#!/bin/bash

# Create schema for this report. Its name is based on the current git commit.
# We set search_path for this user for the below psql calls, which we'll reset
# at the end.
hash=$(git show --format=format:%h --no-patch)
psql -c "create schema git_$hash"
psql -c "alter user current_user set search_path to git_$hash"
psql -c "create table if not exists includes (type text, file text, header text); truncate table includes;"

IFS=$'\n'

# Grep all the backend .c files for the "#include" lines and insert them into the
# includes table.
for i in $(find src/backend \( -path "*libstemmer*" -prune \) -o -name \*.c -print); do
	insert_lines=""
	for line in $(git grep -h '^#include ".*\.h"' $i); do
		file=$(echo $i | sed -e 's/src.backend.//')
		line=$(echo $line | sed -e 's/#include "\([^"]*\)".*/\1/')

		if [ ! -z "$insert_lines" ]; then
			insert_lines=$insert_lines,
		fi
		insert_lines="$insert_lines ('source', '$file', '$line')"
	done
	# Insert in batches; too slow otherwise
	if [ ! -z "$insert_lines" ]; then
		psql -c "insert into includes values $insert_lines"
	fi
done

# Grep the .h files for the #include lines.  This code is almost identical to
# above, could stand refactoring.
for i in $(find src/include \( -path "*libstemmer*" -prune \) -o -name \*.h -print); do
	insert_lines=""
	for line in $(git grep -h '^#include ".*\.h"' $i); do
		file=$(echo $i | sed -e 's/src.include.//')
		line=$(echo $line | sed -e 's/#include "\([^"]*\)".*/\1/')
		if [ ! -z "$insert_lines" ]; then
			insert_lines=$insert_lines,
		fi
		insert_lines="$insert_lines ('header', '$file', '$line')"
	done
	if [ ! -z "$insert_lines" ]; then
		psql -c "insert into includes values $insert_lines"
	fi
done

# Some header files are not interesting
psql -c "delete from includes where header in ('c.h', 'postgres.h', 'postgres_fe.h', 'utils/elog.h')"

# Create a table with all the direct and indirect dependencies between header
# files
psql -c "create table allheaderdeps as
with recursive headertree(file, include, depth) as (
  select file, header as include, 1 from includes where type = 'header'
union
  select headertree.file, header, depth+1
    from headertree, includes 
   where includes.file = headertree.include
)
select * from headertree"

# Create a table with all the dependencies -- as above, but also consider all
# .c files.
psql -c "create table alldeps as
with recursive headertree(file, include, depth) as (
  select file, header as include, 1 from includes
union   select headertree.file, header, depth+1
    from headertree, includes 
   where includes.file = headertree.include
)
select * from headertree"

# Create a view to show the includes most used altogether.
# XXX I think this is slightly bogus ... perhaps the `alldeps` table should
# really include only .c files?  Not sure.
psql -c "create view topheaders as select a.*, b.numincludes from (select include, count(*) as used_by_source
from alldeps group by include) a join
(select file, count(*) as numincludes from allheaderdeps group by file) 
b on (b.file = a.include) /* where used_by_source >= 0 */ order by 2 desc"

# reset search_path
psql -c "alter user current_user reset search_path"