[clang] [NFC][clang] Quote literal builtin attribute markers in Builtins.def docs (PR #160080)

Erich Keane via cfe-commits cfe-commits at lists.llvm.org
Wed Sep 24 06:15:20 PDT 2025


erichkeane wrote:

> > > > > But I'm not sure what we could use reasonably.
> > > > 
> > > > 
> > > > We could replace it w/ e.g. `*` and add a note along the lines of ‘where `*` is any integer/index/whatever it’s actually supposed to be’
> > > 
> > > 
> > > Asterisk seems pretty overloaded here... maybe the same except with `{num}` or something like that?
> > 
> > 
> > What about using two letters, e.g., `"nn"`, to mark `nonnull`?
> 
> I experimented with using `"nn"` as the marker for builtin nonnull, but that caused 137 test failures (in SemaHLSL, SemaSPIRV, CodeGenSPIRV, CodeGenHLSL, etc.). Using `"{null}"` reduced it to 3 failing tests:
> 
> ```
> Clang :: CodeGen/AArch64/strictfp-builtins.c
> Clang :: CodeGen/X86/strictfp_builtins.c
> Clang :: CodeGen/strictfp_builtins.c
> ```
> 
> The most stable option so far is **`"NN"`**, where all tests pass.

Ah, I was not suggesting we switch the `N` away from null, I'm not sure it is a particularly good idea to have ONE be 2 characters, particularly with how it is parsed.  I am afraid that above, all you did is test the thoroughness of our testing :) 

I was suggesting the `{num}` or whatever for the cases where the documentation is currently using `N` to mean `some random number`.

https://github.com/llvm/llvm-project/pull/160080


More information about the cfe-commits mailing list