Micro-coding standard question

LLVM Project
1

I recall from a while back that LLVM coding style preferred using a literal char instead of a single-character string when writing to an output stream. That is:

OS << ' ';  // Instead of OS << " ";
OS << '\n'; // Instead of OS << "\n";

Some of my code reviewers gave the same feedback as well, but this is not mentioned anywhere in the coding standard. I am assuming the only reason to prefer a character instead of a string is performance (else the string is more idiomatic). Currently the code has a mix of both, and I wanted to check if there is any reason to use single char over a string. In optimized builds, it would seem the performance is likely same.

We don’t need to document this necessarily in the CS but would be good to establish if there is a preference or both are acceptable, or maybe the preference is now to use the more idiomatic string.

Thanks

2

The single character calls a different overload of <<, which gets inlined but the string one does not

3

Unless the code is performance critical (which most stream printing is not, but there are some rare cases) my guideline would be to match the surrounding style:

Most uses of << are also going to involve the use some multi-character "string" literals, in which case I’d consistently use "string" literals for everything, even if use of character literals is possible in some cases. Basically, don’t mix " and ' for the sake of micro-optimization.

2 Likes