I measured the size of some big modules, with B::Stats and B::C, statically compiled.
This amounts to about to the same size as dynamic usage via perl, with all options and modules used at once. I.e. the worst-case scenario.
Interestingly Module::Build is much harder to compile - you'd need 6GB RAM at least -
and much bigger in the end-result than Moose, which compiles/compresses really fine.
h1. Modules
h2. Module::Build
@perl -e'use Module::Build: print q(k)'@
| ops | 68K |
| files | 48 |
| src lines | 18K |
| compiled c lines | 1.5M |
| perl compiler memory | 360M |
| gcc compiler memory | **3.4G** |
| executable size | 37M |
h2. Moose
| ops | 91K |
| files | 109 |
| src lines | 28K |
| compiled c lines | 889K |
| perl compiler memory | 218M |
| gcc compiler memory | 1.6G |
| executable size | 23M |
h2. DateTime::TimeZone
| ops | 2.3K |
| files | 6 |
| src lines | 2.2K |
| compiled c lines | 130K |
| perl compiler memory | 19M |
| gcc compiler memory | 415M |
| executable size | 32M |
h1. Typical Apps
h2. Biggest cPanel app
_The recipe for less than 2MB RAM and gcc < 4.5 is:_
cc_harness -E -o$@-cpp.c $@.c
cc_harness -O0 -fno-var-tracking -fno-stack-protector \
-fno-tree-loop-optimize \
-fpreprocessed -fno-exceptions $@-cpp.c
| ops | **467K** |
| files | 539 |
| src lines | 95K |
| compiled c lines | 1.5M |
| perl compiler memory | 352M |
| gcc compiler memory (32bit -O0) | 2.0G |
| executable size | **61M** |
h2. Typical cPanel app
| ops | 120K |
| files | 156 |
| src lines | 28K |
| compiled c lines | 1M |
| perl compiler memory | 120M |
| gcc compiler memory (32bit -O0) | 734M |
| executable size | 42M |
h2. dzil --help
| ops | 167K |
| files | 214 |
| src lines | 48K |
| compiled c lines | 18M |
| perl compiler memory | 485M |
| gcc compiler memory | **4.5G** |
| executable size | 48M |
| B::C time | 182s |
| cc time | s |
h1. Minimal Web apps
_A simple two-state form "like this":https://gist.github.com/387207 without Android._
h2. Plack
| ops | 78K |
| files | 98 |
| src lines | 29K |
| compiled c lines | 388K |
| perl compiler memory | 95M |
| gcc compiler memory | 1.0G |
| executable size | 11M |
| B::C time | 11s |
| cc time | 72s |
h2. Dancer
| ops | 123K |
| files | 143 |
| src lines | 40K |
| compiled c lines | 517K |
| perl compiler memory | 159M |
| gcc compiler memory | 1.1G |
| executable size | 14M |
| B::C time | 20s |
| cc time | 97s |
h2. Mojo (daemon)
| ops | 151K |
| files | 150 |
| src lines | 46K |
| compiled c lines | 622K |
| perl compiler memory | 153M |
| gcc compiler memory | 1.4G |
| executable size | 17M |
| B::C time | 29s |
| cc time | 114s |
h2. Catalyst
_catalyst.pl Catalyst::Test; script/catalyst_test_create.pl. Same numbers with the server_
| ops | 199K |
| files | 214 |
| src lines | 57K |
| compiled c lines | 941K |
| perl compiler memory | 267M |
| gcc compiler memory | 1.7G |
| executable size | 26M |
| B::C time | 44s |
| cc time | 169s |
h1. Minimal OO frameworks
h2. Moose
| ops | 91K |
| files | 109 |
| src lines | 28K |
| compiled c lines | 889K |
| perl compiler memory | 218M |
| gcc compiler memory | 1.6G |
| executable size | 23M |
h2. Mouse
| ops | 20K |
| files | 23 |
| src lines | 6.5K |
| compiled c lines | 120K |
| perl compiler memory | 19M |
| gcc compiler memory | 427M |
| executable size | 3.3M |
h2. Moo
| ops | 6.4K |
| files | 16 |
| src lines | 3.7K |
| compiled c lines | 96K |
| perl compiler memory | 17M |
| gcc compiler memory | _error_ (326M) |
| executable size | _error_ (2.7M) |
h2. Mo
| ops | 3.7K |
| files | 8 |
| src lines | 2.6K |
| compiled c lines | 67K |
| perl compiler memory | - _(too tiny, not measurable)_ |
| gcc compiler memory | 346M |
| executable size | 1.8M |
h1. Summary
If you compile an app like Module::Build, i.e dzil (which means in the convoluted java-like style like Module::Build) you need a really big build machine, and developers will not be able to compile it by themselves, because they have not enough RAM. Errors will only appear daily (from the build machine), not immediately.
Moose alone is okay to use, dzil for sure not.
A minimal web app with Plack, Dancer, Mojo and even Catalyst look all okay.
SW metrics to explain the style, why the numbers are so different in the next blog post.
So if you plan to do "modern perl" you'll either take care, like using Mouse or Moo, or do something like prefork or FCGI and do not care.
Measured numbers are for x86_64 perl5.14.2-nt, gcc -Os -msse4.2 -march=corei7
| | | gcc | c | exe |
| | subs | mem[M] | lines | size |
| Dist::Zilla | 4309 | 4500 | 1320K | 46M |
| Module::Build | 4017 | 3400 | 1500K | 37M |
| CPAN | 4142 | 3300 | 1319K | 33M |
| Catalyst | 4508 | 1700 | 941K | 26M |
| ExtUtils::CBuilder | 1633 | 1552 | 595K | 15M |
| Moose | 2523 | 1600 | 889K | 23M |
| Mojo | 2559 | 1400 | 622K | 17M |
| Dancer | 2368 | 1100 | 517K | 14M |
| Plack | 1945 | 1000 | 388K | 11M |
| ExtUtils::MakeMaker | 1233 | 957 | 335K | 8.4M |
| DateTime | 1277 | 840 | 293K | 6.8M |
| DateTime::TimeZone | 542 | 415 | 130K | 3.1M |
| DateTime::Locale | 471 | 464 | 138K | 3.4M |
| Mouse | 743 | 427 | 120K | 3.3M |
| Moo | 572 | 326 | 96K | 2.7M |
| Mo | 388 | 346 | 67K | 1.8M |
h1. Makefile tricks
If you want to check in your Makefile how much memory is available and
disable some gcc optimizations accordingly you can something like this:
Makefile snippet
bc. AVAILMEM=$(shell availmem.sh)
bigapp: $@.pl
if [ ${AVAILMEM} -lt 2000000 ]; then \
echo "less than 2GB free RAM, disable some gcc optimizations"; \
/bin/rm -f $@ $@-cpp.c $@.c; \
perl -c $@.pl; \
perl -MO=C,-O3,-o$@.c $@.pl; \
cc_harness -E -o$@-cpp.c $@.c; \
cc_harness -O0 -fno-var-tracking -fno-stack-protector \
-fno-tree-loop-optimize -fpreprocessed \
-fno-exceptions $@-cpp.c -o$@; \
/bin/rm -f $@-cpp.c $@.c; \
else \
perlcc $(PERLCC_OPTS) $@.pl -o $@; \
fi
My availmem.sh on linux:
bc. #!/bin/sh
memtotal=$(perl -ane'print $F[1] if /^MemTotal:/' /proc/meminfo)
memused=$(perl -ane'print $F[1] if /^Active:/' /proc/meminfo)
echo $(($memtotal - $memused))
Plack
p -MB::Stats t/plack.pl
pb script/perlcc -v2 --time -O3 t/plack.pl -S -ot/plack
files=98 lines=28891 ops=78214
perl res 95m
cc1 res 1.0g
c time: 11.407679
cc time: 71.652737
Dancer
p -MB::Stats t/dancer.pl
pb script/perlcc -v2 --time -O3 -S t/dancer.pl
files=143 lines=40177 ops=123197
perl res 159m
cc1 res 1.1g
c time: 20.45
cc time: 96.88
Mojo
(daemon)
files=150 lines=46779 ops=151577
perl res 153m
cc1 res 1.4g
c 29s
cc 114s
Catalyst
script/catalyst_test_create.pl (no server)
files=214 lines=57770 ops=199103
perl res 267m
cc1 res 1.7g
c time: 44s
cc time: 169s
CPAN
c time: 78.491689
cc time: 255.88538
mouse
pb script/perlcc -v2 --time -O3 -S -ot/mouse -e'use Mouse; print q(ok)'
19m
427m
c time: 2.758865
cc time: 23.37078
number of subs:
for m in Module::Build Moose Mouse Moo Mo DateTime::TimeZone Dist::Zilla Plack::Runner;use Plack::Request Dancer Mojo Catalyst
do
echo -n "$m ";
perlcc -v4 --Wb=-Ds,-c -e'use Moose' 2>&1 |egrep -iv "^ (New |Warning)"|sort -u|wc -l
echo "";
done
gcc c exe
subs mem[M] lines size
Dist::Zilla 4309 4500 18000K 48M
Module::Build 4017 3400 1500K 37M
CPAN 4142 3300 1319K 33M
Catalyst 4508 1700 941K 26M
ExtUtils::CBuilder 1633 1552 595K 15M
Moose 2523 1600 889K 23M
Mojo 2559 1400 622K 17M
Dancer 2368 1100 517K 14M
Plack 1945 1000 388K 11M
ExtUtils::MakeMaker 1233 957 335K 8.4M
DateTime 1277 840 293K 6.8M
DateTime::TimeZone 542 415 130K 3.1M
DateTime::Locale 471 464 138K 3.4M
Mouse 743 427 120K 3.3M
Moo 572 326 96K 2.7M
Mo 388 346 67K 1.8M