[PATCH] D69764: [clang-format] Add East/West Const fixer capability

Mon Aug 9 07:12:58 PDT 2021

MyDeveloperDay added a comment.

In D69764#2934483 <https://reviews.llvm.org/D69764#2934483>, @erichkeane wrote:

> In D69764#2934473 <https://reviews.llvm.org/D69764#2934473>, @MyDeveloperDay wrote:
>
>> In D69764#2934378 <https://reviews.llvm.org/D69764#2934378>, @erichkeane wrote:
>>
>>> I've just been watching this from the sideline, but the cases where this breaks code are unacceptable for this tool, it is a complete direction change for the tool, and making that direction change silently on a review of a 15 month patch, where TWO code owners have said 'no' for that reason is absurd.
>>>
>>> I use this tool daily as a part of my 'upload' script, having it silently bust code between when I validate it and when I upload it is terrible, and makes the tool unusable for my purposes.  If we change this direction without a full RFC, my next step is going to be an RFC to remove clang-format from the check-in requirements of the entire LLVM project.
>>
>> This and other potentially other mutating options would and MUST in my view ALWAYS be 100% "off by default" for all default style options (as -fix is for clang-tidy), it would be a purely "opt in" basis. (via .clang-format or command line)
>>
>> I personally use this in a non modifying way "using the -dry-run mode" to catch new"east/const violations" and report failure back rather than "change the code itself"
>>
>> I would not expect clang-format usage to change unless someone specially opted in to using it.
>
> That seems just as bad, if not worse.  Clang-format isn't an analysis tool, its a format tool.  If you have an option that can only reasonably be run in 'dry-run' mode, it seems that putting it in a 'format' tool is the wrong place.

This is exactly what the "llvm-premerge checks" do!, why can't it also be an analysis tool? your own usage scenario may not be the same as everyone elses, that doesn't make it wrong!

> If you have an option that can only reasonably be run in 'dry-run' mode

This isn't true, I successfully "east consted" all of clang causing no build errors and from what I can tell no unit test failure either, I'm just saying using it in dry-run mode is an equally useful usage scenario.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69764/new/

https://reviews.llvm.org/D69764