view dynodump/README @ 3063:d30cd499e445

[xemacs-hg @ 2005-11-13 10:48:01 by ben] further error-checking, etc. alloc.c, lrecord.h: Move around the handling of setting of lheader->uid so it's in set_lheader_implementation() -- that way, even non-MC-ALLOC builds get useful uid's in their bare lrecords. Redo related code for strings so the non-ascii count that is stored in the uid isn't hosed. events.c: Save and restore the uid around event zeroing/deadbeefing. lisp.h: Set the correct value of MAX_STRING_ASCII_BEGIN under MC_ALLOC. lisp.h: rearrange the basic code handling ints and chars. basic int stuff goes first, followed by basic char stuff, followed in turn by stuff that mixes ints and chars. this is required since some basic defn's have become inline functions. XCHAR and CHARP have additional error-checking in that they check to make sure that the value in question is not just a character but a valid character (i.e. its numeric value is valid). print.c: debug_p4 now has a useful UID in all cases and uses it; but it also prints the raw header address (previously, you just got one of them). text.h: some basic char defn's that belonged in lisp.h have been moved there. valid_ichar_p() is moved too since the inline functions need it.
author ben
date Sun, 13 Nov 2005 10:48:04 +0000
parents 376386a54a3c
children
line wrap: on
line source

		DYNODUMP
		--------

Dynodump, not to be confused with DinoTurd (as seen in Jurassic Park),
is a shared object that provides one function:

	int dynodump(char *new_file);

dynodump(), called from a running program will write a new executable
in new_file a la unexec() in GNU Emacs.  The difference lies in the
relocations.

dynodump() will create an image with any relocations (which were
performed by the run-time dynamic linker) undone.  This allows the new
image to be run in a different environment.  There is, however, one
potentially major caveat.  If a symbol reference gets updated during
the running of the calling program, its updated value will be lost.
An example (with additional blank lines for legibility):

	$ cat lib.c
	char _foo[] = "hello";
	char _bar[] = "world";

	$ cc -G -o lib.so lib.c

	$ cat prog.c
	extern char _foo, _bar;
 
	int beenhere = 0;
	char * foo = &_foo;
	char * bar = &_bar;
 
	int
	main(void)
	{
	    (void) printf("%d: foo = %x\n", beenhere, foo);
	    (void) printf("%d: bar = %x, ", beenhere, bar);
 
	    if (!beenhere) {
	        beenhere = 1;
	        bar++;
	        dynodump("newfile");
	    }
	    (void) printf("%x\n", bar);
	}

	$ cc -o prog prog.c -R. lib.so dynodump.so

	$ ./prog
	0: foo = ef7503cc
	0: bar = ef7503d2, ef7503d3

	$ ./newfile
	1: foo = ef7503cc
	1: bar = ef7503d2, ef7503d2

Notice that in the run of newfile, bar points at "world" instead of
the perhaps expected "orld".

Dynodump supports sparc, intel, and power pc architectures.

Dynodump is buildable with GNU make and gcc.  If it works for you
with these tools, let me know.

unexec()
--------

Also supplied is an unexsol2.c which belongs in the XEmacs src
directory with the other unex*.c files.  The src/s/sol2.h should be
have the following added to it:

	#ifdef UNEXEC
	#undef UNEXEC
	#endif
	#define UNEXEC unexsol2.o

This unexec() will attempt to dlopen() the dynodump.so to find the
dynodump function.  The default is "../dynodump/dynodump.so" which
should be appropriate for the typical XEmacs build (unless you used
configure to set up separate build and source trees).  You may change
it by setting the DYNODUMP environment variable to the full path of
the dynodump shared object.

Other notes:

If you're using the 4.0 C compiler from Sun, you should arrange to
shut off the incremental linker.  This is done with the -xildoff
option to the compiler.  The common.mk and testsuite/Makefile files
have commentary and ready made lines for uncommenting for this
purpose.

If you're interested in playing with the UltraSPARC specific options
to the aforementioned compiler, then the same makefiles have some
commentary and flags settings to point you in that direction.

Questions:

dynodump() was developed by Rod.Evans@Eng.Sun.COM and
Georg.Nikodym@Canada.Sun.COM.  If you have questions, feel free to ask
them but be aware that Rod, "don't know jack about emacs."