Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.mrao.cam.ac.uk/~bn204/oof/software-architecture.html
Дата изменения: Mon Apr 4 13:47:52 2016
Дата индексирования: Sun Apr 10 07:40:07 2016
Кодировка: ISO8859-5

Поисковые слова: п п п п п п п п п п п п п п п п п п п п п п п
Description of the OOF software architecture — Bojan Nikolic web pages (r. 329)

Description of the OOF software architectureТЖ

OverviewТЖ

The design of the OOF software is driven in part by the desire that a single system serves as both the development environment for new algorithms and investigation of new data sets and as the production environment which is used in routine on-line and off-line data reduction.

The architecture of the OOF data reduction software can be summarised in following points:

  1. The implementation is in modern C++ and Python, with the binding between the two automatically generated by the SWIG program
  2. The functionality is split between five relatively loosely coupled modules which allow re-use of functionality in other projects
  3. A well defined set of external libraries is used and generally newer versions of these libraries are adopted quickly after release
  4. The build system is implemented using the autoconf/automake/autotools tool chain
  5. Revision control is carried out using the Bazaar system

Mixed C++ and Python implementationТЖ

This mixed language model is an essential feature of the package that makes development and extending the present system much easier.

The functionality is divided between C++ and Python in the following way:

  • The C++ layer provides the bulk of the numerical processing routines, including:

    • Interfacing to FFT libraries (FFTW), implementing far-field to aperture plane transformations and convolutions
    • Implementation of the Zernike polynomials and their caching
    • Implementation of specialised 2d interpolations, again with caching
    • Implementations of models of aperture-plane illumination
    • Routines for simulating OOF data-sets

    Also implemented in the C++ layer are interfaces to some required libraries, such as:

    • I/O of FITS file binary tables (using cfitsio)
    • Plotting (pgplot)
  • The Python provides user-facing wrapper routines:

    • Driver scripts for reduction of entire data sets
    • Writing and reading of results of analysis
    • Automated plotting

The use of Python however also allows following very important facilities:

  1. Efficient and ergonomic inspection of the state of C++ objects at any intermediate stage of computation

  2. “On-line” decision on the algorithms to be run through use of virtual functions.

    For example, the model for the aperture-plane consists of a model of the illumination and of the phase. Each of these is specified as a pointer to an object which is a sub-class of the right base-class (one for each of these two models). The subsequent minimisation calls virtual functions provided by these objects.

Loosely coupled modulesТЖ

The rationale for adopting multiple module architecture is simply that the OOF holography application was able to re-use some code originally developed for other applications, while it was also clear that some of the code developed for OOF will be (and now is) useful for other, unrelated applications.

Therefore the functionality is split (as well as I could do it) along domain lines as follows:

  • BNLib – basic C++ support functionality and interfaces to the GSL
  • BNFits – A C++ wrapper around FITSIO libraries
  • Astromap – facilities for astronomical maps (i.e., 2d arrays with attached coordinate systems). This includes:
    • Simple manipulations (add, multiply, fill with random numbers, simple statistics)
    • FFTs and convolution
    • Interpolation
    • Publication quality plotting (via pgplot)
  • BNMin1 – minimisation, non-linear least-squares fitting and general statistical inference (now includes MCMC and Nested Sampling)
  • OOF – Out-of-focus specific algorithms and code

External librariesТЖ

The following libraries are used:

  1. Boost
  2. GSL
  3. PGPLOT
  4. CFitsIO
  5. PyFITS
  6. Numpy (originally numarray)

Build systemТЖ

The de-facto standard Autotools build system is used and has been found to work well on all Linux installations.

Estimate of the sizeТЖ

At the time of writing, the modules that make up the OOF data reduction software contain 12000 lines of code (this excludes auto-generated files, comments and whitespace lines). Of that approximately 9000 are in C++, 2500 are in Python, and the remainder are Makefile and configure.ac scripts.

Obviously, it should be pointed out that some of this code relates to features quite unconnected with OOF data reduction and some of it relates to features that were developed in connection with OOF but are not in production use.