pdflatex with precompiling does not support .Tex file with Chinese filename

Question

First I am sure, pdflatex can compile Tex file with Chinese file name without problem. For example, the Tex source content is as follows

\documentclass[UTF8,hyperref]{ctexart}
\begin{document}
    中文
\end{document}

save it as a file named "中文.tex" and run

pdflatex 中文.tex

everything is OK.

Now, in order to speed up the compilation, we make pre-compilation according to this link. That is to edit a file named "pre.tex" whose content is as follows

\documentclass[UTF8,hyperref]{ctexart}

and run

pdflatex -ini -jobname="pre" "&pdflatex pre.tex\dump"

to generate a .fmt file called "pre.fmt"

now, we use this precompiled .fmt file in out tex source like

%& pre
\begin{document}
    中文
\end{document}

if we save it as a file named "en.tex", and pdflatex en.tex works fine. But if we save it as a file named "中文.tex", then pdflatex 中文.tex will show errors like

What is wrong here? why precompile doesn't support Tex file with Chinese file name?

I upload the file "中文.tex" download here

Answer 1

You can see the cause here

\edef\x{中文.tex}\show\x
\documentclass[UTF8,hyperref]{ctexart}
\edef\x{中文.tex}\show\x
\begin{document}
    中文
\end{document}

Compare the two \show

Initially the bytes are all essentially inert and so the filename gets passed to the filesystem and would work, but after the class has been loaded the bytes have been made active and given definitions that enable typesetting of the characters but this means they expand to the wrong thing for the filename.

---

I don't have all the fonts to fully test this but I think it works if you change your format template to

\documentclass[UTF8,hyperref]{ctexart}

% set high chars active at start of job
% After the commandline has been read, but before Tex starts reading the
% file contents, the tokens in \everyjob are run. 
% This loop sets every character between 127 and 255 to be catcode 13
% "active" which is the catcode they normally have after inputenc package
% or the ctexart class code.
\everyjob{{\count0=127
\loop
\global\catcode\count0=13
\ifnum\count0<255
\advance\count0 1
\repeat
}}

% set high chars inactive now for filename reading
% This loop sets every character between 127 and 255 to be catcode 12
% "other" which is the normal catcode for punctuation. this means that
% no special interpretation is done and the filename works as in a normal 
% format (where high characters are also not active until inputenc or
% similar is loaded. 
% Unlike the previous loop which is delayed in \everyjob
% This loop is executed here to make characters safe before the \dump
{\count0=127
\loop
\global\catcode\count0=12
\ifnum\count0<255
\advance\count0 1
\repeat
}

So that the 8bit characters are not active until \everyjob is run, after the filename has been read.