[PATCH] D134337: [clang] [Driver] More flexible rules for loading default configs

Wed Oct 19 02:43:18 PDT 2022

sepavloff added a comment.

In D134337#3866267 <https://reviews.llvm.org/D134337#3866267>, @mgorny wrote:

> In D134337#3865905 <https://reviews.llvm.org/D134337#3865905>, @tstellar wrote:
>
>> In D134337#3865753 <https://reviews.llvm.org/D134337#3865753>, @mgorny wrote:
>>
>>> In D134337#3865541 <https://reviews.llvm.org/D134337#3865541>, @tstellar wrote:
>>>
>>>> I know I'm a little late here, but having a default config file that's always loaded makes triaging issues much harder, because now every time someone files a bug, we need to ask for the contents of their config directory.
>>>
>>> I don't see that as much worse than asking them for the values of `*DEFAULT*` CMake options or whether they have patched their clang because the driver code makes so many specific assumptions that it didn't work for them. At least `clang -v` tells us whether they are actually using any configs.
>>
>> At least as a distro maintainer, I know the CMake options used for the packages I'm trying to support.
>
> As a distro maintainer, you can decide to build clang without system/user config directories set (which is the default), in which case only explicit `--config` will work.

Strictly speaking in this case the default config file may be loaded from the binary directory but the latter is controlled by distro.

Users also may split configuration file into pieces and include them using `@file` construct, in this case single config file is not enough. I think clang could print driver options in output of `-v`. Another way is to implement a special command line option, say `-print-args`, which would print all driver options.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134337/new/

https://reviews.llvm.org/D134337