pgfkeys doesn't play well with edef keys

[pgf-tikz-bot]

Migrated from SourceForge
Author: mirandole
Timestamp: 2014-04-06 10:24:37.816000

Example :

\documentclass {minimal}
\usepackage [utf8]{inputenc}
\usepackage {etex}
\usepackage {pgfkeys}

\loggingall

\pgfkeysedef {/cle}{\noexpand \macro }
\pgfkeys {/cle/.append code=\MACRO }

\expandafter \show \csname pgfk@/cle/.@cmd\endcsname

% \pgfk@/cle/.@cmd  is  "\noexpand \macro \MACRO "

confer pgfkeys.code.tex line 608 :

\long\def\pgfkeysedef#1#2{%
  \long\edef\pgfkeys@temp##1\pgfeov{#2}%
  \pgfkeyslet{#1/.@cmd}{\pgfkeys@temp}%
  \pgfkeyssetvalue{#1/.@body}{#2}%
} 

/.@body is NOT expanded in the definition !!

[pgf-tikz-bot]

Migrated from SourceForge
Author: gaborit
Timestamp: 2017-01-05 14:18:40.931000

https://sourceforge.net/p/pgf/bugs/_discuss/thread/3fba4423/f5f2/attachment/patch.txt
Here is my proposed patch on pgfkeys.code.tex (from v3.0.1a).

I add the \pgfkeyssetevalue macro and modify the \pgfkeysedef and \pgfkeysedefargs macros to use it.

PS: you may thank wipet on stackexchange that finds the way to define \pgfkeyssetevalue (http://tex.stackexchange.com/q/346779/14500).

[pgf-tikz-bot]

Migrated from SourceForge
Author: gaborit
Timestamp: 2017-01-08 00:38:18.074000

https://sourceforge.net/p/pgf/bugs/_discuss/thread/3fba4423/b608/attachment/pgfkeys-tests.tex
https://sourceforge.net/p/pgf/bugs/_discuss/thread/3fba4423/b608/attachment/pgfkeys.code.tex.patch
Here is a better patch.

PS: with a test document.

[pgf-tikz-bot]

Migrated from SourceForge
Author: gaborit
Timestamp: 2017-01-09 15:48:05.858000

https://sourceforge.net/p/pgf/bugs/_discuss/thread/3fba4423/a81a/attachment/pgfkeys.code.tex
Here is my last proposition for pgfkeys.code.tex. I provide the entire file.

[pgf-tikz-bot]

Migrated from SourceForge
Author: mo-gul
Timestamp: 2018-12-24 11:42:05.018000

• status: open --> closed-fixed

[pgf-tikz-bot]

Migrated from SourceForge
Author: mo-gul
Timestamp: 2018-12-24 11:45:06.019000

Thank you for reporting and providing the patch. It is included now.

[polgab]

Here my new test file (with #669 example) with .txt extension as github don't support .tex attached file!
pgfkeys-tests.txt

Can you reopen this issue?

Thanks.

[hmenke]

Do you also have actual use-cases? Most of the stuff that you sent looks very contrived.

[polgab]

You are right!
This particular issue seems fixed so you can close it.
I'm sure there are still bugs or limits that need to be fixed or, at a minimum, documented. I will look for good use-cases and then I will submit a new issue.

[hmenke]

Open issues don't look nice on the bug tracker but they don't hurt. I'll leave this open so we don't forget about it.

[muzimuzhi]

(Not sure if I should open a new issue, but since this one is open ...)

In \pgfkeyssetevalue, we actually want to define \pgfk@#1/.@body to be the expanded result of #2, with doubling of hashes unchanged.

• The current implementation (v3.1.5b) eliminates the doubled hashes, hence \pgfkeys{/ekey/.ecode={\def\noexpand\test##1{<##1>}}} defines \pgfk@/ekey/.@body to be \def\test#1{#1}, which should be \def\test##1{##1}.
• \expanded meets our requirement, but it is not fully supported by various engines until texlive 2019.

  % c.f. https://tex.stackexchange.com/a/509706
  \long\def\pgfkeyssetevalue#1#2{%
    \expandafter\edef\csname pgfk@#1\endcsname
    {\unexpanded\expandafter{\expanded{#2}}}%
  }

• Luckily, \expanded can be emulated. Both the LaTeX3 team (see texdoc source3, sec. 6.6) and Ulrich Diez on TeX-SX (see this answer) have provided such emulation.
• The new implementation of \pgfkeyssetevalue using \expanded passes the pgfkeys-tests.txt in pgfkeys doesn't play well with edef keys #305 (comment). I have not read and tested the emulated \expanded yet (I will, in the next couple of days).

---

A full example demonstrating the problem in current implementation and the effect of \expanded:

Example

\documentclass{article}
\usepackage{geometry}
\usepackage{xcolor}
\usepackage{pgfkeys}

% helper macros
% execute \pgfkeys{#1}, then inspect #2/.@cmd and #2/.@body
\def\showKey#1#2{
  \pgfkeys{#1}
  \begingroup
  \large\bfseries After \textcolor{key}{\detokenize{#1}}
  \endgroup
  \par
  \smallskip
  \showcs{pgfk@#2/.@cmd}
  \showcs{pgfk@#2/.@body}
  \medskip
}

\def\showcs#1{
  \makebox[100pt][l]{\expandafter\string\csname #1\endcsname} =
  \expandafter\meaning\csname #1\endcsname\par
}

\begin{document}
\parindent=0pt
\ttfamily

\section{Handler \texttt{.code}}
\colorlet{key}{orange!90!black}

\showKey{/key/.code={\def\test##1{<##1>}}}{/key}
\showKey{/key/.add code={}{\test{#1}}}{/key}

\section{Handler \texttt{.ecode}, current implementation}
\colorlet{key}{magenta!90!black}

\showKey{/ekey/.ecode={\def\noexpand\test##1{<##1>}}}{/ekey}
\showKey{/ekey/.add code={}{\test{#1}}}{/ekey}

% c.f. https://tex.stackexchange.com/a/509706
\long\def\pgfkeyssetevalue#1#2{%
  \expandafter\edef\csname pgfk@#1\endcsname
  {\unexpanded\expandafter{\expanded{#2}}}%
}

\section{Handler \texttt{.ecode}, new implementation using \texttt{\string\expanded}}
\colorlet{key}{cyan!90!black}

\showKey{/ekey/.ecode={\def\noexpand\test##1{<##1>}}}{/ekey}
\showKey{/ekey/.add code={}{\test{#1}}}{/ekey}

\end{document}

[hmenke]

Yes, this issue has been a source of a lot of headache. I have thought about using \expanded here previously, which is the only proper solution, but have so far held off on it for the compatibility reasons you mentioned. Maybe I'll ship that for TeX Live 2021.

[muzimuzhi]

Perhaps the compatibility reason is weaker than you thought, because it only happens when a user uses

• an engine other than luatex (this is common), and
• an engine prior to texlive 2019 (this is not rare), and
• an updated pgf (this is rare).

To sum up, I think most users who upgrade pgf, will also have latest tex engines installed. And users who have old engines installed, will most likely not solely upgrade package(s).