[lld] r207141 - [PECOFF] Define implicit symbols for exported ones.

[Rui Ueyama]

On Thu, Apr 24, 2014 at 2:38 PM, Nick Kledzik <kledzik at apple.com> wrote:

>
> On Apr 24, 2014, at 1:57 PM, Rui Ueyama wrote:
>
> On Thu, Apr 24, 2014 at 1:29 PM, Chandler Carruth <chandlerc at google.com>wrote:
>
>>
>> On Thu, Apr 24, 2014 at 1:12 PM, Rui Ueyama <ruiu at google.com> wrote:
>>
>>> Using __imp_ symbols locally is I think not a good coding style. One
>>> should just take an address using "&" operator rather than appending
>>> __imp_ prefix. However, there are programs in the wild that depends
>>> on this link.exe's behavior, so we need this feature.
>>>
>>
>> I wonder if we should have a warning about this?
>>
>
> Ideally, yes, but there's no good way to print a warning message when some
> symbol is actually linked. In this patch I create __imp_ symbols
> unconditionally for all dllexported symbols. Creating them is not visible
> to user -- using them should be warned.
>
> It is possible to create the __imp_ atoms lazily.  You have to add a
> linker internal File object which acts like a static archive. It only
> returns atoms when asked for an __imp_ symbol.  The File object could then
> warn when an ImpPointerAtom is instantiated.
>
> I'm not sure if the warning is worth the extra machinery this would
> require.  Is there a (size) cost to always generating all
> the ImpPointerAtom atoms?  The lazy way only instantiates ones that are
> used.
>

The cost of eagerly generating __imp_ symbols is low -- they are created
only for symbols with __declspec(dllexport).


>
> -Nick
>
>
>
> I also wonder whether the cause in the wild is source code being shared
>> between the DLL and the users of the DLL and being written "generically".
>>
>
> The program I spotted in the wild is ffmpeg. That has /include:__imp_*sym* where
> *sym* is a defined function in ffmpeg (/include is equivalent to
> --undefined). Not sure why it has such option as it is basically no-op to
> link.exe.
>
> I don't think writing code generically cannot really justify to use __imp_
> symbols, as you can just use "&", so I'm still wondering what this feature
> is really for...
>  _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140424/d0f56d50/attachment.html>