OKlibrary  0.2.1.6
MakeSystem.hpp File Reference

Plans for the core part of the build system, the make- or make-like engine. More...

Go to the source code of this file.


Detailed Description

Plans for the core part of the build system, the make- or make-like engine.

Todo:
Makepp
  • See http://makepp.sourceforge.net/ .
  • A disadvantage is the use of Perl, but perhaps soon with Git we also get this dependency.
  • However, currently the make-functions "call, error, warning" are not supported, so it seems we can't use it in its current form.
  • Otherwise its features (automatic scanning for include-files, a stronger system for hierarchical builds, handling of conditional compilation) should be very useful.
  • Though it seems to be a rather small project.
Todo:
DONE (we don't use CMake ourselves, and let the Linux distribution provide it) CMake
  • The following needs a complete update, considering whether CMake gained more power in these 3 years, and regarding the changes in our buildsystem.
  • The basic idea of using cmake despite of the problems described below (see also the e-mail to the CMake-mailing list from 7.6.2007): Use make to run the cmake-process! In Detail:
    • As now, every module where we wish to have build-functionality contains a symbolic link to a generic makefile (a new one --- just a very simple one (hopefully)). Perhaps we call this makefile the "master makefile".
    • Such a module contains a .build-subdirectory (seems better to hide it, since normally one doesn't want to look into it).
    • Creating a new directory is now a target for the master makefile: It creates the .build-directory, places inside it the cmake-file, and runs the cmake-process.
    • For the normal tasks like "all" or "check" the master makefile just calls the (created) makefile in .build.
    • While adding a new file is now also a task of the master makefile, which calls the cmake-makefile in .build (to recreate the makefile in .build).
    • For this to work we need the capability of cmake to place the cmake-makefile as well as the makefile somewhere else than in the source-directory.
    • Having the .build-directory in the source-tree creates considerable space overhead (relevant when copying the source-tree, as it happens when using it on the laptop), and likely it is not advisable to have build-information in the source-tree: So, given that we are free where to place the cmake-file, we should place the build-directory in system_directories/aux (or system_directories/build?).
    • What is the cmake-file in .build? Three possibilities:
      1. A copy of the generic cmake-file.
      2. A symbolic link to the generic cmake-file.
      3. A created cmake-file (created from a generic template and some parameters).
      Obviously, creating a customised cmake-file is most powerful, but one needs good reasons for that --- and a good design, with a clear idea about the different levels of indirection!
    • The aim is to use in this way all the build-in facilities of cmake, while overcoming its restricted power.
    • If cmake has not enough power for computing (and manipulating) file-lists, then the master makefile could compute them and pass them to cmake; a problem here could be that this must happen via the command line (?), and then there is the restricted command-line length?!?
    • The additional level of indirection (given by the master makefile) seems necessary, since we want complete automisation, while cmake itself does nothing than creating makefile --- and the makefiles created are not "dynamic" but "static" (rather restricted in power). The abstract division of work between the three layers can be understood as follows:
      1. The master-makefile-level does not do real building, but only its fileprocessing capabilities are needed (so actually a shell-script could be used?!).
      2. The cmake-level has the knowledge of how to build things.
      3. Finally, the created makefiles are purely static, and master dependencies (and, or course, actually perform the build-actions).
      One should think about using a shell-script instead of a master-makefile! The main disadvantage seems to be that the syntax "make target" couldn't be used.
  • Can cmake handle our "fractal" or "recursive" directory structure, where we have arbitrary nested subdirectories, inside we find standard functional (sub-sub-)directories like "tests", "testobjects", "demos", "apps", "messages", "plans", "docus", and from these functional directories the source-code files are to be extracted, and name-surgery is applied to compute the corresponding directories under system_directories ?!?
  • Can cmake handle the uniformity of our system, where for OKlib we only have 2 makefiles (OKlibBuilding/Makefile and Recursive.mak), and every directory contains (if wished) only a symbolic link to one of those (utilising then the definitions.mak files) ?!? We do not want to clutter the source directory with code for the build system, but at any point we want to be able to issue a make-command.
  • Can cmake handle the dynamic aspects of our build system, where with every make-command all the programs and libraries to be compiled and all executables to be executed are first determined dynamically, namely what exists currently in the source-code tree?!?
    1. The possibilities to manipulate variables seem very restricted?? There seem to exist no functions???
    2. What is needed is a "meta-make", which from configuration values, parameter values and the current state of the file-system first creates all targets, corresponding names and paths etc., and then creates the make-file. One would guess that this exactly should be possible with cmake?? With make we have the possibility due to the two-stages concept, but it's hard to create the dependencies on the target.
    3. So ADD_EXECUTABLE and ADD_LIBRARY depend on the output of a shell-command (typically "find") --- this doesn't seem possible? But with FOREACH and WHILE it should be possible.
  • It seems one doesn't have fine-grained control over the separation of the cmake-specific files and the library- and binary-files? Apparently cmake expects the user to issue a final "install" command, which will put the lib- and bin-files to their "final" destination?!? But we have a continuous process!
  • It seems that the basic idea between the higher-level "cmake" and the lower-level "make" is here, that every time a new files is introduced (or something similar), one must run cmake, while otherwise one can just run make. Seems not unreasonable, but of course one introduces a further possibility for making mistakes ?!?
  • It seems that the special makefile for external sources is the best candidate for employing cmake; see "Using CMake" in Buildsystem/ExternalSources/plans/general.hpp.
  • Investigate how our main tasks can be handled by cmake.
  • How would the work flow be with the new system (the 2 levels, the cmake- and the make-level)?
  • The cmake-files should use decomposition into smaller cmake-files (this should now be easier), and we also need a library of cmake-macros.
  • Is ccmake useful? DONE (just a program for setting configuration variables, so yes, it is useful for that purpose)
Todo:
Competition extraction tools
  • These tools, for a given program build a package, which contains all necessary source code and can be build using make (producing then the executable).
  • A problem here are the link-libraries. Without them, one could use the ability of g++ to produce compilation units (using the option "-E" like "g++ -E Program.cpp -o Program.ii"), putting everything into one big file.
  • Knowing the link-libraries, just compiling Program.ii with the right .o-files yields the program.
  • So when we know all compilation units, we can preprocess them all and putting them into the package; the package just compiles the units separately, and then links them together.

Definition in file MakeSystem.hpp.