Question

I’m cleaning up a large C++ code base where I need all variables of type “vector” need to be changed to “std::vector”. Skipping over #include and comments in the code. And most importantly, if the expression is already written as “std::vector”, don’t convert it to “std::std::vector”

That is:

#include <vector>
vector<Foo> foolist;
typedef vector<Foo> FooListType;
std::vector<Foo> otherfoolist;
int main()
{
    // this vector is for iteration
    for (vector <Foo>::iterator itor = foo.begin...)

Converts to

#include <vector>
std::vector<Foo> foolist;
typedef std::vector<Foo> FooListType;
std::vector<Foo> otherfoolist;
int main()
{
    // this vector is for iteration
    for (std::vector<Foo>::iterator itor = foo.begin...)

So far, I have this narrowed down to two sed commands

sed -r 's/vector\s{0,1}</std::vector</g' < codefile > tmpfile
sed 's/std::std/std/' < tmpfile > codefile

The first sed matches “vector< and “vector <” and converts either to “std::vector<“.

The second sed fixes the side effect of converting “std::vector<” into “std::std::vector<“.

How can merge the two different regex expressions above so I can have a single sed command that fixes the code correctly.

I’ve tried reading online about lookahead and lookbehind, but my eyes are starting to burn out.

Answer 1

You can make the first regexp also match a possible std:: by using

sed -r 's/[std:]*vector\s{0,1}</std::vector</' < codefile > tmpfile

btw: you can make the changes in place by adding -i and just passing the file as command line parameter:

sed -i -r 's/[std:]*vector\s{0,1}</std::vector</' codefile