\fontsize{0}{10pt} のエラー？

[aminophen]

不思議なので issue にしておきます。

https://twitter.com/aminophen/status/1240941195620040705

\documentclass{article}
\begin{document}
\fontsize{0}{10pt}
\selectfont
\end{document}

→ latex, pdflatex: エラーなし

→ platex, uplatex, lualatex, xelatex: エラー発生

! Improper `at' size (0.0pt), replaced by 10pt.

[kmaed]

pdflatex でも lmodern など何かフォントを読もうとすると同じエラーになります．

external@font ->rm-lmr5 at0.0pt

! Improper `at' size (0.0pt), replaced by 10pt.

[aminophen]

pdflatex でも lmodern など何かフォントを読もうとすると同じエラー

本当ですね。latexbug-compatible な fix-cm でも出せました：

\RequirePackage{latexbug}
\documentclass{article}
\usepackage{fix-cm}
\begin{document}
\tracingonline1
\tracingmacros1
\fontsize{0}{10pt}
\selectfont
\end{document}

[aminophen]

少し違う例ですが

\documentclass{article}
\usepackage{minijs}
\begin{document}
\fontsize{-1}{10pt}
\selectfont
\end{document}

だと

! LaTeX Error: Font JT1/mc/m/n/-1 not found.
! LaTeX Error: Font JY1/mc/m/n/-1 not found.

になり，ここで \usepackage{minijs} を消すと

LaTeX Font Warning: Font shape `JT1/mc/m/n' in size <-1> not available
(Font)              size <5> substituted on input line 5.
LaTeX Font Warning: Font shape `JY1/mc/m/n' in size <-1> not available
(Font)              size <5> substituted on input line 5.

になります。fix-cm の有無でも同様の違いが起きます。

[aminophen]

素の latex で起こせる MWE にするとこんな感じですね。

\documentclass{article}
\DeclareFontShape{OT1}{cmr}{m}{n}{<-10> cmr8}{}
\begin{document}
\nonstopmode
\fontsize{0}{15pt}\selectfont
% => ! Improper `at' size (0.0pt), replaced by 10pt.
\fontsize{-1}{13pt}\selectfont
% => ! LaTeX Error: Font OT1/cmr/m/n/-1 not found. をひたすら吐き続ける
\end{document}

[aminophen]

こうすれば回避できそう：

\makeatletter
\def\check@range#1-#2>#3<#4\@nnil{%
  \ifnot@nil{#3}{%
     \def\reserved@f{\extract@rangefontinfo<#4\@nnil}%
     \upper@bound0#2\p@
     \ifdim\upper@bound=\z@ \upper@bound\maxdimen\fi
     %%% addition BEGIN
     \ifdim \f@size \p@<1sp
       \def\f@size{1}%
     \fi
     %%% addition END
     \ifdim \f@size \p@<\upper@bound
       \lower@bound0#1\p@
       \ifdim \f@size \p@<\lower@bound
        \else
          \set@simple@size@args#3<#4\@nnil
          \execute@size@function\sizefn@info
         \ifx\external@font\@empty
         \else
           \let\reserved@f\@empty
         \fi
       \fi
     \fi
     \reserved@f}}
\makeatother

[aminophen]

Reported: https://github.com/latex3/latex2e/issues/309

[h-kitagawa]

こうすれば回避できそう：

これだと，

\DeclareFontShape{OT1}{cmr}{m}{n}{<-10> cmr8}{}
...
\fontsize{0}{15pt}\selectfont % (1)
\fontsize{-1}{13pt}\selectfont % (2)

の (1), (2) で cmr8 at 1.0pt を選択してしまうのが不自然な気がします．

[zr-tex8r]

えーと、「何が正しい動作なのか」がよく理解できていないのですが…。

• TeXの\fontにおいてフォントサイズを0ptや−1ptにしようとすると、どのエンジンでもエラーになる。
• なので、LaTeXの\fontsizeにおいてフォントサイズを0ptや−1ptにしようとした場合は、エンジンの種類や当該シェープの\DeclareFontShapeの定義の如何に関わらず、必ずエラーになるべきである。

ということで合っていますか？

[FrankMittelbach]

I'm not sure I understand the issue. Asking for a font at size 0 is wrong, so I find the error showing up quite reasonable. The fact this doesn't happen if your declaration only have a fixed set of sizes is also reasonable im my opinion. If you ask for \fontsize{foo}{10} you even get a worse low-level error

./gh309.tex:7: Missing number, treated as zero.
<to be read again> 
                   f

but I think even that is acceptable, because it is really not LaTeX's task to guard against all kind of invalid input in order to keep things reasonably fast.

Or am I missing something?

[aminophen]

cmr8 at 1.0pt を選択してしまうのが不自然

確かに \fontsize{0.5}{15pt} だと 0.5pt になるのに \fontsize{0}{15pt} だとより大きな 1.0pt になるので変な気はします。幾つにフォールバックすれば良かったのでしょう？

「何が正しい動作なのか」がよく理解できていないのですが…。

• どのエンジンでも \font プリミティブの動作は「0pt や −1pt にしようとするとエラー」です。
• 現状の NFSS は「"0pt" はそのまま \font に渡す，"−1pt" はその手前でよく分からない LaTeX エラー "Font OT1/cmr/m/n/-1 not found." を出す」という挙動です。
• 私の意見は「LaTeX の \fontsize は \font プリミティブの動作を知っているべき。NFSS が 0pt や −1pt をそのまま \font に渡すべきではない。」です。

「もっとわかりやすいエラーを出す」か「エラーなく可能な最小サイズにフォールバックする」のどちらかになれば，私としてはOKです。

[aminophen]

@FrankMittelbach Your test file "tlb0747.lvt" contains

\newcommand\tst[1]{\fontsize{#1}{13pt}\selectfont \showthe\font}
\tst{-1}
\tst{0}

so I guess those are considered a valid input at least for standard LaTeX. OTOH,

\newcommand\tst[1]{\fontsize{#1}{13pt}\selectfont \showthe\font}
\tracingonline1
\tracingmacros1
\tst{-1}
\tst{0}
\stop

runs perfectly on "latex" but throws errors on "platex":

external@font ->tmin10 at0.0pt
! Improper `at' size (0.0pt), replaced by 10pt.

external@font ->min10 at0.0pt
! Improper `at' size (0.0pt), replaced by 10pt.

(Here tmin10 and min10 are names of Japanese-specific fonts.)

If they are considered a valid input for standard LaTeX, I hope this should be valid also on pLaTeX, even when number of loaded fonts are different.

[FrankMittelbach]

Am 20.03.20 um 15:23 schrieb Hironobu Yamashita:

確かに |\fontsize{0.5}{15pt}| だと 0.5pt になるのに |\fontsize{0}{15pt}| だとより大きな 1.0pt になるので変な気はします。幾つにフォールバックすれ ば良かったのでしょう？

\fontsize{0.5}{15pt} asks for a font at size 0.5pt (with a baselineskip of 15pt)

\fontsize{0}{15pt} asks for a font at size 0pt (agan with a baselineskip of 15pt)

so what do you mean by "it is larger"?

the baselineskip has nothing to do with the font size, the NFSS interface has only bundled the two because you more or less always want to set/alter the baseline skip when you set the font size.

so bottom line asking for a font which is 0 is just plain wrong. I'm really a bit confused here.

[aminophen]

so what do you mean by "it is larger"?

"it is larger" is referring to the result of my proposal of "fallback" \def\f@size{1}; so I mean \def\f@size{1} is not good.

[zr-tex8r]

• 警告を出して正の値にフォールバックする」というのは結局「文書作成者の期待と大きく異なる結果にしかならない。
  • 負の値になるのは明らかにミスである。
  • 何か意図があってフォントサイズをゼロにしようとしたとしても、それは実現されない。
• 大昔から（どのエンジンでも）エラーになる場合があった。

なのだから、もし改善するのであれば

• 早い段階で、明瞭なエラーを出す。
• リカバリで値を（\fontに合わせて）10ptにする（その後の不審な挙動を抑止するため）。

が妥当だと思いますね。

---

• “Fallback to some positive value (possibly with warning)” would result in the output that is totally different from what the author intends.
  • Negative values are simply wrong.
  • Even when the author chooses zero for some intention, that will be never realized.
• There have been cases for decades that such values issues errors (on any engines).

Then I think a reasonable fix will be:

• Issue a clear and informative error,
• Then change the size to 10pt (as \font does) as recovery so as to avoid haphazard afterwards.

[FrankMittelbach]

Hironobu

@FrankMittelbach https://github.com/FrankMittelbach Your test file "tlb0747.lvt" contains

\newcommand\tst[1]{\fontsize{#1}{13pt}\selectfont \showthe\font} \tst{-1} \tst{0}

so I /guess/ those are considered a valid input at least for standard LaTeX. OTOH,

no you can't conclude any thing like that :-)

I have the habit of testing a lot of strange and wrong input to see how code behaves, so that fact that I'm testing for a font with a negative size doesn't mean it is acceptable input only that in 1994 (where nearly all fonts where metafont based and and only available in discrete sizes) I checked what the code back then did.

\newcommand\tst[1]{\fontsize{#1}{13pt}\selectfont \showthe\font} \tracingonline1 \tracingmacros1 \tst{-1} \tst{0} \stop

runs perfectly on "latex" but throws errors on "platex":

it only runs perfectly with discrete fonts it never did with Type1 fonts when they came about and it is rather pointless to epect that a fotn with 0 or negative size should do anything sensible.

If they are considered a valid input for standard LaTeX, I hope this should be valid also on pLaTeX, even when number of loaded fonts are different.

as I said it is not valid input and it only gets "sort of corrected" for discrete fonts and not for scalable ones.

[aminophen]

as I said it is not valid input and it only gets "sort of corrected" for discrete fonts and not for scalable ones.

Thanks for clarification; then I should mark the test file "tlb0747.lvt" (= runs perfectly on "latex" but throws errors on "platex") as "known difference" between latex and platex.

[h-kitagawa]

範囲指定に対する fallback が奇妙？

\documentclass{article}
\DeclareFontShape{OT1}{cmr}{m}{ta}{<10> cmr17}{}
\DeclareFontShape{OT1}{cmr}{m}{tb}{<-3> cmr17}{}
\DeclareFontShape{OT1}{cmr}{m}{tc}{<6-> cmr17}{}
\begin{document}
\nonstopmode
\def\TEST#1{%
  \fontshapeforce{ta}\fontsize{#1}{15pt}\selectfont% (A)
  \fontshapeforce{tb}\fontsize{#1}{15pt}\selectfont% (B)
  \fontshapeforce{tc}\fontsize{#1}{15pt}\selectfont% (C)
}
\TEST{5}\TEST{0}\TEST{-1}
\end{document}

Results:

argument of \TEST	5	0	-1
(A)	cmr17 at 10.0pt	cmr17 at 10.0pt	cmr17 at 10.0pt
(B)	! LaTeX Error: Font OT1/cmr/m/tb/5 not found.	! Improper `at' size | ! LaTeX Error: Font OT1/cmr/m/tb/-1 not found.
(C)	! LaTeX Error: Font OT1/cmr/m/tc/5 not found.	! LaTeX Error: Font OT1/cmr/m/tc/0 not found.	! LaTeX Error: Font OT1/cmr/m/tc/-1 not found.

例えば「使えるサイズ指定の中で最良近似を使う」という方針にするなら，次のようになるのが望ましい？

argument of \TEST	5	0	-1
(A)	cmr17 at 10.0pt	cmr17 at 10.0pt	cmr17 at 10.0pt
(B)	cmr17 at 3pt	*	*
(C)	cmr17 at 6pt	cmr17 at 6pt	cmr17 at 6pt

*: invalid font size とかなんとか LaTeX レベルのエラー

[FrankMittelbach]

If you run that test with the .fd files for cm fonts I can't imagine that it fails in platex why should it? It fails because you also arrange for Japanese fonts to go alongside with it, isn't that the reason?

The same would happen with luatex (that uses LM fonts by default --- it works only because we run with a special fontsetup to avoid that most of the time, otherwise anytest showing \showoutput would be different across engines). But in other cases we explicitly have tlg files for individual engines generated like this:

l3build save -eluatex tlb0747

I guess you could do the same when needed

[aminophen]

It fails because you also arrange for Japanese fonts to go alongside with it, isn't that the reason?

Right, and of course I knew that. I was comparing the real-life pLaTeX format (including .fd files for Japanese fonts) against the real-life LaTeX format, to ensure that pLaTeX's extensions to NFSS does not break any valid inputs of LaTeX.

I guess you could do the same when needed

Yes I'm considering that.

[aminophen]

I was comparing the real-life pLaTeX format (including .fd files for Japanese fonts) against the real-life LaTeX format, to ensure that pLaTeX's extensions to NFSS does not break any valid inputs of LaTeX.

I raised this issue simply because I felt it strange that the source file which passes LaTeX fails on pLaTeX. Ideally pLaTeX should be upward compatible with LaTeX...

[FrankMittelbach]

well, I would say it is upward compatible. I don't think that you really need to concern yourself mimiking how LaTex handles bad input. I mean

\fontsize{0.1}{10}\selectfont

runs through latex and platex in both cases selecting 5pt CM not 0.1pt CM (so already that is an error situation (somehow handled but not well). Selecting a nearby font was really meant for cases where the difference is small, eg rounding errors or maybe a point but not 4 or 5

in ptex your j-fonts will actually be selected at 0.1pt so you get a rather strange mix :-) but again the ask for 0.1 was already not a reasonable one if CM is the font so ... tough

if you switch to LM or any other scalable font then both implementations work the same, so I think you are fine.