2

I use a lot of unicode symbols, auto-inserted with a TeXStudio macro (e.g. "\alpha " --> "α"). This makes it easier for me to read my .tex file. These (~100) characters are declared in a .sty file called in the preamble, e.g.

\DeclareUnicodeCharacter{03B1}{\alpha}

Could this slow down compilation of the document?

Leucippus
  • 1,636
Patrick Ferguson
  • 153
  • 6
  • (in pdfLaTeX) Yes, it does slow down the compilation because defining: \DeclareUnicodeCharacter{03B1}{\alpha} makes the character α be a command that expands to \alpha. However the overhead of expanding α to \alpha is insignificant in a real document, so if it makes typing easier, there's no real reason for you to don't use that. – Phelype Oleinik May 29 '20 at 18:16
  • To be clear, you are saying the compilation time is (minimally) increased solely by the inclusion of \DeclareUnicodeCharacter{03B1}{\alpha} in the .sty file, and thus isn't increasing in the repeated use of α in the document? – Patrick Ferguson May 29 '20 at 18:46
  • 3
    No, both increase the compilation time. Roughly speaking, \DeclareUnicodeCharacter{03B1}{\alpha} is like \newcommand{α}{\alpha}, then each use of α eventually becomes \alpha. But really, that should make no visible difference in the time it takes to build the document. Just to give you an idea, a document with 7000 α takes 0.118 s to run on my machine, whereas the same document with 7000 \alpha takes 0.112 s. – Phelype Oleinik May 29 '20 at 18:56
  • 1
    @PhelypeOleinik how many cups of coffee can you make in that .006s though? – David Carlisle May 29 '20 at 19:08
  • 1
    @DavidCarlisle Not nearly enough :-) – Phelype Oleinik May 29 '20 at 19:32

2 Answers2

1

I made an experiment: the code

\documentclass{article}

\begin{document}

\ExplSyntaxOn
\prg_replicate:nn {100000} { $\alpha$$\alpha$$\alpha$$\alpha$$\alpha$$\alpha$$\alpha$$\alpha$$\alpha$$\alpha$\par }
\ExplSyntaxOff

\end{document}

run with time pdflatex -draftmode (that simply doesn't produce the PDF file) shows

real  0m1.491s
user  0m1.377s
sys   0m0.073s

If I change to

\documentclass{article}
\DeclareUnicodeCharacter{03B1}{\alpha}

\begin{document}

\ExplSyntaxOn
\prg_replicate:nn {100000} { $α$$α$$α$$α$$α$$α$$α$$α$$α$$α$\par }
\ExplSyntaxOff

\end{document}

the terminal shows

real  0m3.213s
user  0m3.070s
sys   0m0.090s

The run takes about twice as long. In a previous experiment, with just 100000 occurrences of alpha, the run took 20% longer.

I don't think there's anything to be worried about for a normal document.

If I repeat the same experiment with unicode-math and LuaLaTeX, the execution times are comparable (but much longer, of course).

egreg
  • 1,121,712
  • I'm not really sure I fully understand the benchmark results. 100,000 times $\alpha$ (total 100k) takes half the time as 100,000 times 10 times $α$ (total 1M), meaning that $α$ is about five times faster than $\alpha$. Yet, 100,000 times $α$ (the 'previous experiment') is slower than 100,000 times $\alpha$. So where does this difference come from? – Marijn May 30 '20 at 11:38
  • @Marijn It is not linear… – egreg May 30 '20 at 12:37
  • Is there a reason you used $α$$α$$α$$α$$α$$α$$α$$α$$α$$α$ rather than simply $α$? – Patrick Ferguson May 30 '20 at 13:33
  • @PatrickFerguson I did to produce just 100000 paragraphs, but with a total of a million instances. – egreg May 30 '20 at 13:45
  • 1
    @egreg Why didn't you do the same for $\alpha$? – Patrick Ferguson May 30 '20 at 13:49
  • @PatrickFerguson Oh, sorry, I didn't paste the correct version! – egreg May 30 '20 at 15:27
  • @Schrödinger'scat My language uses double negations… – egreg May 30 '20 at 20:10
0

If you’re using LuaLaTeX or XeLaTeX with the Unicode encoding, no. They use Unicode natively, and passing the code point for α over to the font renderer is actually faster than looking up the macro \alpha, detecting the current font, and converting that to the same codepoint.

In an 8-bit TeX engine, it’s the other way around: inputenc will set the Unicode character active and look up what command to insert into the document, so it would have been slightly faster to just enter the command.

However, the time of a human being reading your source is much more valuable than the microseconds that takes the computer.

Davislor
  • 44,045