In which order should I load inputenc and fontenc?

Question

When I’m using pdfLaTeX, my preamble usually contains

\usepackage[utf8]{inputenc}

and

\usepackage[T1]{fontenc}

Should these two be loaded in any particular order? What problems could arise if they’re loaded in the “wrong” order?

On a related topic, I know that fontenc should come after font packages, as described in Should the fontenc package be loaded before or after font packages?, so I’d move font packages along with inputenc – unless there’s a catch as well, which would indicate something like
1) font package(s) 2) inputenc 3) fontenc?

Answer 1

No the order doesn't matter as inputenc contains some code to avoid "order problems".

Some background: inputenc with option utf8 doesn't set up definitions for every unicode character as everyone knows who got once the message ! Package inputenc Error: Unicode char \u8:... not set up for use with LaTeX.

Instead it loads a .dfu for each font encoding declared by fontenc (naturally only if the dfu exists) so that there exist input definitions for the chars that can be printed with this font encodings.

fontenc puts all the font encodings it has been called with or which have been loaded by default in a list \cdp@list. inputenc process this list and loads the corresponding .dfu (e.g. T1.dfu) and then redefines \DeclareFontEncoding@ so that from now on every \DeclareFontEncoding will load the .dfu directly.

Answer 2

Before trying to answer the question I would like to make a general statement that font encoding and keyboard mappings is an area which is still in a mess. This is not only with LaTeX but also with the web. How many pages still come up with funny characters? The nearest thing to a solution is the use of UTF-8. For the TeX/LaTeX family the answer is XeLaTeX or LuaLaTeX.

For texts that their language is predominantly English, I would stick with the extended ASCII encoding at the cost of typing a few commands for accented letters here and there. This in my opinion is a safer way and ensures that journal archives have an easier time in the longer run.

As to the question, I would place \inputenc first as shown in the below minimal.

\documentclass{article}
\usepackage[utf8]{inputenc} 
\def\testa{Umlauts: äöüß}   
%\usepackage[T1]{fontenc}    
%\usepackage{textcomp}
%\usepackage{lmodern}
%\usepackage[utf8]{inputenc} 
\begin{document}

\testa

äöüß 

\end{document}

Reasoning, defensive programming means I want to catch errors as soon as possible. Contrary to what one would expect, this works with and without \fontenc. (Since LaTeX loads both the T1 and OT1 encoding files on its own).