[PATCH] D96572: [Clang][ASan] Introduce `-fsanitize-address-destructor-kind=` driver & frontend option.

Mon Feb 15 08:57:28 PST 2021

jansvoboda11 added a comment.

In D96572#2561216 <https://reviews.llvm.org/D96572#2561216>, @delcypher wrote:

> @jansvoboda11 I noticed that the `Options.td` file is making use of some fancy mixins that mean the option gets automatically marshalled by the frontend into the codegen options :). I tried this out using the patch to this patch and all my tests seem to pass. Should I be using these mixins (I don't know how stable they are because this looks like a new feature). Am I using them properly? It seems that the driver still has to parse the option manually (which is actually what I want) but the frontend automagically parses the option and modifies the right codegen option for me.

Hi, yes, this is a new feature. The option generation is not being enforced in any way yet, but the automatic marshalling should work. You can test it by running your `clang -cc1` commands with the `-round-trip-args` flag (or `clang` with `-Xclang -round-trip-args`). With this option, the original `cc1` command line is first parsed into a dummy `CompilerInvocation`, which is used to generate new command line. The new command line is what is then used to create the real `CompilerInvocation`.

As you've correctly discovered, this does not interact with the driver in any way.

I'll be sending an email to cfe-dev about this shortly. Your marshalling code seems correct to me. If something does not work, check out the patch with documentation D95790 <https://reviews.llvm.org/D95790> and feel free to let me know.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D96572/new/

https://reviews.llvm.org/D96572