Make biblatex treat different author names as same author in text citations

Question

Some guy has written a few books, but he is inconsistent from book to book when it comes to including and/or abbreviating his middle name. When I cite this author in the text, I would like biblatex to treat all different versions of his name as the same author. I can enforce this for the sorting in the bibliography with sortname, but what do I do to enforce this for text citations?

The following example illustrates how the text citation treats "Paul J[ohn] Smith" as a different author from "Paul John Smith" and "Paul Smith". The first citation should appear as (Smith 2002), and the second citation should appear as (2001a, 2001b, 2002).

\documentclass{article}
\usepackage{csquotes}
\usepackage[
        bibstyle = authoryear,
        citestyle = authoryear-comp,
        dashed = false,
        sorting = nyt,
        sortcites = false,
        language = american,
        abbreviate = false,
        backend = biber]{biblatex}
\usepackage{filecontents}
\begin{filecontents}{\jobname.bib}
@BOOK{smith2001a,
    AUTHOR = "Paul John Smith",
    TITLE = "My first book",  
    YEAR = "2001"}

@BOOK{smith2001b,
    AUTHOR = "Paul Smith",
    TITLE = "My second book",  
    YEAR = "2001",
    SORTNAME = "Paul John Smith"}

@BOOK{smith2002,
    AUTHOR = "Paul J[ohn] Smith",
    TITLE = "My third book",  
    YEAR = "2002",
    SORTNAME = "Paul John Smith"}
\end{filecontents}
\addbibresource{\jobname.bib}
\begin{document}
\noindent
Some clever guy said that \parencite{smith2002}.
In fact, Paul Smith has talked about this several times \parencite*{smith2001a,smith2001b,smith2002}.
\printbibliography
\end{document}

Answer 1

There is now a biblatex macro interface to the sourcemapping feature, you can put this in your preamble:

\DeclareSourcemap{
  \maps[datatype=bibtex]{
    \map[overwrite]{
       \step[fieldsource=author, match=\regexp{Paul\s+(?:J\S+\s+)?Smith}, final]
       \step[fieldset=shortauthor, fieldvalue={Smith, Paul John}]
       \step[fieldset=sortname, fieldvalue={Smith, Paul John}]
    }
  }
}

You may want to tune the regular expression, I made it quite specific so as not to catch anything it shouldn't.

Answer 2

This is just to show how PLK's solution above does the trick in practice:

\documentclass{article}
\usepackage{csquotes}
\usepackage[
        bibstyle = authoryear,
        citestyle = authoryear-comp,
        dashed = false,
        sorting = nyt,
        sortcites = false,
        language = american,
        abbreviate = false,
        backend = biber]{biblatex}
\usepackage{filecontents}
\begin{filecontents}{\jobname.bib}
@BOOK{smith2001a,
    AUTHOR = "Paul John Smith",
    TITLE = "My first book",  
    YEAR = "2001"}

@BOOK{smith2001b,
    AUTHOR = "Paul Smith",
    TITLE = "My second book",  
    YEAR = "2001",
    SORTNAME = "Paul John Smith"}

@BOOK{smith2002,
    AUTHOR = "Paul J[ohn] Smith",
    TITLE = "My third book",  
    YEAR = "2002",
    SORTNAME = "Paul John Smith"}
\end{filecontents}
\addbibresource{\jobname.bib}
\DeclareSourcemap{
  \maps[datatype=bibtex]{
    \map[overwrite]{
       \step[fieldsource=author, match=\regexp{Paul\s+(?:J\S+\s+)?Smith}, final]
       \step[fieldset=shortauthor, fieldvalue={Smith, Paul John}]
       \step[fieldset=sortname, fieldvalue={Smith, Paul John}]
    }
  }
}
\begin{document}
\noindent
Some clever guy said that \parencite{smith2002}.
In fact, Paul Smith has talked about this several times \parencite*{smith2001a,smith2001b,smith2002}.
\printbibliography
\end{document}