fmt/README.rst

105 lines
3.7 KiB
ReStructuredText

format
======
Small, safe and fast printf-like formatting library for C++
Benchmarks
----------
Compile time and code bloat
~~~~~~~~~~~~~~~~~~~~~~~~~~~
The script ``bloat_test.sh`` from the `tinyformat
<https://github.com/c42f/tinyformat>`__ repository tests compile time and
code bloat for nontrivial projects. It generates 100 translation units
and uses ``printf()`` or its alternative five times in each to simulate
a medium sized project. The resulting executable size and compile time
(g++-4.7.2, Ubuntu GNU/Linux 12.10, best of three) is shown in the following
tables.
**Non-optimized build**
====================== ================== ==========================
test name total compile time executable size (stripped)
====================== ================== ==========================
libc printf 2.8s 44K (32K)
std::ostream 12.9s 84K (60K)
format 16.0s 152K (128K)
tinyformat 20.6s 240K (200K)
boost::format 76.0s 888K (780K)
====================== ================== ==========================
**Optimized build (-O3)**
====================== ================== ==========================
test name total compile time executable size (stripped)
====================== ================== ==========================
libc printf 3.5s 40K (28K)
std::ostream 14.1s 88K (64K)
format 25.1s 552K (536K)
tinyformat 56.3s 200K (164K)
boost::format 169.4s 1.7M (1.6M)
====================== ================== ==========================
Printf and std::ostream win here which is not surprising considering
that they are included in the standard library. Tinyformat has somewhat
slower compilation times compared to format. Interestingly optimized
executable size is smaller with tinyformat then with format and for
non-optimized build its vice versa. Boost::format has by far the
largest overheads.
Speed tests
~~~~~~~~~~~
The following speed tests results were generated by building
``tinyformat_test.cpp`` on Ubuntu GNU/Linux 12.10 with
``g++-4.7.2 -O3 -DSPEED_TEST -DHAVE_FORMAT``, and taking the best of three
runs. In the test, the format string ``"%0.10f:%04d:%+g:%s:%p:%c:%%\n"`` or
equivalent is filled 2000000 times with output sent to ``/dev/null``; for
further details see the `source
<https://github.com/vitaut/tinyformat/blob/master/tinyformat_test.cpp>`__.
============== ========
test name run time
============== ========
libc printf 1.26s
std::ostream 2.02s
format 2.20s
tinyformat 2.51s
boost::format 10.40s
============== ========
As you can see boost::format is much slower than the alternative methods; this
is confirmed by `other tests <http://accu.org/index.php/journals/1539>`__.
Tinyformat is quite good coming close to iostreams. Unfortunately tinyformat
cannot be faster than the iostreams because it uses them internally.
Running the tests
~~~~~~~~~~~~~~~~~
To run the tests you first need to get the format repository with submodules::
$ git clone --recursive git://github.com/vitaut/format.git
Then go to the format directory and generate Makefiles with CMake::
$ cd format
$ cmake .
Next use the following commands to run the speed test::
$ make speed_test
or the bloat test::
$ make bloat_test
Acknowledgments
---------------
The benchmark section of this readme file and the performance tests are taken
from the excellent `tinyformat <https://github.com/c42f/tinyformat>`__ library
written by Chris Foster. boost::format is acknowledged transitively since
it had some influence on tinyformat.
TODO: SafeFormat, Clang DiagnosticHandler