[Tutor] How it is better than java
steve at pearwood.info
Tue Sep 20 02:27:12 CEST 2011
Ashish Gaonker wrote:
> My obvious thinking is : Java being compiled language , must be faster then
> a interpreted language.
There are three misunderstandings with that statement.
Languages are neither "compiled" or "interpreted". Languages are syntax
and grammar. Implementations are either compiled, or interpreted, or
both: for example, there are C interpreters and C compilers. And the
quality of both can vary significantly.
Asking which language is faster is like asking which is faster, Ford or
Toyota? That depends on the particular model, and the conditions, and
the skill of the driver.
It is ironic that you contrast Java as "compiled" and Python as
"interpreted", because that is *marketing*. When Java first came out,
Sun didn't want people describing it as "interpreted", which it was, so
they popularized the term "byte-code compiler" just so that they could
talk about Java being compiled. Java would compile your Java source code
to byte-code which was then interpreted by a virtual machine.
That is *exactly* what Python does: it compiles Python source code to
byte-code which is interpreted by a virtual machine, just like Java.
What do you think the .pyc files contain, and what the compile()
function does? And yet, even back in the 1980s, everybody swallowed
Sun's marketing and called Java a compiled language and Python an
interpreted language. This is a testament to Sun spending millions in
Byte-code compilation is a lot older than Java. Pascal used something
similar in the early 1970s, called a p-machine. Even Apple's Hypertalk
did the same thing, only they called it "tokenized" code instead of
compiled. Java's big innovation was to convince people to use the term
"compiler" for what was functionally identical to an interpreter.
Of course, Sun (now owned by Oracle) put in a lot of money into Java.
Millions. Today, Java does have implementations which compile source
code to machine code. But there are Python implementations that do the
same, such as Nuitka and Compyler. (I don't know how successful or good
Secondly, what do you mean by "faster"? Faster to write? Faster to
compile? Faster to run? Faster for the engine to start up? Even today,
after Sun has spent tens of millions on Java development, the Java
Runtime Environment is a big, heavyweight machine that takes a long time
to start up: cold starts can easily take 30 seconds. That makes Java
completely unsuitable for small, lightweight tasks: in the time it takes
for a *single* Java program just to start up, you could possibly run a
dozen Python programs or a hundred interpreted bash scripts.
But again, that's an *implementation*, not a hard rule about Java. There
is at least one third-party JRE which claims to have startup times twice
as fast as the Sun/Oracle JRE.
Either way, once you take startup time into account, sometimes Python
scripts are not only faster to write and faster to maintain, but faster
to run as well.
Thirdly, there is no rule of nature that a compiled program to do a job
must be faster than an interpreted program to do the same thing. This
depends entirely on the quality of implementation of both: a poor
compiler may easily generate bad, slow code that takes longer to run
than a wickedly fast and efficient interpreter. E.g. a compiled version
of bubblesort will still be slower than an interpreted version of quicksort.
Nothing drives this home more than PyPy, a Just In Time optimizing
version of Python. PyPy uses a JIT compiler to run code sometimes FASTER
than the equivalent program in optimized C.
Yes. Faster than C. You read that right.
Of course, benchmarks are notoriously flawed, especially micro-
benchmarks. What they *really* answer is not "which language is faster?"
(a nonsense question, as I have tried to explain) but "which
implementation is faster with these particular settings on this
particular task?", a much more practical question.
As exciting as it is to see Python code run faster than C code, it
shouldn't really surprise anyone that a JIT dynamic compiler with cross
module optimization beats a static compiler without it. What *is*
surprising is that a small group of open-source developers have been
able to build an optimizing JIT compiler for Python of such quality.
Or at least, it is surprising to people who think that quality code can
only come from big proprietary companies with huge budgets.
What's really exciting though is that now PyPy can be fast enough for
large programs that traditionally needed to be written in C can now be
(sometimes!) written in Python:
Who cares whether Java, or C, is "faster" than Python? The important
question should be, is Python fast enough for the job I have to do?
> Can you share some more especially as compared to Java / .net (two primarily
> used languages in enterprise language & web based applications)
You can look at Jython (Python for the JRE) and IronPython (Python for
.Net). There is also an older implementation, Python For .Net, which
runs the standard CPython implementation on .Net, but I don't think it
is still maintained -- IronPython has taken over its niche. There's also
JPype, which claims to give full access to Java libraries in Python.
More information about the Tutor