Ordering of ports move to libc

Ordering of ports move to libc

[Joseph S. Myers]

After 2.19 has branched and master reopened for 2.20 development, we would 
like to move ports architectures to libc, as described at 
<https://sourceware.org/ml/libc-alpha/2014-01/msg00373.html> (which 
includes the architecture maintainers doing the moves for their 
architectures and checking that the before-and-after disassembly of 
installed shared libraries is identical).

I've checked what cases there are where one ports architecture uses files 
from another.  In such cases, either the moves need to be 
dependency-ordered, or else the #include direction needs to be reversed at 
move time.  To avoid unnecessary churn and potential conflicts when two 
people try to move their architectures at once, I suggest 
dependency-ordering if the depended-on architecture can be moved quickly.

The following architectures use linux-generic (both via Implies files and 
#include): aarch64, tile.  So I suggest that linux-generic is moved early.

There are the following other dependencies (for example, arm uses files 
from alpha):

aarch64: arm
arm: alpha
microblaze: alpha
mips: alpha arm hppa

Thus, I suggest moving alpha early - and once that's moved, I'll move arm.  
hppa is blocked by not building rather than by dependencies; I'll reverse 
the #include direction between sysdeps/unix/sysv/linux/hppa/umount.c and 
sysdeps/unix/sysv/linux/mips/mips64/umount.c unless hppa is fixed soon 
(fixing the hppa build would unblock it for move to libc).

As well as alpha, ia64 and m68k are unblocked for moving as soon as master 
reopens, and tile is unblocked as soon as linux-generic has moved (or they 
could move at the same time).  I intend to remove am33 as soon as master 
reopens.

When moving a port, as well as moving the files you should properly update 
the lists of ports in the libc README file as well.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: Ordering of ports move to libc

[Chris Metcalf]

I will plan to move linux-generic and tile relatively promptly once master has reopened for 2.20.

I'm not sure how to judge whether the move breaks aarch64; presumably it "should" work via the include paths being correct.  The best strategy may just be for me to coordinate with Marcus Shawcroft or whomever will be doing the aarch64 move.  Since it won't break anyone else, perhaps it's reasonable just to line up availability for both of us and then I'll push tile once it works, then have Marcus quickly validate aarch64 isn't broken, then do the aarch64 move when it makes sense for him.

On 1/30/2014 1:29 PM, Joseph S. Myers wrote:
> After 2.19 has branched and master reopened for 2.20 development, we would 
> like to move ports architectures to libc, as described at 
> <https://sourceware.org/ml/libc-alpha/2014-01/msg00373.html> (which 
> includes the architecture maintainers doing the moves for their 
> architectures and checking that the before-and-after disassembly of 
> installed shared libraries is identical).
>
> I've checked what cases there are where one ports architecture uses files 
> from another.  In such cases, either the moves need to be 
> dependency-ordered, or else the #include direction needs to be reversed at 
> move time.  To avoid unnecessary churn and potential conflicts when two 
> people try to move their architectures at once, I suggest 
> dependency-ordering if the depended-on architecture can be moved quickly.
>
> The following architectures use linux-generic (both via Implies files and 
> #include): aarch64, tile.  So I suggest that linux-generic is moved early.
>
> There are the following other dependencies (for example, arm uses files 
> from alpha):
>
> aarch64: arm
> arm: alpha
> microblaze: alpha
> mips: alpha arm hppa
>
> Thus, I suggest moving alpha early - and once that's moved, I'll move arm.  
> hppa is blocked by not building rather than by dependencies; I'll reverse 
> the #include direction between sysdeps/unix/sysv/linux/hppa/umount.c and 
> sysdeps/unix/sysv/linux/mips/mips64/umount.c unless hppa is fixed soon 
> (fixing the hppa build would unblock it for move to libc).
>
> As well as alpha, ia64 and m68k are unblocked for moving as soon as master 
> reopens, and tile is unblocked as soon as linux-generic has moved (or they 
> could move at the same time).  I intend to remove am33 as soon as master 
> reopens.
>
> When moving a port, as well as moving the files you should properly update 
> the lists of ports in the libc README file as well.
>

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: Ordering of ports move to libc

[Roland McGrath]

A grep suggests that the arm dependencies on alpha are just msgctl.c,
semctl.c, and shmctl.c.  Is that all you found too?

Given the relative activity of the maintenance of the two ports, I think it
makes more sense to reverse the dependency and move arm first.  I also
think it will help for arm specifically to be the first to move just
because you happen to be its maintainer: you are the person who has done
the most to understand and address the issues in doing a move, so you're
the most likely to do it right (perhaps including solving problems you
didn't think of and discuss here beforehand) and set the example for how to
do it with the least hassle.


Thanks,
Roland

Re: Ordering of ports move to libc

[Joseph S. Myers]

On Thu, 30 Jan 2014, Roland McGrath wrote:

> A grep suggests that the arm dependencies on alpha are just msgctl.c,
> semctl.c, and shmctl.c.  Is that all you found too?

Yes, that's all I found for arm (and microblaze).  (The mips dependency on 
alpha is ipc_priv.h, but that's silly - the alpha file just includes the 
powerpc one, so mips should just include the powerpc one directly and I'll 
make that fix after 2.19 is out.)

I'm happy to move arm first if desired.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: Ordering of ports move to libc

[Roland McGrath]

> I'm happy to move arm first if desired.

Unless Richard expresses a desire and commitment to do alpha first and
really fast for some reason, I think you should.

Re: Ordering of ports move to libc

[Richard Henderson]

On 01/30/2014 03:31 PM, Roland McGrath wrote:
>> I'm happy to move arm first if desired.
> 
> Unless Richard expresses a desire and commitment to do alpha first and
> really fast for some reason, I think you should.
> 

Please don't let me hold up any of this.  I'm quite happy for the arm/alpha
dependency to be reversed.


r~