[LLVMdev] Optimization of string.h calls

[David Blaikie]

On Sat, Sep 20, 2014 at 12:31 PM, Iritscen <iritscen at yahoo.com> wrote:

> Yes, well that is a good point w.r.t. strings as typically passed to
> string.h functions.  I guess in this case I had figured that the use of
> __FILE__ could easily be optimized since it becomes a string constant.  Am
> I wrong?
>

I'm not sure I follow.

Imagine the following:

First file:
  void f1(const char* c) {
    puts(c - 4);
  }

Second file:
  void f2() {
    f1("the name" + 4);
  }

The compiler can't optimize away the 'the ' prefix while compiling the
second file because it can't know whether the users of that pointer might
subtract from it, walking into well defined memory they should be able to
examine/print/etc.


>  Is the issue that __FILE__ might occur multiple times in a source file
> and thus become a merged constant?  Perhaps there is simply not much use
> for a string-optimizing feature since reducing program size isn’t a goal
> for most developers these days?
>
> Then, more to the point, I wonder if there shouldn’t be a macro in
> clang/llvm for returning only the current source file name, e.g.
> __FILE_NAME__.  I don’t know what the feeling is on introducing macros that
> do not exist in gcc, since it seems that up until now the approach has been
> quite conservative.  Does no one else have a desire to easily suppress the
> full hard drive paths that will show up in their binaries with the use of
> __FILE__?  Or perhaps everyone simply has their own existing solutions to
> this, such as a build system that can pass clang the files by relative
> path, or placing the source code on its own volume?
>

Yeah, I forget/don't recall how we solve this in-house, but I'm sure we
have a way - perhaps someone else will chime in.


>
>
> On Sep 20, 2014, at 10:41 AM, David Blaikie <dblaikie at gmail.com> wrote:
>
> I imagine that would be difficult. If you pass the resulting pointer out
> to unknown code/functions, there's nothing to stop that code from walking
> backwards from the pointer and having well defined behavior of observing
> the prefix you tried to hide.
>
> Essentially the compiler would have to be able to see and analyze all uses
> of that pointer in one go before it could optimize away the prefix.
> On Sep 19, 2014 8:41 PM, "Iritscen" <iritscen at yahoo.com> wrote:
>
>> Hello all.  Is there a way to get llvm/clang at build-time to optimize a
>> string.h call so that the final form of the string is saved in the binary?
>> For instance, for a statement like...
>>
>> strrchr(__FILE__, '/') + 1
>>
>> …and where clang is called on this code’s source file with “clang
>> /some/long/path/file.c”, can I end up with only “file.c” stored in the
>> binary on disk?  I think you can see that I am attempting to avoid full
>> paths from my machine ending up in the program.  My IDE, Xcode, always
>> passes full paths to clang when building.  Thanks.
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140920/1038c02d/attachment.html>