Editing Perl (section)

== Performance ==

[[The Computer Language Benchmarks Game]] compares the performance of implementations of typical programming problems in several programming languages.<ref>{{cite web |url=http://alioth.debian.org/projects/benchmarksgame/ |title=Alioth: The Computer Language Benchmarks Game: Project Info |access-date=2011-01-13 |publisher=[[Alioth]] |archive-url=https://web.archive.org/web/20130325192723/https://alioth.debian.org/projects/benchmarksgame/ |archive-date=March 25, 2013 |url-status=dead |df=mdy-all}}</ref> The submitted Perl implementations typically perform toward the high end of the memory-usage spectrum and give varied speed results. Perl's performance in the benchmarks game is typical for interpreted languages.<ref>{{cite web |url=http://benchmarksgame.alioth.debian.org/u32/which-programs-are-fastest.php?v8=on&lua=on&jruby=on&php=on&python3=on&yarv=on&perl=on |archive-url=https://web.archive.org/web/20130517145336/http://benchmarksgame.alioth.debian.org/u32/which-programs-are-fastest.php?v8=on&lua=on&jruby=on&php=on&python3=on&yarv=on&perl=on |url-status=dead |archive-date=2013-05-17 |title=Which programs are fastest? |access-date=2011-01-13 |work=Computer Language Benchmarks Game |publisher=[[Alioth]]}}</ref>

Large Perl programs start more slowly than similar programs in compiled languages because Perl has to compile the source every time it runs. In a talk at the [[Yet Another Perl Conference|YAPC::Europe 2005]] conference and subsequent article "A Timely Start", Jean-Louis Leroy found that his Perl programs took much longer to run than expected because the perl interpreter spent significant time finding modules within his over-large include path.<ref>{{cite web |url=http://www.perl.com/pub/a/2005/12/21/a_timely_start.html |title=A Timely Start |last1=Leroy |first1=Jean-Louis |date=2005-12-01 |publisher=O'Reilly |access-date=May 22, 2006 |archive-date=June 13, 2006 |archive-url=https://web.archive.org/web/20060613025623/http://www.perl.com/pub/a/2005/12/21/a_timely_start.html |url-status=live}}</ref> Unlike Java, Python, and Ruby, Perl has only experimental support for pre-compiling.<ref>{{cite web |url=https://metacpan.org/module/NWCLARK/perl-5.8.8/ext/B/B/Bytecode.pm#KNOWN-BUGS |title=B::Bytecode Perl compiler's bytecode backend |author1=Beattie, Malcolm |author2=Enache Adrian |name-list-style=amp |year=2003 |publisher=[[CPAN]] |access-date=September 16, 2013 |archive-date=March 30, 2021 |archive-url=https://web.archive.org/web/20210330011632/https://metacpan.org/pod/release/NWCLARK/perl-5.8.8/ext/B/B/Bytecode.pm#KNOWN-BUGS |url-status=dead}}</ref> Therefore, Perl programs pay this overhead penalty on every execution. The run phase of typical programs is long enough that [[amortized]] startup time is not substantial, but benchmarks that measure very short execution times are likely to be skewed due to this overhead.<ref>{{Cite book|last1=Schwartz|first1=Randal|url=https://books.google.com/books?id=va1PSgaO4xIC&q=Therefore,+Perl+programs+pay+this+overhead+penalty+on+every+execution.+The+run+phase+of+typical+programs+is+long+enough+that+amortized+startup+time+is+not+substantial,+but+benchmarks+that+measure+very+short+execution+times+are+likely+to+be+skewed+due+to+this+overhead.|title=Learning Perl|last2=foy|first2=brian|last3=Phoenix|first3=Tom|date=2011-06-23|publisher="O'Reilly Media, Inc."|isbn=978-1-4493-0358-7|language=en}}</ref>

A number of tools have been introduced to improve this situation. The first such tool was Apache's [[mod_perl]], which sought to address one of the most-common reasons that small Perl programs were invoked rapidly: [[Common Gateway Interface|CGI]] [[World Wide Web|Web]] development. [[ActiveState|ActivePerl]], via Microsoft [[ISAPI]], provides similar performance improvements.<ref>{{Cite book|last1=Stein|first1=Lincoln|url=https://books.google.com/books?id=qyzTI_eAeHUC&q=A+number+of+tools+have+been+introduced+to+improve+this+situation.+The+first+such+tool+was+Apache's+mod+perl,+which+sought+to+address+one+of+the+most-common+reasons+that+small+Perl+programs+were+invoked+rapidly:+CGI+Web+development.+ActivePerl,+via+Microsoft+ISAPI,+provides+similar+performance+improvements.|title=Writing Apache Modules with Perl and C: The Apache API and Mod_perl|last2=MacEachern|first2=Doug|date=1999|publisher="O'Reilly Media, Inc."|isbn=978-1-56592-567-0|language=en}}</ref>

Once Perl code is compiled, there is additional overhead during the execution phase that typically isn't present for programs written in compiled languages such as C or C++. Examples of such overhead include [[bytecode]] interpretation, reference-counting memory management, and dynamic type-checking.<ref>{{Cite book|last1=Bekman|first1=Stas|url=https://books.google.com/books?id=UDabAgAAQBAJ&q=Once+Perl+code+is+compiled,+there+is+additional+overhead+during+the+execution+phase+that+typically+isn't+present+for+programs+written+in+compiled+languages+such+as+C+or+C++.+Examples+of+such+overhead+include+bytecode+interpretation,+reference-counting+memory+management,+and+dynamic+type-checking.|title=Practical Mod_perl|last2=Cholet|first2=Eric|date=2003|publisher="O'Reilly Media, Inc."|isbn=978-0-596-00227-5|language=en}}</ref>

The most critical routines can be written in other languages (such as [[C (programming language)|C]]), which can be connected to Perl via simple Inline modules or the more complex, but flexible, [[XS (Perl)|XS]] mechanism.<ref>{{cite web |url=https://metacpan.org/module/Inline |title=Inline - metacpan.org |access-date=2011-01-26 |last=Ingerson |first=Brian |publisher=[[CPAN]] |url-status=dead |archive-url=https://web.archive.org/web/20130613163723/https://metacpan.org/module/Inline |archive-date=June 13, 2013}}</ref>