Suppressing a plus from setlocalversion

Suppressing a plus from setlocalversion

[Pavel Roskin]

Hello!

There is one annoyance in the build system that I cannot work around.
Before I attempt hacking the code, I'd like to make sure I'm not
missing anything obvious.

Suppose that I have a kernel version 3.1-rc2.  I want to test some patch
for some code that is compiled as a module.  I change the code and
create a patch with STGit.  I recompile the modules, install them by
"make module_install", unload the module I patched, load it again
and find that there is no effect from my change.  It turns out that the
modules were installed to "/lib/modules/3.1.0-rc2+" and the plus was
added by scripts/localversion because the source tree is now modified.

So I recompile the whole kernel, reboot and continue working on my
patch.  Many days later, I'm about to send my patch, but I want to
retest the system without it, so I "stg pop" it, recompile the
module, install it and find that the unpatched module behaves just like
the patched one.  Of course, it turns out that the unpatched
module ended up in "/lib/modules/3.1.0-rc2", whereas modprobe keeps
loading the modified module from "/lib/modules/3.1.0-rc2+"

CONFIG_LOCALVERSION_AUTO is unset, but that doesn't stop
scripts/localversion from adding the plus.  Setting CONFIG_LOCALVERSION
doesn't help.  Setting LOCALVERSION on the make command line helps, but
it needs to be done every time, and it's too easy to forget.  I don't
want to write a wrapper around make, it would be too intrusive
(actually, it's my solution for now).

Adding anything to localversion* doesn't suppress the plus.  Keeping
some patch in STGit all the time is quite annoying, I'll eventually end
up sending it to some mailing list by mistake.  STGit cannot hide
applied patches.  Besides, I'll need to be aware of the patch when
switching branches.

File .scmversion is only read from the source directory, so having it
is equivalent to having a modified source all the time, which is what
I'm trying to avoid.  Sure, I can keep .scmversion not in git, but I
risk to clean it up with "git clean" eventually.

If there is no simple solution with the existing code, maybe
non-empty CONFIG_LOCALVERSION should suppress the plus just like
LOCALVERSION?

Or maybe CONFIG_LOCALVERSION_AUTO should be not yes/no, but
full/minimal/no?

-- 
Regards,
Pavel Roskin

Re: Suppressing a plus from setlocalversion

[David Rientjes]

On Mon, 15 Aug 2011, Pavel Roskin wrote:

> Hello!
> 
> There is one annoyance in the build system that I cannot work around.

That's surprising since you actually state the workaround, i.e. using 
LOCALVERSION= with make, yourself.

> Before I attempt hacking the code, I'd like to make sure I'm not
> missing anything obvious.
> 
> Suppose that I have a kernel version 3.1-rc2.  I want to test some patch
> for some code that is compiled as a module.  I change the code and
> create a patch with STGit.  I recompile the modules, install them by
> "make module_install", unload the module I patched, load it again
> and find that there is no effect from my change.  It turns out that the
> modules were installed to "/lib/modules/3.1.0-rc2+" and the plus was
> added by scripts/localversion because the source tree is now modified.
> 

Yes, because the kernel you've produced is no longer 3.1.0-rc2, it is 
modified.  If you want to do something else locally, then you can always 
use LOCALVERSION=.

The + was added to stop falsely identifying kernels as a specific version 
when they are modified by default.  Sorry if you forget to override it 
sometimes.

Re: Suppressing a plus from setlocalversion

[Pavel Roskin]

Quoting David Rientjes <rientjes@google.com>:

> On Mon, 15 Aug 2011, Pavel Roskin wrote:
>
>> Hello!
>>
>> There is one annoyance in the build system that I cannot work around.
>
> That's surprising since you actually state the workaround, i.e. using
> LOCALVERSION= with make, yourself.

OK, I don't consider that an acceptable solution, as it could  
interfere with other build systems.

However, as soon as I posted this message, I found a better solution,  
namely setting LOCALVERSION= in the environment (both for the normal  
user and for root).  This is less likely to have unexpected effects,  
since the environment variables don't override the variables  
explicitly set in makefiles.

> Yes, because the kernel you've produced is no longer 3.1.0-rc2, it is
> modified.  If you want to do something else locally, then you can always
> use LOCALVERSION=.
>
> The + was added to stop falsely identifying kernels as a specific version
> when they are modified by default.  Sorry if you forget to override it
> sometimes.

OK, I have an acceptable solution for my needs now, and if others need  
it, hopefully they would find this post.

-- 
Regards,
Pavel Roskin

Re: Suppressing a plus from setlocalversion

[Stefan Richter]

On Aug 15 Pavel Roskin wrote:
> Quoting David Rientjes <rientjes@google.com>:
> 
> > On Mon, 15 Aug 2011, Pavel Roskin wrote:
> >
> >> Hello!
> >>
> >> There is one annoyance in the build system that I cannot work around.
> >
> > That's surprising since you actually state the workaround, i.e. using
> > LOCALVERSION= with make, yourself.
> 
> OK, I don't consider that an acceptable solution, as it could  
> interfere with other build systems.
> 
> However, as soon as I posted this message, I found a better solution,  
> namely setting LOCALVERSION= in the environment (both for the normal  
> user and for root).  This is less likely to have unexpected effects,  
> since the environment variables don't override the variables  
> explicitly set in makefiles.

Having this globally as an environment variable seems to be a more
intrusive alternative; there are no namespaces.

You can wrap the make command lines that you often use into shell
aliases.  I often use command lines like 'nice make -j8' or
'nice make -j8 C=1 CF="-D__CHECK_ENDIAN__"' by alias.

> > Yes, because the kernel you've produced is no longer 3.1.0-rc2, it is
> > modified.  If you want to do something else locally, then you can always
> > use LOCALVERSION=.
> >
> > The + was added to stop falsely identifying kernels as a specific version
> > when they are modified by default.  Sorry if you forget to override it
> > sometimes.

I for one am glad that I don't have to remember to force a localversion
all the time.  All of the kernels that I runtime-test have their source
managed by quilt, not git.  Thankfully, scripts/setlocalversion
acknowledges that quilt users know what they are doing. :-)
-- 
Stefan Richter
-=====-==-== =--- =----
http://arcgraph.de/sr/

Re: Suppressing a plus from setlocalversion

[Pavel Roskin]

Quoting Stefan Richter <stefanr@s5r6.in-berlin.de>:

> On Aug 15 Pavel Roskin wrote:
>> Quoting David Rientjes <rientjes@google.com>:
>>
>> > On Mon, 15 Aug 2011, Pavel Roskin wrote:
>> >
>> >> Hello!
>> >>
>> >> There is one annoyance in the build system that I cannot work around.
>> >
>> > That's surprising since you actually state the workaround, i.e. using
>> > LOCALVERSION= with make, yourself.
>>
>> OK, I don't consider that an acceptable solution, as it could
>> interfere with other build systems.
>>
>> However, as soon as I posted this message, I found a better solution,
>> namely setting LOCALVERSION= in the environment (both for the normal
>> user and for root).  This is less likely to have unexpected effects,
>> since the environment variables don't override the variables
>> explicitly set in makefiles.
>
> Having this globally as an environment variable seems to be a more
> intrusive alternative; there are no namespaces.

I know, it's not an ideal solution.

> You can wrap the make command lines that you often use into shell
> aliases.  I often use command lines like 'nice make -j8' or
> 'nice make -j8 C=1 CF="-D__CHECK_ENDIAN__"' by alias.

__CHECK_ENDIAN__ is addressed by a separate patch :)

I guess I need to write a wrapper that would be suitable for my needs.  
  Still, I'd prefer the build system do what I want, at least in some  
aspects.

>> > Yes, because the kernel you've produced is no longer 3.1.0-rc2, it is
>> > modified.  If you want to do something else locally, then you can always
>> > use LOCALVERSION=.
>> >
>> > The + was added to stop falsely identifying kernels as a specific version
>> > when they are modified by default.  Sorry if you forget to override it
>> > sometimes.
>
> I for one am glad that I don't have to remember to force a localversion
> all the time.  All of the kernels that I runtime-test have their source
> managed by quilt, not git.  Thankfully, scripts/setlocalversion
> acknowledges that quilt users know what they are doing. :-)

Actually, it's simply unaware of what you are doing.

Also, setlocalversion is unaware that the wireless-testing tree is  
tagged every day.  It doesn't add the plus for the tagged revisions,  
even though they are very different from the version in the source  
Makefile.

But I'd rather not be on the mercy of a script not knowing something.

-- 
Regards,
Pavel Roskin