Disadvantage of make cscope

In the post, we uses the infrastructure by u-boot or Linux makefile to build the cscope database. But the disadvantage of that approach is that there are duplicated files for the same functions, so it is difficult the find a single path for the specific build.

E.g. in the current build

pandy@ubuntu:~/u-boot-2018.09$ cat cscope.files | wc -l
5342

The database includes 5K files.

Improvement

The approach is to limit the files to the ones used in the real build. We use the utility readelf to list all the files.

readelf --string-dump=.debug_str u-boot | sed -n '/\/\|\.c/{s/.*\]  //p}' > cscope.files

This will create a list of files which is used in the compilation and included in the elf file.

Then use following command to create cscope database:

cscope -bqk

The advantage of this method is that this method includes only files which are used in actual compilation.

Header files

The problem of this method is that it does not include header files. In order to overcome this issue, I need to find the all header files used in the compilation.

gcc has an option -H, which will print all used header files used in compilation.

In u-boot, environment var KCFLAGS could be used to specify customized option. Add following environment var:

export KCFLAGS=-H

Prepare for a clean build:

make clean

Capture all printing:

make > headers.txt 2>&1

Filter out all header files:

cat headers.txt  | sed -n 's/.* \(.*\.h\)/\1/p' | sort | uniq > headers_filted_uniq.txt

Create cscope.files as usualy:

readelf --string-dump=.debug_str u-boot | sed -n '/\/\|\.c/{s/.*\]  //p}' > cscope.files

Then add all header files:

cat headers_filted_uniq.txt >> cscope.files

Finally create cscope database:

cscope -bqk