically promoted to
%ld if/when Obj is a Wide int? Otherwise, tcl programmers are forced to
concern themselves with type in a "typeless" language. This doesn't
seem right to me.

See, if the wide int was actually indeed larger than LONG_MAX (>32-
bits) then an error would have been produced legitimately. This means
at the tcl level one must concern themselves with %d vs %ld. But these
notions were invented to deal with different types (int vs long). Tcl is
not supposed to be "typed".

[Donal K. Fellows]

Logged In: YES
user_id=79902

There are altogether too many ways for me to list right now
in which I'm not satisfied with the current handling of
64-bit values on 32-bit platforms. This is *not* the most
surprising of them. :^(

[Donal K. Fellows]

Logged In: YES
user_id=79902

The fix for Bug 868489 seems to have cleared this up as well
(or at least it all works for me now I've patched that.)