[llvm] r200858 - Revert "Fix an invalid check for duplicate option categories."

Alexander Kornienko alexfh at google.com
Fri Feb 7 09:49:05 PST 2014


On Fri, Feb 7, 2014 at 6:29 PM, Jordan Rose <jordan_rose at apple.com> wrote:

>
> On Feb 6, 2014, at 1:50 , Alexander Kornienko <alexfh at google.com> wrote:
>
> On Thu, Feb 6, 2014 at 5:03 AM, Jordan Rose <jordan_rose at apple.com> wrote:
>
>> It would actually make sense for the plugin not to link LLVM libraries
>> known to be in Clang, but on the flip side it would also make sense for the
>> general option category to not be a static variable.
>>
>> <rant>Honestly, my personal opinion is that the whole CommandLine library
>> is really messed up, and is using global variables and static
>> initialization in pretty much exactly the ways a C++ style guide would tell
>> you not to. :-) (You'll notice how I made a stack-based cl::Option variant
>> when adding unit tests for my recent change.)</rant>
>>
>> Okay, rant aside, I think we should get loadable bundles building without
>> linking in LLVM libraries. That doesn't help if two plugins *share* a
>> static library that the main executable doesn't include, but Support
>> certainly isn't going to be one of those.
>>
>
> Are you going to take care of the plugin rework? Is it fine to go with the
> intermediate solution in the meanwhile? It's strictly better, than the
> status quo, as it resolves crashes when using -help in the
> assertions-enabled build with a certain STL implementation.
>
>
> *sigh* Yes, all right. :-)
>
> (The *sigh* is that I haven't touched those examples in over a year.)
>

Thanks!
Committed revision 200981.


> Jordan
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140207/c58b6dde/attachment.html>


More information about the llvm-commits mailing list