Confusion over uselargefiles.cbu and uselfs.cbu (the first one

[p5sagit/p5-mst-13.2.git] / README.mint

##########################################################################
# *** README.mint
##########################################################################

If you want to build perl yourself on MiNT (or maybe on an Atari without
MiNT) you may want to accept some advice from somebody who already did it...

There was a perl port for Atari ST done by ++jrb bammi@cadence.com.
This port tried very hard to build on non-MiNT-systems.  For the
sake of efficiency I've left this way.  Yet, I haven't removed bammi's
patches but left them intact.  Unfortunately some of the files that
bammi contributed to the perl distribution seem to have vanished?

So, how can you distinguish my patches from bammi's patches?  All of
bammi's stuff is embedded in "#ifdef atarist" preprocessor macros.
My MiNT port uses "#ifdef __MINT__" instead (and unconditionally
undefines "atarist".  If you want to continue on bammi's port, all 
you have to do is to swap the "-D" and "-U" switches for "__MINT__"  
and "atarist" in the variable ccflags.

However, I think that my version will still run on non-MiNT-systems
provided that the user has a Eunuchs-like environment (i.e. the
standard envariables like $PATH, $HOME, ... are set, there is a
POSIX compliant shell in /bin/sh, and...)

Known problems
==============

The problems you may encounter when building perl on your machine
are most probably due to deficiencies in MiNT resp. the Atari
platform in general.

First of all, if you have less than 8 MB of RAM you shouldn't
even try to build Perl yourself.  Better grab a binary pre-compiled
version somewhere.  Even if you have more memory you should take
some care.  Try to run in a fresh environment (without memory
fragmented too much) with as few daemons, accessories, xcontrol
modules etc. as possible.  If you run some AES you should 
consider to start a console based environment instead.

A problem has been reported with sed.  Sed is used to create
some configuration files based on the answers you have given
to the Configure script.  Unfortunately the Perl Configure script
shows sed on MiNT its limits.  I have sed 2.05 with a stacksize
of 64k and I have encountered no problems.  If sed crashes
during your configuration process you should first try to
augment sed's stacksize:

	fixstk 64k /usr/bin/sed

(or similar).  If it still doesn't help you may have a look
which other versions of sed are installed on your system.
If you have a KGMD 1.0 installation you will find three
in /usr/bin.  Have a look there.

Perl has some "mammut" C files.  If gcc reports "internal
compiler error: program cc1 got fatal signal 10" this is very
likely due to a stack overflow in program cc1. Find cc1 
and fix its stack.  I have made good experiences with 

	fixstk 2 cc1

This doesn't establish a stack of 2 Bytes only as you might
think.  It really reserves one half of the available memory
for cc1's stack.  A setting of 1 would reserve the entire
memory for cc1, 3 would reserve three thirds.  You will have
to find out the value that suits to your system yourself.

BTW, cc1 is maybe a little hard to find.  It is generally installed
as
	/usr/local/lib/gcc-lib/<platform>/<gcc-version>/cc1

where <platform> is probably something like "m68k-atari-mint"
and <version> is the gcc version you use (find out with
"gcc --version").  Maybe "gcc-lib" is not installed in
"/usr/local/lib" but "/usr/lib" on your system.

Now run make (maybe "make -k").  If you get a fatal signal 10 
increase cc1's stacksize, if you run out of memory you should 
either decrease the stacksize or follow some more hints:

Perl's building process is very handy on machines with a lot
of virtual memory but may result in a desaster if you are short
of memory.  If gcc fails to compile many source files you should
reduce the optimization.  Grep for "optimize" in the file
config.sh and change the flags.

If only several huge files cause problems (actually it is not a
matter of the file size resp. the amount of code but depends on
the size of the individual funtions) it is useful to bypass
the make program and compile these files directly from the
command line.  For example if you got something like the
following from make:

	CCCMD = gcc -DPERL_CORE ....
	...
	...: virtual memory exhausted

you should hack into the shell:

	gcc -DPERL_CORE ... toke.c

Please note that you have to add the name of the source file
(here toke.c) at the end.

If none of this helps, you're helpless.  Wait for a binary
release.  If you have succeded you may encounter another problem
at the linking process.  If gcc complains that it can't find
some libraries within the perl distribution you probably have
an old linker.  If it complains for example about "file not
found for xxx.olb" you should cd into the directory in
question and

	ln -s libxxx.a xxx.olb

This will fix the problem.

This version (5.00402) of perl has passed most of the tests on my system:

Failed Test  Status Wstat Total Fail  Failed  List of failed
------------------------------------------------------------------------------
io/pipe.t                    10    2  20.00%  7, 9
io/tell.t                    13    1   7.69%  12
lib/complex.t               762   13   1.71%  84-85, 248-251, 257, 272-273,
                                              371, 380, 419-420
lib/io_pipe.t                10    1  10.00%  9
lib/io_tell.t                13    1   7.69%  12
op/magic.t                   30    2   6.67%  29-30
Failed 6/152 test scripts, 96.05% okay. 20/4359 subtests failed, 99.54% okay.

Pipes always cause problems with MiNT, it's actually a surprise that
most of the tests did work.  I've got no idea why the "tell" test failed,
this shouldn't mean too big a problem however.

Most of the failures of lib/complex seem to be harmless, actually errors
far right to the decimal point...  Two failures seem to be serious:
The sign of the results is reversed.  I would say that this is due
to minor bugs in the portable math lib that I compiled perl with.

I haven't bothered very much to find the reason for the failures
with op/magic.t and op/stat.t.  Maybe you'll find it out.

##########################################################################

Another possible problem may arise from the implementation of  the "pwd" 
command.  It happened to add a carriage return and newline to its output 
no matter what the setting of $UNIXMODE is.  This is quite annoying since many
library modules for perl take the output of pwd, chop off the
trailing newline character and then expect to see a valid path in
that.  But the carriage return (last but second character!) isn't
chopped off.  You can either try to patch all library modules (at
the price of performance for the extra transformation) or you can
use my version of pwd that doesn't suffer from this deficiency.

The fixed implementation is in the mint subdirectory.  Running
"Configure" will attempt to build and install it if necessary
(hints/mint.sh will do this work) but you can build and install it
explicitly by:

	cd mint
	make install

This is the fastest solution.

Just in case you want to go the hard way: perl won't even build with a
broken pwd!  You will have to fix the library modules
(ext/POSIX/POSIX.pm, lib/Cwd.pm, lib/pwd.pl) at last after building
miniperl.

A major nuisance of current MiNTLib versions is the implementation
of system() which is far from being POSIX compliant.  A real system()
should fork and then exec /bin/sh with its argument as a command
line to the shell.  The MiNTLib system() however doesn't expect
that every user has a POSIX shell in /bin/sh.  It tries to work
around the problem by forking and exec'ing the first token in its argument
string.  To get a little bit of compliance to POSIX system() it
tries to handle at least redirection ("<" or ">") on its own 
behalf.  

This isn't a good idea since many programs expect that they can
pass a command line to system() that exploits all features of a
POSIX shell.  If you use the MiNTLib version of system() with
perl the Perl function system() will suffer from the same deficiencies.

You will find a fixed version of system() in the mint subdirectory.
You can easily insert this version into your system libc:

	cd mint
	make system.o
	ar r /usr/lib/libc.a
	ranlib /usr/lib/libc.a

If you are suspicious you should either back up your libc before
or extract the original system.o from your libc with 
"ar x /usr/lib/libc.a system.o".  You can then backup the system.o
module somewhere before you succeed.

Anything missing?  Yep, I've almost forgotten...  
No file in this  distribution without a fine saying.  Take this one:

	"From a thief you should learn: (1) to work at night;
	(2) if one cannot gain what one wants in one night to
	try again the next night; (3) to love one's coworkers
	just as thieves love each other; (4) to be willing to
	risk one's life even for a little thing; (5) not to 
	attach too much value to things even though one has 
	risked one's life for them - just as a thief will resell
	a stolen article for a fraction of its real value; 
	(6) to withstand all kinds of beatings and tortures
	but to remain what you are; and (7) to believe your
	work is worthwhile and not be willing to change it."

			-- Rabbi Dov Baer, Maggid of Mezeritch

OK, this was my motto while working on Perl for MiNT, especially rule (1)...

Have fun with Perl!

Guido Flohr
--
mailto:gufl0000@stud.uni-sb.de
http://stud.uni-sb.de/~gufl0000