freebsd-ports/devel/gecc/files

The group enabled cluster compiler (gecc) is a tool to make build faster.  This
tool is inspired by distcc (<URL:http://distcc.samba.org>) and ccache
(<URL:httpd://ccache.samba.org>).  It helps in two ways:

- not doing unneeded compiles and
- distributing builds on a cluster of hosts.

These two optimizations are unrelated and both of them are optional to gecc.
Not doing unneeded compiles means that gecc caches the build of object files
and distributing means putting compiles on more than one host (not a big
surprise).

For a more detailed look, please refer to the Web pages at
<URL:http://gecc.sf.net>, which are included in the CVS repository.  Take a
look at the htdocs directory.

building:

I have compiled gecc with gcc 2.95.3.  Then I got a report that gecc does not
compile with gcc-3.2 (thanks for that), so I fixed gecc to compile with both.
There are problems using the std::string::compare member functions. They have
changed from gcc 2.95.x to gcc 3.y (I have been told that gcc3 is correct, but
I have not checked with the ANSI standard).  My normal CXX is gcc 2.95.3.  If I
break compatibilty with gcc-3.x again, please drop me a note and I will fix it,
if anybody cares.

The usual commands:

	./configure
	make

should build two binaries:  gecc and geccd.  The client is gecc.  It's called
instead of your usual C/C++ compiler, which means gcc in almost every case.
There are two ways to call it:

1.) gecc gcc file.c -c -o file.o -O2 ...
That is, prepend your usual command line with "gecc".  gecc will find out which
C/C++ compiler you would have called. It crawls your PATH for the next token on
the command line ("gcc" in our example).  It will take a look, which version it
is, by calling it with "--version" as the only parameter.  The name and the
version should identify the compiler.  There are plans to take also care of the
architecture of your host, so cross compiles could be done (this is one of my
needs, but not implemented now).

2.) If you make a link from the name of your compiler to gecc, then gecc will
behave as above. The advantage is that this works with libtool, which is a bit
picky about chained tools. I have a ~/bin directory, which is the very first in
my $PATH. There are links like this:

	c++ -> gecc
	g++ -> gecc
	gcc -> gecc
	cc -> gecc

Also, gecc is in the path.

Either way, the source file is preprocessed and a hash of the preprocessed file
and your command line is calculated.  This hash is looked up in a
on-disk-cache.  For a cache hit, the object file, the stderr, and the compiler
return code of the original compiler run are taken from cache and the original
result is "reproduced".  On a cache miss the compiler is called and the result
is recorded for the cache.

If there are nodes registered to help in the compiling than for every cache
miss, the host to do the compilation is calculated by a scheduler algorithm
(right now it's round robin, but there will be feedback on the basis of
compilation speed).

testing:
To test gecc, you need to start a geccd (the gecc daemon).  For example:

	geccd -C /tmp/gecc-cache --compile -d

will start geccd in debugging mode (-d), that is, without forking. A Ctrl-C
will terminate the program. Then, on another shell, you need to set some
environment variables:

either:
	export GECCD_SOCKFILE=/tmp/geccd.sockfile
or:
	export GECCD_HOSTNAME=localhost
	export GECCD_PORT=42042

Now do this:

	gecc gcc hello.c -o hello.o

Assuming you have a hello.c, it will be compiled to hello.o.  You can see geccd
act and print debugging stuff on stderr. If you now remove the hello.o and
repeat the same command line, hello.o will be taken from cache.

If you have a second machine available, then you could start a second geccd on
this second machine. Let's assume the first machine is named dilbert and the
second is named asok. Then the command line on asok will be:

	geccd --compile -a dilbert -A 42042 --compile --port 42042 -d

this means announce yourself to a geccd, running on dilbert (port -A 42042) and
help him to compile. Right now it is important to start the geccd on the helper
host _after_ the main geccd (yes, this has to change).  If you now do
compilation of more source files, then they will be distributed on all
machines.

The compile nodes don't need the includes or libs installed, but the same
version of the compiler.  If not all compile nodes have all needed compilers
installed, that's OK, since they only get jobs for the compilers they announce
to the main geccd.  For this, they scan the $PATH and collect all known compiler
binaries, which right now is all binaries that gecc installs.

If you try this out, please be so nice as to mail me your feedback (either
positive or negative, any feedback is better than none).  You can reach me as
<j.beyer@web.de>.

	Yours,
	Joerg