Re: [dpdk-dev] [PATCH v2] ABI: Add abi checking utility

[Neil Horman <nhorman@tuxdriver.com>]

On Wed, Mar 04, 2015 at 04:15:18PM +0100, Thomas Monjalon wrote:
> 2015-03-04 09:39, Neil Horman:
> > On Wed, Mar 04, 2015 at 01:54:49PM +0100, Thomas Monjalon wrote:
> > > Hi Neil,
> > > 
> > > I remove parts that I agree and reply to those which deserve more discussion.
> > > 
> > > 2015-03-04 06:49, Neil Horman:
> > > > On Tue, Mar 03, 2015 at 11:18:47PM +0100, Thomas Monjalon wrote:
> > > > > 2015-02-02 13:18, Neil Horman:
> > > > > > +# Validate that we have all the arguments we need
> > > > > > +if [ ! -d ./.git ]
> > > > > > +then
> > > > > > +	log "WARN" "You must be in the root of the dpdk git tree"
> > > > > > +	log "WARN" "You are in $PWD"
> > > > > > +	cleanup_and_exit 1
> > > > > > +fi
> > > > > 
> > > > > Why not cd $(dirname $0)/.. instead of returning an error?
> > > > 
> > > > Why would that help in finding the base of the git tree.  Theres no guarantee
> > > > that you are in a subdirectory of a git tree.  I suppose we can try it
> > > > recursively until we hit /, but it seems just as easy and clear to tell the user
> > > > whats needed.
> > > 
> > > No I'm saying that you could avoid this check by going into the right
> > > directory from the beginning.
> > > We know that the root dir is $(dirname $0)/.. because this script is in
> > > scripts/ directory.
> > > 
> > That only helps if you start from the right directory.  If you run this command
> > from some other location, your solution just breaks.
> 
> Why it would break? $(dirname $0) is always reachable because you launched $0.
> The only exception is for the case the PATH variable is used to find the DPDK
> scripts/ directory (should not happen).
> 
Ah!  Sorry, misunderstood, for some reason I was conflating $0 with $PWD.  Yes,
this will work and I'll update it

> > > > > > +# Make sure we configure SHARED libraries
> > > > > > +# Also turn off IGB and KNI as those require kernel headers to build
> > > > > > +sed -i -e"$ a\CONFIG_RTE_BUILD_SHARED_LIB=y" config/defconfig_$TARGET
> > > > > > +sed -i -e"$ a\CONFIG_RTE_EAL_IGB_UIO=n" config/defconfig_$TARGET
> > > > > > +sed -i -e"$ a\CONFIG_RTE_LIBRTE_KNI=n" config/defconfig_$TARGET
> > > > > 
> > > > > Why not tuning configuration after make config in .config file?
> > > > > 
> > > > Because this way we save a reconfig (from a developer viewpoint), you should run
> > > > make config again after changing configs, and so this way you save doing that.
> > > 
> > > No, you run make config once and update .config file. That's the recommended
> > > way to configure DPDK.
> > > defconfig files are default configurations and should stay read-only.
> > 
> > They get overwritten when we do the git resets.  Its silly to modify your config
> > file after you run make config, in the event the make target has to re-read any
> > modified options and adjust dependent config files accordingly.  I understand
> > that doesn't happen now, but its common practice for every open source project
> > in existance.
> 
> I'm not sure to understand. Maybe an example would help.
> By the way, your method works.
For example, the linux kernel.  The .config file that is generated in the root
directory is converted to an autoconf.h in parallel with its generation, for
applications to key off of.  If you change something in .config, you need to run
make config again so that those changes are reflected into the other
auto-generated files.  Thats common practice.  So its counter intuitive to
assume that altering the generated .config file is automatically recognized by
the rest of the build, without a subsequent make config (be it explicit or and
implicit dependency of the make all target).

Neil


> 
>