[cmake-developers] ExternalProject: Use native paths as substitute for directory tokens

David Cole DLRdave at aol.com
Thu Aug 20 17:20:42 EDT 2015


It's exactly what I am concerned about:

You're asking to change the behavior of something for EVERYONE to
solve a problem which you have encountered. If you change it the way
you have proposed, you will cause problems for others. It has worked
the way it is now since ExternalProject_Add was introduced in CMake
2.8. Changing it unconditionally the way you propose is simply not
feasible for backwards compatibility.

I think commands that take native paths ought NOT to use the <*_DIR>
replacement values, and instead, ought to pass in variables that
contain the native paths in the first place.


David C.



On Thu, Aug 20, 2015 at 2:58 PM, James Johnston
<johnstonj.public at codenest.com> wrote:
> Hi,
>
> Funny you are mailing the list about this, since I just ran into this same issue today building something totally different from Boost...  In this case it's a build tool that thinks the "/U" in "C:/Users" is a new command line argument, that isn't recognized - and then the subsequent "s" also ends up unrecognized... and it all fails...  And it has nothing to do with the working directory, so _Add_Step(WORKING_DIRECTORY) isn't a possible workaround for me.
>
> I think the issue with globally making this change to the existing tokens is that there could be some external tool/program that is EXPECTING to get CMake paths, not native paths.  Who knows?  I am guessing that is what David Cole was concerned about.
>
> Maybe the right answer is to introduce some NEW tokens while leaving the behavior of the old ones unchanged.  E.g. <BINARY_DIR_NATIVE> etc.  It would be good if the patch updates the documentation of ExternalProject and clearly states the path format of <BINARY_DIR> vs <BINARY_DIR_NATIVE>.  Then the user can pick whichever one suits them best, depending on the tool being invoked.
>
> Furthermore, sometimes <BINARY_DIR_NATIVE> still needs to be replaced with a CMake path, not native path.  For example, if the token is being found in a property like WORKING_DIRECTORY that eventually gets passed to add_custom_command(WORKING_DIRECTORY) then I'm guessing still has to be a CMake path.  I am guessing this is what David Cole was also concerned about.
>
> I still think your original method of building Boost is a bit unusual and would be better served by _Add_Step with a custom working directory - because that's the publicly documented/standard way of changing the working directory, but that is up to you.  :)
>
> Best regards,
>
> James Johnston
>
>
> ---- On Thu, 20 Aug 2015 14:37:08 +0000  Stefan Kislinskiy <s.kislinskiy at Dkfz-Heidelberg.de> wrote ----
>
>  > Hi,
>  >
>  > thank you for our suggestions. I am aware that I can solve my example differently and that it might look not directly connected the proposal, but well, it is an example just to show a single case and why it matters. :) I did not want to discuss the example itself. Working around here would just resolve a symptom.
>  >
>  > My point is the overall problem that would persist: A big part of ExternalProject is to issue commands for predefined and custom steps. Those commands are supposed to be executed by the shell/command line. According to the documentation and the source code of ExternalProject, directory tokens are mainly supposed to be replaced in commands. It is my understanding, that it is a bug, if CMake isn't able to assemble these commands correctly. This would include usage of the correct path style of the OS for shell/command line commands. As directory tokens are replaced internally right before a shell/command line command is assembled, I can't see why this would be kind of "API-breaking". You cannot interfere in your CMake code with these internal replacements.
>  >
>  > Therefore I would still prefer my solution as it is pretty simple without adding even more features to ExternalProject and in my opinion without breaking code in the wild. It is a true bug fix instead of a feature request for working directories, which is a different topic that just coincidentally arised because of my specific example I guess. The features you described wouldn't fix the actual bug.
>  >
>  > As you were not sure if my approach would even fix my problems: It does of course and this is what I am currently doing and what I tested extensively before creating the patch. :) Regarding your quote from the add_custom_command documentation I can tell you that this is how things are currently done in ExternalProject and always were as far as I know, for example (from ExternalProject.cmake):
>  >
>  >   add_custom_command(
>  >     OUTPUT ${stamp_file}
>  >     BYPRODUCTS ${byproducts}
>  >     COMMENT ${comment}
>  >     COMMAND ${command}
>  >     COMMAND ${touch}
>  >     DEPENDS ${depends}
>  >     WORKING_DIRECTORY ${work_dir}
>  >     VERBATIM
>  >     )
>  >
>  > Best regards,
>  > Stefan
>  >
>  >
>  > -----Original Message-----
>  > From: cmake-developers [mailto:cmake-developers-bounces at cmake.org] On Behalf Of James Johnston
>  > Sent: Donnerstag, 20. August 2015 15:37
>  > To: cmake-developers at cmake.org
>  > Subject: Re: [cmake-developers] ExternalProject: Use native paths as substitute for directory tokens
>  >
>  > > -----Original Message-----
>  > > From: cmake-developers [mailto:cmake-developers-bounces at cmake.org]
>  > > On Behalf Of Kislinskiy, Stefan
>  > > Sent: Thursday, August 20, 2015 09:02
>  > > To: David Cole
>  > > Cc: cmake-developers at cmake.org
>  > > Subject: Re: [cmake-developers] ExternalProject: Use native paths as
>  > > substitute for directory tokens
>  > >
>  > > Hi David,
>  > >
>  > > Example excerpt (it is not possible to change the working directory
>  > > for
>  > the
>  > > CONFIGURE_COMMAND as it is fixed to the BUILD_DIR, which might not be
>  > > sufficient):
>  >
>  > This doesn't really directly have to do with your proposal, but what if an option was added to change the working dir of the CONFIGURE_COMMAND?  E.g.
>  > WORKING_DIRECTORY_CONFIGURE.  And suppose you'd have it recognize the various tags like <SOURCE_DIR>, etc.  This might be useful to add to other steps as well, and would be more portable than your solution which is using cmd.exe-specific commands.  You'd want to audit for any resulting breakage (e.g. does ExternalProject make assumptions that the working directory of CONFIGURE is always the binary dir? - e.g. a relative path being used somewhere.  And probably only allow specification of WORKING_DIRECTORY_CONFIGURE if a CONFIGURE_COMMAND was also specified, as the built-in commands certainly assume the default working dir.)
>  >
>  > In your situation though, I'm not sure it's strictly needed.  From your sample, it looks like you're building boost.  In your case what if you:
>  >
>  >  * Use ExternalProject_Add_Step to bootstrap.  You can specify a WORKING_DIRECTORY here.  Note one problem: you can't do out of source build of b2, which breaks user expectations.
>  >  * Then use ExternalProject_Add_Step to build Boost.
>  >
>  > Yes, using _Add_Step is somewhat of a workaround, but in this case, I've found it wasn't much of a burden at all.  In fact the only case I can think of where it WOULD be a burden would be if the configure step is CMake.  But then you wouldn't need to change the working directory; changing it would break CMake.  In practice nobody will want to change WORKING_DIRECTORY unless it's a custom command and then it's easy to use _Add_Step anyway.
>  > That said, it might still be considered a little undesired and so maybe my proposal above would be a better way to handle it.
>  >
>  > Corrections from maintainers and others on the above commentary are welcome...
>  >
>  > >
>  > > set(bootstrap_cmd "<SOURCE_DIR>/bootstrap${shell_ext}"
>  > > ${bootstrap_toolset})
>  > >
>  > > if(WIN32)
>  > >   set(bootstrap_cmd pushd "<SOURCE_DIR>" COMMAND ${bootstrap_cmd}
>  > > COMMAND popd)
>  > > endif()
>  > >
>  > > ExternalProject_Add(Boost
>  > >   ...
>  > >   CONFIGURE_COMMAND ${bootstrap_cmd}
>  > >   ...
>  > > )
>  >
>  > From add_custom_command:  "If more than one COMMAND is specified they will be executed in order, but not necessarily composed into a stateful shell or batch script."
>  >
>  > So I am not sure your approach will work for you even if you fix the issue with path slashes.
>  >
>  > James
>  >
>
> --
>
> Powered by www.kitware.com
>
> Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ
>
> Kitware offers various services to support the CMake community. For more information on each offering, please visit:
>
> CMake Support: http://cmake.org/cmake/help/support.html
> CMake Consulting: http://cmake.org/cmake/help/consulting.html
> CMake Training Courses: http://cmake.org/cmake/help/training.html
>
> Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>
> Follow this link to subscribe/unsubscribe:
> http://public.kitware.com/mailman/listinfo/cmake-developers


More information about the cmake-developers mailing list