[pgffor] "remember=\x as \lastx" without "(initially ...)" reports \lastx undefined

[muzimuzhi]

Brief outline of the bug
The pgf manual (3.1.5b, sec. 89, p. 1004) says

By default the value of <variable> is zero for the first iteration, however, the optional (initially <value>) statement, allows the <macro> to be initially defined as <value>.

In my understanding, if (initially <value>) is missing, <macro> is initially defined as 0. But <macro> is reported to be undefined.

Minimal working example (MWE)

\documentclass{article}
\usepackage{pgffor}

\begin{document}
% based on the example in doc of option "remember", with "(initially A)" removed
\foreach \x [remember=\x as \lastx] in {B,...,H}
  {$\overrightarrow{\lastx\x}$, }
\end{document}

The above example reports

! Undefined control sequence.
<argument> \lastx 

[hmenke]

Possible patch (needs testing)

--- a/tex/generic/pgf/utilities/pgffor.code.tex
+++ b/tex/generic/pgf/utilities/pgffor.code.tex
@@ -737,6 +737,7 @@
     \pgfutil@append@tomacro{\pgffor@remember@code}{\noexpand\def\noexpand#2{#2}}%
     \def\pgffor@test{#4}%
     \ifx\pgffor@test\pgfutil@empty%
+        \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{0}}%
     \else%
         \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{#4}}%
     \fi%

[hmenke]

I think this was broken by this commit e22a1aa which means it has been broken for 10 years.

[hmenke]

Interesting, I discovered an undocumented syntax in there:

\documentclass{article}
\usepackage{pgffor}
\begin{document}
\def\lastx{A}
\foreach \x/\y [remember={\lastx=\x}] in {B,...,H}
  {$\overrightarrow{\lastx\x}$, }
\end{document}

But that really doesn't work so well. The following behaves a little bit funky:

\documentclass{article}
\usepackage{pgffor}
\begin{document}
\def\lastx{A}
\def\lasty{0}
\foreach \x/\y [remember={\lastx=\x;\lasty=\y;}] in {B/1,...,H/6}
  {$\overrightarrow{\lastx\x\lasty\y}$, }
\end{document}

In fact, if I write \lastx=\x; \lasty=\y; (note the space before \lasty) it blows up with

! Missing control sequence inserted.
<inserted text> 
                \inaccessible 
l.10   {$\overrightarrow{\lastx\x\lasty\y}$, }
                                              
?

[muzimuzhi]

Possible patch (needs testing)

--- a/tex/generic/pgf/utilities/pgffor.code.tex
+++ b/tex/generic/pgf/utilities/pgffor.code.tex
@@ -737,6 +737,7 @@
     \pgfutil@append@tomacro{\pgffor@remember@code}{\noexpand\def\noexpand#2{#2}}%
     \def\pgffor@test{#4}%
     \ifx\pgffor@test\pgfutil@empty%
+        \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{0}}%
     \else%
         \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{#4}}%
     \fi%

The current fix 08041e4 (additionally) allows input with empty <value> in (initially <value>), see

\documentclass{article}
\usepackage{pgffor}

\begin{document}
% \lastx is initialized to zero as well
\foreach \x [remember=\x as \lastx (initially )] in {B,...,H}
  {$\overrightarrow{\lastx\x}$, }
\end{document}

That might not be the expected effect and will make the logic too complex to be fully documented. The following attempt avoids that:

--- a/tex/generic/pgf/utilities/pgffor.code.tex
+++ b/tex/generic/pgf/utilities/pgffor.code.tex
@@ -727,7 +727,7 @@
 }

 \def\pgffor@remember@parse@old#1#2\pgffor@stop{%
-    \pgffor@remember@@parse@old#1#2\pgffor@stop as#1(initially )\pgffor@@stop}
+    \pgffor@remember@@parse@old#1#2\pgffor@stop as#1(initially 0)\pgffor@@stop}

 \def\pgffor@remember@@parse@old#1#2as#3#4\pgffor@@stop{%
     \pgffor@remember@@@parse@old{#1}{#3}#2#4\pgffor@stop\pgffor@@stop}

[hmenke]

Your proposed patch is much better. I have queued it for inclusion.

[muzimuzhi]

Many thanks. :)

[muzimuzhi]

A follow-up question: Do we need to allow \lastx to be manually initialized to empty?

• In \foreach \x in <list>, empty item is not removed from <list>, hence \x is allowed to be empty for corresponding iteration(s).
• In \foreach \x[remember=\x as \lastx (initially )] in <list>, \lastx is undefined for the first iteration, rather than initialized to empty.

If \lastx is allowed to be manually initialized to empty, then the following attempt is helpful:

--- a/tex/generic/pgf/utilities/pgffor.code.tex
+++ b/tex/generic/pgf/utilities/pgffor.code.tex
@@ -735,11 +735,7 @@
 \def\pgffor@remember@@@parse@old#1#2#3(initially #4)#5\pgffor@stop#6\pgffor@@stop{%
     \pgfutil@append@tomacro{\pgffor@assign@after@code}{\edef#2{#1}}%
     \pgfutil@append@tomacro{\pgffor@remember@code}{\noexpand\def\noexpand#2{#2}}%
-    \def\pgffor@test{#4}%
-    \ifx\pgffor@test\pgfutil@empty%
-    \else%
-        \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{#4}}%
-    \fi%
+    \pgfutil@append@tomacro{\pgffor@assign@once@code}{\def#2{#4}}%
 }

This changes the logic to (in pseudo-code)

if (initially <value>) is missing
    use 0 for <value>
else
    use <value>, no matter it is empty or not

Full example

\documentclass{article}
\usepackage{pgffor}

\begin{document}
% this prints "<>, <1>, <2>, <3>, <>,"
\foreach \x in {,1,2,3,} { <\x>, }

% currently, this raises an error "\lasti is undefined"
% if the attempt applied, \lastx is initialized to empty 
%   and this prints "<1>, <12>, <23>,", 
\foreach \x[remember=\x as \lastx (initially )] in {1,2,3}  { <\lastx\x>, }
\end{document}

[hmenke]

If you are posting a patch, please always paste the full diff. The one that you posted is corrupt because there are three more lines missing, so I couldn't directly apply it and had to fix it by hand.

In fact, you don't have to patches at all, you can just file a pull request here on GitHub or you can use git send-email to post your patches to the mailing list at [email protected]

[muzimuzhi]

Thanks again. I'll take care of it. Also, I'm not confident enough to directly open a pr.

[Mo-Gul]

@muzimuzhi, just give it a try. That will be a good chance to learn something.
(I regularly have messed up things and I am still here...)