Bug #1028
closedcpdup: not 64-bit clean
0%
Description
cpdup cannot be built on 64-bit Linux (probably not on 64-bit DragonFly either,
once that becomes available).
Build log attached (using Fedora 9, GCC 4.3. Same problem occurs with GCC 4.2
and, as far as I remember, 4.1)
Files
Updated by dillon over 16 years ago
:New submission from Michel Salim <salimma@fedoraproject.org>:
:
:cpdup cannot be built on 64-bit Linux (probably not on 64-bit DragonFly eit=
:her,
:once that becomes available).
:
:Build log attached (using Fedora 9, GCC 4.3. Same problem occurs with GCC 4=
:.2
:and, as far as I remember, 4.1)
The only actual error is occuring inside a system include file, which
I have no control over.
The 64<->32 bit warnings just require casts. cpdup is installing a
32 bit protocol identifier as a pointer in the hclink code, and
converting it back on callbacks. It should work as-is but it does
need a clean-up to get rid of the 64 bit warnings. If someone wants
to do that and submit a patch that would be great.
-Matt
Matthew Dillon
<dillon@backplane.com>
Updated by msylvan over 16 years ago
Here is a patch implementing the explicit casts. I've used size_t rather than
int for the integral conversions of pointers.
Updated by msylvan over 16 years ago
The 64-bit build errors are due to a namespace clash (or rather C's lack of
one): GLIBC's stat.h internally uses bits/stat.h, which on machines with 64-bit
wordsizes uses __unused for its own purpose. I've attached a patch that renames
every use of __unused in cpdup to __cpdup_unused; it works fine on Linux (the
build instructions will have to be modified accordingly)
Updated by msylvan over 16 years ago
One last thing: could a COPYING or LICENSE file be packaged with the next
release of cpdup? It tends to be expected on Linux packages.
Thanks,
--
Michel
Updated by joerg over 16 years ago
%zd or %zu?
(u)intmax_t + %jd/%ju is better here.
Joerg
Updated by msylvan over 16 years ago
On 32-bit systems we pay a penalty for using intmax_t, as the pointer size will
only be 32-bit. size_t is always the same size as a pointer, as both are big
enough to range over all of addressable memory on a system
Updated by dillon over 16 years ago
One can always use intptr_t. size_t is not a good abstraction for a
pointer-sized int.
-Matt
Matthew Dillon
<dillon@backplane.com>
Updated by msylvan over 16 years ago
Aha! Of course.. I was not familiar with it. Everything else is OK with both
patches, though?
Updated by dillon over 16 years ago
:
:
:Michel Salim <salimma@fedoraproject.org> added the comment:
:
:Aha! Of course.. I was not familiar with it. Everything else is OK with both
:patches, though?
The __cpdup_unused is ok. Not fun, but I can't think of another solution
at the moment.
Lets not use size_t for the casts. Does linux have an intptr_t ? Can
we use that?
-Matt
Matthew Dillon
<dillon@backplane.com>
Updated by msylvan over 16 years ago
Linux does, it just did not occur to me at that time.
Updated by erik-wikstrom over 16 years ago
intptr_t (and uintptr_t) is C99 so I would suspect that most modern
systems have them.
Updated by msylvan about 16 years ago
Just checking -- has the patch (modified to use intptr_t) been committed? I can
re-submit a corrected patch if necessary. Would like to drop the patch we use in
Fedora when the next version of cpdup comes out.
Updated by swildner almost 16 years ago
I've committed the casts but I'm rather reluctant about the __unused part.
Updated by sjg over 14 years ago
No update for some months, assume swildner's commit resolved this. Please
re-open if that is not the case.