tmroeder added a comment.
In https://reviews.llvm.org/D54657#1302086, @joerg wrote:
> I don't understand the point here. Why would you want to include
> pre-processing-only commands in the compilation database?
Sorry for not being clear enough.
I don't want to include entries with -E. However, in the Linux kernel build,
some build commands with -E occur after the regular build step, and that means
that without the append option, the output .json file for those objects gets
overwritten with a useless entry.
I originally thought that the simplest fix was to add a flag that appended to a
json file instead of overwriting it. However, thinking about it some more, it
would be nice if DumpCompilationDatabase refused to write such entries, since
then I wouldn't care about the append version of this flag. I could drop the
rest of this code (and fix the tests) if I could instead add the following
after the check for -### at the top of DumpCompilationDatabase.
if (C.getArgs().hasArg(options::OPT_E))
return;
What do you think? I don't want to break anyone who depends on the current
behavior of -MJ, but I think it's reasonable for DumpCompilationDatabase to
output only entries it can use.
Repository:
rC Clang
https://reviews.llvm.org/D54657
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
