[llvm-dev] RFC: changing variable naming rules in LLVM codebase

[via llvm-dev]

I had posted something in the code review but Chris suggested doing it 
here instead, which makes sense. Also I have to remember that the
discussion is specifically about spelling variables, not changing any
other spelling conventions.

Looking at names of "variables" there's reasonable support for making 
them visually more distinct from other kinds of names.  Regarding 
making data-member names distinct from local variables, some people 
don't see why it should matter, others find it helpful; given this 
neutral-to-helpful spectrum, going with the kind-of helpful convention 
seems preferable.

So:

- Local variables and formal parameters should be lower_case, with
  one exception: Variables/parameters that have lambda/function
  type should follow the function-name spelling rules.
- Class data members should have an "m_" prefix, so m_lower_case.
- Initialisms and other abbreviations would be considered words for 
  this purpose, so we have names such as:
    tli   // Local variable for TargetLoweringInfo
    m_cgm // Data member for CodeGenModule
- I don't have a good suggestion for file-static/global variables.
  Some people have suggested a "g_" prefix for globals, or possibly
  an "s_" prefix for class-static data.

Regarding the transition:

Some people have worried that the churn will cause blame issues.
I respectfully point out that in my own archaeology I have to deal
with lots of clang-format/indentation/other random semantically
meaningless refactoring, this is just one more.  Also the point is
not to optimize for git-blame but to optimize for reading what is
there at the moment.

A more focused and shorter transition period will create a lot of
short-term churn but get us to the good endpoint sooner.  Doing
conversions per-file or per-class (rather than per-function [too
small] or per-library [too big]) are probably the way to go.  
Given we are changing the names used for _data_, and we try to 
practice good data-hiding, the impact of the conversion of any 
given class *ought* to be reasonably confined.

If someone can make clang-tidy help with this, that's awesome.

I'm almost afraid to make the next suggestion, but here goes:
In more complicated/wide-impact cases, it would be possible to
stage a data-member name conversion into "small-bang" iterations
using a C++ tactic like this:
    class Foo {
      int m_bar; // The renamed member.
      int &Bar = m_bar; // TEMPORARY alias using the old name.
    };
This would have to be done sparingly and for good reason, such as
when the names are known across many components/subprojects and 
doing them all at once would be really too much. Someone would
have to commit to getting it all done and removing the aliases in
a reasonably short period of time.  Needing to do this trick would
be (IMO) strong evidence of poor software design and a place to
focus some refactoring effort.

HTH,
--paulr