[PATCH 2/2 v5] [BZ #10283] localedef: align fixed maps to SHMLBA

[Mike Frysinger]

On Wednesday 19 June 2013 19:02:42 Roland McGrath wrote:
> You're not going to convince me that SHMLBA counts as "generic".
> The code should be written to make MAP_FIXED_ALIGNMENT optional
> and not do any extra alignment if it's not defined in libc-mmap.h.

the generic code (locale/programs/locarchive.c) is already written to assume 
that MAP_FIXED requires at least PAGE_SIZE alignment.  it's been this way 
since the code was introduced (Apr 2009).  i guess you'd submit that this code 
has simply always been wrong.  however, writing the generic code to work with 
no alignment at all so it'll work on a hypothetical machine that, almost 
certainly (hell i'll bet my hat on it) will never exist (at least that will be 
capable of running glibc), is entirely pointless.  especially when no one can 
actually validate it does anything useful (like compile let alone execute).

> The generic libc-mmap.h should be empty (just comments).
> sysdeps/unix/sysv/linux/libc-mmap.h can use SHMLBA.

if we follow your logic, even linux/libc-mmap.h isn't the right place.  the 
fact that Linux (and every single OS out there) happens to have SHMLBA that 
reflects the right value is incidental.  so we should keep a copy of this 
header file in every OS & arch combo.

we've dealt with this copy & paste hell already with things like the bits 
headers.  having one master version which, as i documented in another e-mail, 
does the right thing for every OS/arch glibc supports (and even the ones it 
doesn't) keeps the maintenance cost down.  it doesn't even have a maze of 
ifdefs like we've accepted with other headers.
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20130620/98751d4e/attachment.sig>