repover.txt

开放源码的编译器open watcom 1.6.0版的源代码

Repository Use in Program Development
=====================================

Potential User Benefits:
------------------------

* faster turn-around during the program-development cycle

* better code optimization

* increased functionality: version control, group control



Potential Detailed Benefits:
----------------------------

* reduce storage redundancy of compiled files ==> faster compilation and
  linking
  
* reduce # of compilations/links following an edit

* reduce link time by providing information for incremental linking

* better optimization by providing information about items not fully
  defined in a compilation unit
  
* increased functionality:
    * versioning should be pervasive across all aspects of repository
    * group control: protection across group of developers (like PVCS, RCS)
        * should have project, sub-project capabilities
    * should also apply to sources
      
      
Background (compilation and linking):
-------------------------------------

* current technology emits in object file:
    - compiled code
    - debugging information
    - browsing information (C++)
    - generated functions (C++)
    - internal data for included definitions (C++)
    
* all of above can be redundant (present in many object files) when they
  pertain to items shared (by inclusion directives) by a number of source
  modules
  
* result is slowed compilation and linking when redundant information is
  produced or processed
  
* a compiler can query the repository to decide that redundant information
  need not be generated during a specific compilation
  
* a linker processes less information by the elimination of redundancy
    - using a repository may consume less system resource than conventional
      file-based storage of object files

* while optimizing, a compiler can query the repository about items compiled
  in other compilation units, resulting in better code


Background (Make engine):
-------------------------

* conventional technology requires dependency rules to be file based
    - hand-entered for small to medimum C projects and for non-compiler
      dependencies
    - automatically generated (based on inclusion) for larger or more complex
      projects
    - file dates are used to decide when re-compilation is required
      
* when automatic generation is used, many files get re-compiled needlessly,
  because of the grossness of the dependency rule
  
* compiler can add dependencies (definitions and uses) to the repository to
  build a finer set of rules for processing by a Make engine
  - e.g., compiler adds item during source compilation:
            "used field x from structure y with offset z"
          compiler adds item for header file, following an edit:
            " field x in structure y has offset Z"
          source file has to recompliled if z is different than Z

* result will be recompilation of fewer source files
 
* note: mechanism can apply to any compilation system, not only C, C++