Am Tue, 7 Apr 2015 23:22:16 +0200 schrieb Frank Mittelbach:

> Am 07.04.2015 um 16:01 schrieb Ulrike Fischer:
>> I run into a problem with T5-encoding and a font family which
>> doesn't support this encoding.
>>
>> I did sent a question on tex.SX:
>>
>> http://tex.stackexchange.com/questions/237188/corrupted-nfss-tables
> 
> I answered it there

Yes, I saw it. Thanks.

>>
>> In my opinion the fd-file for the "fallback" font is loaded to late
>> in \[log in to unmask]
> 
> 
> the problem is really that preamble is still setup time so "fallback 
> font may not be the final one etc etc

Yes, I thought that this is the reasoning why \process@table is
called a \begin{document}. 

 
>> I found a workaround but I'm in no way sure
>> that it hasn't some unwanted side-effects.
> 
> I don't think it would have but it is executed unnecessarily often as 
> attempting to load anything only makes sense if the family got changed.
 
> Not sure if this should get changed

In my opinion it should. The core ingredients are an "exotic"
fontencoding (T5, LGR, ....)(e.g. through a language with such an
encoding) and a font family which has no own fd-files for this
encoding. Then you can be bitten quite easily with the bug as the
aux-file with the \selectlanguage is read *before* \process@table
(compile twice):

\documentclass{article}
\usepackage{newpxtext}
\usepackage[vietnam]{babel}
\begin{document}
blub
\end{document}

And if one google a bit for "corrupted NFSS" one can find quite a
number of questions with "cyrillic", "vietnames", "greek".  


> without a lot more analysis as this 
> code has basically stayed untouched for a really really long time but 
> after spending a good deal of time tonight on it I can't see anything 
> wrong with it -- currently running the LaTeX regression test suite 
> against it but I'm not sure how much of exotic NFSS cases this really tests

I don't see what harm it can do to load an fd-file apart from a tiny
loss of speed and memory. Probably it would be okay too to load it
directly in the \DeclareFontSubstituation command in t5enc.def. 

-- 
Ulrike Fischer 
http://www.troubleshooting-tex.de/