reuse.tex

The implementation of perfprof-py is separated in a way that facilitates the creation of a
new backend.
The class {\tt Pdata} is defined to store the parsed data ($\mathcal{P}$,
$\mathcal{S}$, $t_{s,p}$, etc.) and  methods are defined to create the profile data
$r_{p,s}$.
Backends are classes that extend {\tt Pdata} defining a method {\tt plot}
which creates the expected figure.
One should have little difficulty creating their own backend, specially if one
uses a perprof-py backend as a starting point.
However, if one wants to change the profile data definition --- in order
to implement a data profile (see \cite{bib:more2009benchmarking}) ---, one would have to modify one or more methods in {\tt Pdata} directly or re-implement the backends.

The parser opens the input files and creates the information for {\tt Pdata}.
Replacing this parser --- to use with perprof-py backends --- would not be an easy task
since the correct output format should be created.  Nevertheless,
extending it with additional options would be simple enough.

The entry point {\tt perprof-py} essentially collects the options from the command
line and calls the specific backend profiler. This can be completely bypassed
by calling the backend directly. This allows one to create a performance
profile from another python application. In particular, a possibility is the
creation of a graphical user interface (GUI)
or a web server application. Perprof-py modularity 
allows whoever desires to construct this interface to focus entirely on
obtaining the options from the user and passing it to the backend.

Whether one is planning on expanding some of perprof-py functionalities or creating any
new backend or interface, one can contact the authors using the project page on
GitHub~\cite{url:perprof-py}.