Skip to content

[Rcpp-devel] Stripping symbols and Cran

8 messages · Michaël BENESTY, Dan Dillon, Qiang Kou +2 more

Original
Michaël BENESTY
# 
Hi all,

Because of (the great) Rcpp, the size of my package was over 5Mb, which was
an issue for a Cran release.

I read with interest this article
http://dirk.eddelbuettel.com/blog/2017/08/14/#009_compact_shared_libraries

In particular, I was interested in the Makevars modification:

strippedLib: $(SHLIB)
        if test -e "/usr/bin/strip"; then /usr/bin/strip --strip-debug
$(SHLIB); fi

Unfortunately it crashed compilation on Mac OS because for some reason
--strip-debug doesn't exist.

But this small modification works:
strippedLib: $(SHLIB)
        if test -e "/usr/bin/strip" & test -e "/bin/uname" & [[
`uname` == "Linux" ]] ; then /usr/bin/strip --strip-debug $(SHLIB); fi

May be there is a way to make strip works correctly on Mac, but as I
have no size issue on both Windows and Mac OS... I don't care.

I thought it was useful enough to share it with you.

Kind regards,

________________
Micha?l BENESTY
michael at benesty.fr
Please do not print this e-mail unless you really need to.
Dirk Eddelbuettel
# 
Michael,
On 18 September 2017 at 16:55, Micha?l BENESTY wrote:
| Because of (the great) Rcpp, the size of my package was over 5Mb, which was
| an issue for a Cran release.

Are you the OP behind the StackOverflow question here:
https://stackoverflow.com/questions/46280628/object-files-in-r-package-too-large-rcpp

If so we already tried to answer. This is NOT an issue for CRAN but merely a
warning.  Do yourself a quick favour and run

  file.info( system.file("libs", "dplyr.so", package="dplyr") )$size

and you see that some well-known and widely used packages are MUCH bigger.

| I read with interest this article
| http://dirk.eddelbuettel.com/blog/2017/08/14/#009_compact_shared_libraries
| 
| In particular, I was interested in the Makevars modification:
| 
| strippedLib: $(SHLIB)
|         if test -e "/usr/bin/strip"; then /usr/bin/strip --strip-debug
| $(SHLIB); fi
| 
| Unfortunately it crashed compilation on Mac OS because for some reason
| --strip-debug doesn't exist.

Interesting.

It so happens that I just shipped a use of `strip --strip-unneeded` in
RcppClassic, see https://github.com/eddelbuettel/rcppclassic/blob/7cc5c626952d38cc97a99363131fe7ac9d0a7b28/src/Makevars#L33
Does --strip-unneeded work on macOS ?
 
| But this small modification works:
| strippedLib: $(SHLIB)
|         if test -e "/usr/bin/strip" & test -e "/bin/uname" & [[
| `uname` == "Linux" ]] ; then /usr/bin/strip --strip-debug $(SHLIB); fi

Maybe --strip-unneeded is for preferable. Otherwise, yes, one needs to check
for Linux here.

But that is the crux of these Makefile / Makevars tweaks. They get
non-portable real quickly.

The easiest may just be to NOT worry, and just use -W,-S in the local
~/.R/Makevars.  That is what I now do.

| May be there is a way to make strip works correctly on Mac, but as I
| have no size issue on both Windows and Mac OS... I don't care.
| 
| I thought it was useful enough to share it with you.

It is useful.  Dan worked quite a bit on this too. See his repo at

   https://github.com/dcdillon/r-stripper

Dirk
Michaël BENESTY
# 
Hi Dirk,

I am not the author of the question on SO, but the timing is perfect :-)
I am aware of Dplyr size (I read ur blog carefully!!!) but really
wanted to have 0 note.

Just tried --strip-unneeded : not the silver bullet :-(

#> error: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/strip:
unrecognized option: --strip-unneeded


Kind regards,
________________
Micha?l BENESTY
michael at benesty.fr
Please do not print this e-mail unless you really need to.
On Mon, Sep 18, 2017 at 5:19 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
Dan Dillon
# 
What is the output of strip --help (or equivalent) on macOS?  Surely there
is some option that does the same thing.

This doesn't solve your desire for no NOTEs on CRAN because to do it right
you'd have to detect what strip you were calling and pass the right option
to each variant, but it could point you in the right direction.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/attachments/20170918/28bd7ef7/attachment.html>
Dirk Eddelbuettel
# 
On balance, I think you should ask on r-package-devel.

Rcpp cannot help you here as R does not offer portable and reliable hooks. I
recommend (and I am dead serious) to just live with the note.

Dirk
Qiang Kou
# 
On OSX, "-strip-all" or "-strip-debug" should work.

See [1] for more information.

Best,

Qiang Kou

[1] http://releases.llvm.org/2.9/docs/CommandGuide/html/llvm-ld.html
On Mon, Sep 18, 2017 at 11:42 AM, Dan Dillon <dcdillon at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
Note that macOS doesn't include debugging info in the objects so stripping is meaningless there.
I wasn't on the full thread - which package are we talking about?

Cheers,
Simon

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Dirk Eddelbuettel
    

    

      
      #
    

  


  

      
On 18 September 2017 at 17:46, Simon Urbanek wrote:

        
| Note that macOS doesn't include debugging info in the objects so stripping is meaningless there.
| I wasn't on the full thread - which package are we talking about?

Michael started the thread generally musing about large sizes of object files
from C++.  And he has a point :)

And he found my blog post showing the (absurd !!) gains from stripping, tried
it on macOS and had issues.  It so happens that I just updated the stone-old
RcppClassic on the weekend, and in my infinite wisdom include such a call to
strip with the unavailable-on-macOS option :-/   Oops. I broke my build.

Fix coming to a CRAN mirror near you, likely testing for /usr/bin/strip AND
being on Linux.

Dirk