PDF Generation, Bloat and Optimization

#pdf #cfml #coldfusion

The State of myCFML PDF Generation

My current comparison is with CFDocument on the deprecated CF2016 Developer Edition. I'm planning on performing tests with CF2021 & CF2023 soon and will perform this on my personal developer workstation using CommandBox and a testing framework that I'm in the process of developing. I have not personally downloaded or used CF2025 yet¹. I believe that it may be using the same iText library as CF10. (Can anyone confirm this?) I also haven't compared with Lucee 5.3+ Flying Saucer implementation.

This will be a part of a series as I intend on performing a lot more tests as well as sharing tips that we've learned on how to improve overall PDF quality, performance and file size.

Technologies

CFDocument Built-in Adobe ColdFusion function that creates PDF output from a text block containing CFML and HTML. (NOTE: Some ACF overhead may occur due to adding the "Adobe ColdFusion Developer/Trial Edition - Not for Production Use" watermark.)
CFHTMLtoPDF Built-in Adobe ColdFusion function that creates PDFs from HTML using a WebKit based rendering engine. Some ACF overhead may occur due to adding the "Adobe ColdFusion Developer/Trial Edition - Not for Production Use" watermark. I tried testing this on a CF2016 Standard production server, but got the error message "No Service manager is available.". (We don't use CFHTMLTOPDF, so it may be disabled at the CFAdmin level.)
WKHTMLTOPDF (LGPLv3; portable) tends to be faster and generate smaller PDFs. It can also run concurrently and generate PDFs in the background without using a ColdFusion thread or impacting the Java heap memory. Download
Ghostscript (GNU GPL Affero license; portable) is an interpreter for the PostScript(R) language and PDF files. It runs on various embedded operating systems and platforms including Windows, macOS, the wide variety of Unix and Unix-like platforms, and VMS systems. Download

Test an empty page with only a "Hello World" H1 header

There's a huge differences in the generation file size and the Ghostscript optimization doesn't make much of a difference.

Server	Engine	Generation	FileSize	Ghostscript Duration	Ghostscript Filesize	Percent Smaller
CF2016	CFDocument	486 ms	40,837	10 ms	40,068	1.89%
CF2016	CFHTMLTOPDF	1286 ms	77,179	9 ms	98,953	-28.20%
CF2016	WKHTMLTOPDF	223 ms	7,539	9 ms	5,167	31.47%

<!DOCTYPE html>
<html lang="en"><head>
<meta charset="utf-8">
</head>
<body>
<h1>Hello World</h1>
</body>
</html>

Tests with a locally-downloaded 4.7mb NASA JPG image rendered at 400px wide.

The difference regarding file size is a much wider gap here for CFDocument. Using Ghostscript to optimize results in a much smaller CFDocument.

Server	Engine	Generation	FileSize	Ghostscript Duration	Ghostscript Filesize	Percent Smaller
CF2016	CFDocument	8,046 ms	24,311,787	50 ms	74,478	99.693%
CF2016	CFHTMLTOPDF	2,256 ms	120,258	1017 ms	136,211	-11.71%
CF2016	WKHTMLTOPDF	1,638 ms	492,115	45 ms	21,012	95.73%

I also ran this test script on a production server for CFDocument to eliminate the watermark and the filesize was consistent 24mb, but the Ghostscript result was 47,028 (36.86% smaller if no watermark exists) and still twice as large as WKHTMLTOPDF.

<!DOCTYPE html>
<html lang="en"><head>
<meta charset="utf-8">
</head>
<body>
<h1>Hello World</h1>
    <img src="/tempdirectory/images_yyyymmddHHnnss.jpg" width="400" height="248" border="1">
</body>
</html>

¹ I haven't personally interacted with CF2025 beyond accessing it online via CFFiddle. I haven't had any time to read the terms and do not wish to be legally bound by any rules that I may not be fully understand. Back to text