Reduce boost dependency (boost/algorithm/string/split) #22683

Proposing a method to remove the dependency of the boost/algorithm/string/split library. It can be done by including a small function in utils or adding a small alternative code patch in its use cases. The proposed method is much more efficient than the boost/algorithm/string/split. Here is a small demonstration which performs the same task in just 11 microseconds compared to boost library which takes 30 microseconds. (View here) The suggested alternative would reduce boost dependency in the files and thus improve their compilation and execution efficiency.

Following reference states that over big data the boost split performs a lot slower compared to stl implementation: https://stackoverflow.com/questions/5505965/fast-string-splitting-with-multiple-delimiters.

I'm happy to open a PR with these changes, but currently looking for concept ACKs to confirm if this is a good direction.

Boost removal is an eventual goal of the project. We've had similar proposals to remove these Boost components in the past, i.e #13751, you might want to read through that PR for some context. Note that there are also currently PR's open that remove Boost usage, i.e #18130, so you may also be interested in reviewing those.

Proposing a method to remove the dependency of the boost/algorithm/string/split library. It can be done by including a small function in utils or adding a small alternative code patch in its use cases.

... a comprehensive set of unit tests must be included as well. And all the new code must be maintained forever :)

The proposed method is much more efficient than the boost/algorithm/string/split. Here is a small demonstration which performs the same task in just 11 microseconds compared to boost library which takes 30 microseconds.

I don't think the efficiency of the current code is an issue.

Boost removal is an eventual goal of the project.

After adding boost::process it looks like a moving target :)

After adding boost::process it looks like a moving target :)

At least boost::process is optional.

I think before looking into removing header-only boost dependencies, we should be looking into lower hanging fruits such as the filesystem library. See #20744.

Going to close this for now. boost/algorithm/string/split has since been removed: #25057.