• Line Wrapping Benchmark
• Benchmark Setup
• Results

Line Wrapping Benchmark

This benchmark tests the performance of line wrapping and simple general layouting. The Project Gutenberg text of Homer’s Odyssey is used for this purposes.

Benchmark Setup

The text of the Odyssey is arranged on pages of the dimension WIDTHx1000 where WIDTH is set to different values (400, 200, 100 and 50 by default). Additionally, all widths are combined once with the standard PDF Type1 font Times-Roman and once with a TrueType font (DejaVu Sans by default).

In the case of pages with a width of 400 no line wrapping needs to be done because each line in the source text is shorter than 400 points. In the other cases lines need to be actually wrapped and the number of pages increases. With a width of 50 even words need sometimes to be broken.

Each benchmark script can be invoked standalone in the following way: script-executable TXT_FILE WIDTH OUTPUT_FILE [TTF_FILE].

The performance of the libraries hugely depends on how the input text is provided: Some are very fast when processing the whole input file at once, others only when processing the input line by line. The fastest method was always chosen.

The list of the benchmarked libraries:

HexaPDF

Homepage: http://hexapdf.gettalong.org
Language: Ruby
Version: Latest version

HexaPDF works faster if the whole input is given at once but still has acceptable runtimes for line by line input.

Two different ways of general layouting are benchmarked:

L

This version uses the low-level layouting facility HexaPDF::Layout::TextLayouter and processes the whole file at once.

C

This version uses the high-level HexaPDF::Composer to construct the document and processes chunks of 1000 lines at once to reduce peak memory usage.

Prawn

Homepage: https://prawnpdf.org
Language: Ruby
Version: 2.5.0

Prawn is much faster and uses much less memory if the input is provided line by line. However, it still works if the whole input is provided at once.

ReportLab

Homepage: https://www.reportlab.com/opensource/
Language: Python
Version: 4.2.2 + accel extension

ReportLab also needs its input line by line. Otherwise it is much, much slower (at least 60x, then the test run was aborted).

fpdf2

Homepage: https://pyfpdf.github.io/fpdf2/
Language: Python
Version: 2.7.9

As with HexaPDF itself, fpdf2 works equally fine when getting the input as one big string or line by line.

TCPDF

Homepage: https://tcpdf.org/
Language: PHP
Version: 6.7.5

As with Prawn and ReportLab, TCPDF needs its input line by line. Otherwise it is much, much slower when line wrapping needs to be done (the test run was aborted because it took too long).

Results

These benchmark results are from 2025-01-04.