Centralise documentation of CONFIG_ options (and finding unused ones)

Hi All,

As a bit of a lark, I ran the following shell command:

grep -r -h -o -G "#define CONFIG.*" * | \
sed 's/#define[ \t]*//' | \
sed 's/^\([A-Za-z0-9_]*\).*/\1/' | \
sort -u > README.configuration.options

I then ran the result through the following script into README.configuration.options.counted

#!/bin/bash INPUT=doc/README.configuration.options

#!/bin/bash INPUT=doc/README.configuration.options

for i in cat $INPUT
do

	grep -rcs --exclude-dir=doc '\<'${i}'\>' * | \
	cut -d : -f 2 | \
	awk '{sum+=$1} END {printf("%d\t%s\n", sum, f);}' f=$i

done

The result? 5539 lines of pure configuration option goodness :)

Here's a little snippet:

6 CONFIG_100MHz
8 CONFIG_133MHZ_DRAM
7 CONFIG_300MHz
13 CONFIG_405
24 CONFIG_405CR
88 CONFIG_405EP
81 CONFIG_405EX

Hi Graeme,

Le 11/04/2011 15:11, Graeme Russ a ?crit :

I would tend to answer 'yes'. Unused options should not clobber code or documentation.

That would be nice indeed, but it requires all options to have the same semantics over all arches. While this is highly desirable, it may not be true ATM -- just thinking out loud, haven't checked. But even if some options only make sense only for some arch(es), then we could always do it this way:

doc/README.CONFIG_options
doc/README.CONFIG_SYS_options
doc/arch/README.CONFIG_options

Amicalement,

Albert.

Hmm, I have a better idea after I can to the realisation that the central location for configuration options is Ctrl^F in the doc/ directory :)

doc/README.CONFIG_options.unused
doc/README.CONFIG_options.undocumented

I will tweak my scripts to produce these two files and see what the result is.

Regards,

Graeme

Can you post data for which options are not used. While I think automatic removal is harsh, I think something close to automatic is reasonable.

- k

OK, here are some quick stats:

Total CONFIG_ options : 5559
Total CONFIG_ options not documented(*)/ : 4907 Total CONFIG_ options not used in code(+) : 1136 Total CONFIG_ options not used in any board config : 645

(*) README moved into doc/ to make the count of documented options more accurate
(+) i.e. only found in either doc/ or include/configs/

As you can see, there is scope for a LOT of code cleanup within the CONFIG_ namespace

Here is the script:

#!/bin/bash INPUT=doc/README.configuration.options.list OUTPUT1=doc/README.configuration.options.counted.code_usage OUTPUT2=doc/README.configuration.options.counted.doc OUTPUT3=doc/README.configuration.options.counted.config

echo "Catalogue Configuration Options"

grep -rhos "#define CONFIG.*" * | \
	sed 's/#define[ \t]*//' | \
	sed 's/^\([A-Za-z0-9_]*\).*/\1/' | \
	sort -u > $INPUT

echo "Counting usage of configuration options in code" rm $OUTPUT1 2> /dev/null
for i in cat $INPUT
do

	x=${i}'\s'
	grep -rcs --exclude-dir=include/configs --exclude-dir=doc '\<'${i}'\>' * | \
	cut -d : -f 2 | \
	awk '{sum+=$1} END {printf("%d\t%s\n", sum, f);}' f=$i >> $OUTPUT1

done
sort -n $OUTPUT1 > tmp
rm $OUTPUT1 2> /dev/null
mv tmp $OUTPUT1

echo "Counting documented options"
rm $OUTPUT2 2> /dev/null
for i in cat $INPUT
do

	x=${i}'\s'
	grep -rcs '\<'${i}'\>' doc/* --exclude=README.configuration.* | \
	cut -d : -f 2 | \
	awk '{sum+=$1} END {printf("%d\t%s\n", sum, f);}' f=$i >> $OUTPUT2

done
sort -n $OUTPUT2 > tmp
rm $OUTPUT2 2> /dev/null
mv tmp $OUTPUT2

echo "Counting usage of configuration options in board configuration files" rm $OUTPUT3 2> /dev/null
for i in cat $INPUT
do

	x=${i}'\s'
	grep -rcs '\<'${i}'\>' include/configs/* | \
	cut -d : -f 2 | \
	awk '{sum+=$1} END {printf("%d\t%s\n", sum, f);}' f=$i >> $OUTPUT3

done
sort -n $OUTPUT3 > tmp
rm $OUTPUT3 2> /dev/null
mv tmp $OUTPUT3

echo "Done"

Regards,

Graeme

How about Makefiles? Sometimes a CONFIG option is used only in a Makefiles.

Timur Tabi
Linux kernel developer at Freescale

code_usage includes Makefiles

Regards,

Graeme