[Tcl-bugs] [ tcl-Feature Requests-929926 ] Tcl_Alloc should take WideInt or size_t

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

Feature Requests item #929926, was opened at 2004-04-05 19:44
Message generated for change (Comment added) made by dkf
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=360894&aid=929926&group_id=10894

>Category: 40. Memory Allocation
Group: None
Status: Open
Resolution: None
Priority: 3
Submitted By: Mikhail Teterin (kot)
Assigned to: Donal K. Fellows (dkf)
Summary: Tcl_Alloc should take WideInt or size_t

Initial Comment:
Currently Tcl_Alloc causes warnings on 64-bit platforms 
with things like ckalloc(sizeof(foo)), because sizeof returns 
size_t; and will break, when people start using large 
memory machines. Trying to allocate more than 4Gb of RAM 
seems ridiculous now, but just give it time... 

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

>Comment By: Donal K. Fellows (dkf)
Date: 2004-04-07 14:11

Message:
Logged In: YES 
user_id=79902

I use Tcl on big-iron too, so I really do understand.

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

Comment By: Don Porter (dgp)
Date: 2004-04-06 23:34

Message:
Logged In: YES 
user_id=80530

my mistake.  I overlooked your smiley.



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

Comment By: Mikhail Teterin (kot)
Date: 2004-04-06 23:27

Message:
Logged In: YES 
user_id=173641

Of course, I read it. That's how I knew Donal's name :-) 
 
I'm just trying to impress the importance of this 64-bit cleanlines 
(and thus the 64-bit godliness) on you, guys, so that, may be, 
you'll think of a way to bring it in sooner... 
 
The const-poisoning was a great thing -- forcing a review of 
almost entire TCL code. This should be no less usefull for the 
programmers, and a lot more useful for the users (of big 
machines). 

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

Comment By: Don Porter (dgp)
Date: 2004-04-06 23:09

Message:
Logged In: YES 
user_id=80530


Did you read TIP 115 or not?

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

Comment By: Mikhail Teterin (kot)
Date: 2004-04-06 14:28

Message:
Logged In: YES 
user_id=173641

So, if a bug exists for too long, requests to fix it become "feature 
requests"? :-) 
 
Yes, wherever such a thing exists, size_t should be the replacement. 
That's what it was invented for -- many years ago. 
 
Something needs to be done, and quickly (preferably in 8.5 with 
backport available for 8.4 as a patch). Donal K. Fellows is absolutely 
right about the 64-bit platforms becoming more common. My client 
just got a FreeBSD/amd64 system, thanks in no small part to the 
ability to mmap large files directly, and to the possibility to have 12Gb 
of RAM (although we only have 2Gb right now). 

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

Comment By: Donal K. Fellows (dkf)
Date: 2004-04-06 10:17

Message:
Logged In: YES 
user_id=79902

This is a known problem, but can't be fixed in 8.5 because
it stamps incompatably on an enormous number of APIs.  I had
a look into it when doing the TIP#72 implementation, and it
turns out to be even more pernicious than const-poisoning.

The correct type is either 'size_t' or 'unsigned long'
(depending on factors that have nothing to do with technical
issues.)

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

Comment By: Don Porter (dgp)
Date: 2004-04-06 04:50

Message:
Logged In: YES 
user_id=80530

See TIP 115:

http://purl.org/tcl/tip/115.html

for a pending proposal on
this issue.

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

Comment By: Don Porter (dgp)
Date: 2004-04-06 04:47

Message:
Logged In: YES 
user_id=80530

This is a Feature Request
not a bug.

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

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=360894&aid=929926&group_id=10894