1#@!#!123s

: / proc / self / root / usr / share / doc / libdwarf-20130207 /
Filename : README
back
To build libdwarf.a, type
        ./configure
        make
To build libdwarf.so, type
        ./configure --enable-shared --disable-nonshared
        make
To build  both, type
        ./configure --enable-shared 
        make

January 30, 2013: libdwarf.h is no longer in the distribution,
but libdwarf.h.in is identical to libdwarf.h.  'configure'
copies libdwarf.h.in to libdwarf.h and whether libelf.h defines
'struct _Elf' or 'struct Elf' configure attempts to create
libdwarf.h appropriately.  

No real install target is provided here, so 'make install' does
not do much. One can copy either or both of libdwarf.a libdwarf.so 
to somewhere fairly standard (but intended for software you build)  
like '/usr/local/lib'.  Or anywhere else you want to copy it.

To use dwarf or libdwarf, you may want to copy dwarf.h and
a generated libdwarf.h somewhere convenient
(possibly /usr/local/include), 
and you may need to copy the libdwarf to a convenient spot 
(/usr/local/lib is a traditional place for libraries one builds
oneself on Unix and Linux).
This copying is not needed to build dwarfdump.


Multi Threading, or using threads with libdwarf (Thread Safety):
  Nothing in libdwarf does any locking.  Every Dwarf_Debug
  (such as returned by dwarf_init()) is fully independent
  of all other Dwarf_Debug-s.  However, calls to libdwarf can
  change a Dwarf_Debug. So it is unsafe to have two different
  threads accessing a single Dwarf_Debug simultaneously.
  It is therefore sufficient to ensure than any one Dwarf_Debug
  is only accessed from a single thread.

Warnings like
 "warning: cast from pointer to integer of different size"
at compile time are to be expected in dwarf_frame.c and
dwarf_frame2.c.   Do not be alarmed.

Warnings like
 "warning: passing argument 1 of ‘dbg->de_callback_func_c’
discards ‘const’ qualifier from pointer target type
[enabled by default]" at compile time are to be expected in
some pro*.c source files.  Fixing the public prototype could
cause some producer-library user's code to fail to compile
so we live with the warnings for now.


If your headers are not in the expected places,
use the make command line to add flags and include directories.
For example
  ./configure  
  PREINCS="-I /usr/local/share/include" POSTINCS="-I /home/x/include"  make
PREINCS content is inserted before CFLAGS as make(1) is running.
POSTINCS content is added after the CFLAGS value.

To set LDFLAGS (which is used when building a .so and
in building gennames to create some source here), 
do so at configure time, for example:
  ./configure LDFLAGS="-L /var/tmp"
Or use PRELIBS and/or POSTLIBS at 'make' time similar to the use
of PREINCS and POSTINCS.

If you are using the old frame interfaces and depend on
the use of DW_FRAME_CFA_COL you must add  --enable-oldframecol
to the ./configure options to configure libdwarf.
See NEWS and libdwarf2.1.mm/pdf .

To generate SGI IRIX 64 bit offsets (in the producer code)
configure with   --enable-dwarf-format-sgi-irix.
To configure with only 32bit offsets (aka DWARF2) configure
with --enable-dwarf-format-strict-32bit.
By default the producer now generates 32bit offsets by default
but one can turn on DWARF3 64bit offset generation at runtime by ORing
DW_DLC_OFFSET_SIZE_64 onto the flags in the call to 
dwarf_producer_init() (or dwarf_producer_init_b) [when the
address size is specified as 64 bit].

Mac OSX (June 2010): Since MacOSX does not use elf, there is no elf.h
header in the headers provided on MacOSX.
Use a search engine (like google) to find an elf.h you can use.
http://www.rockbox.org/tracker/9006?getfile=16683 
might be useful.
In addition, the archive (ar) program on MacOSX does not
automatically generate some data so modify the generated
Makefile to add -s to the options to ar.

To enable dection of Windows pathnames as full paths
add --enable-windowspath.  Doing this does mean things like
A:foo and \anything are treated as full paths (these are
unlikely path names on a POSIX system but are legal 
POSIX partial paths).

It is possible to request a shared library (libdwarf.so) build with
	--enable-shared
To turn off the build of the archive library (libdwarf.a) specify
        --disable-nonshared
but in this case you must specify --enable-shared or nothing will
build!

TARGET DEPENDENCIES of .debug_frame:
dwarf.h
  These should be revised if you have more than the defined
  63 'normal' registers.  It's not harmful to have these too large!
  Too small will lead to errors reading .debug_frame and .eh_frame.
  DW_FRAME_HIGHEST_NORMAL_REGISTER
  DW_FRAME_LAST_REG_NUM

  These you might revise, but can safely ignore if simply
  using dwarfdump.  If using the producer code you will want
  to get these exactly right for your architecture.
  DW_FRAME_RA_COL
  DW_FRAME_STATIC_LINK
  DW_FRAME_CFA_COL 

libdwarf.h
  The DW_FRAME_REG_INITIAL_VALUE #define should be set to
  the value appropriate to your architecture. See libdwarf.h
  for details.

  If DW_REG_TABLE_SIZE is not set large enough attempts to 
  fill in the .debug_frame tables will get an error. 
  Should be at least as large as DW_FRAME_LAST_REG_NUM.
  If it's too large nothing is harmed (but some extra space taken
  at run time).

If your printf does not support C standard %llx etc,
(such as MSWindows with long long), configure
option --enable-nonstandardprintf
and defines like DW_PR_DUx etc in libdwarf.h
provide a way to configure for that relatively easily.


The .debug_frame is so very architecture dependent
and because the host (where libdwarf/dwarfdump are executed)
and target (the objects read) could be different.
It's currently not supported to have dwarfdump/libdwarf determine
the architecture on-the-fly and do-the-right-thing.
Just setting DW_FRAME_LAST_REG_NUM and DW_FRAME_HIGHEST_NORMAL_REGISTER
and DW_REG_TABLE_SIZE high enough will likely suffice for most
purposes and most compilers/architectures..
See comments in dwarf.h/libdwarf.h.

It's perfectly safe to ignore the above suggestions as long
as libdwarf does not get a DW_DLE_DF_REG_NUM_TOO_HIGH error.
(which would only happen on reading .debug_frame or .eh_frame data).

If you intend to use the libdwarf dwarf-producer code
for .debug_frame information
you must do a thorough analysys and revise dwarf.h
substantially to match the output target architecture.

In general, in the producer code, numbers are copied from and
to integers with memcpy().  In case of endianness problems,
constants set in dwarf_producer_init() can fix the problems.
If one wants to produce a *different-endian* output the best
solution is to change the integer memcpy calls to call thru a
new dbg-based function pointer and have it 'do the right thing'
to adjust endianness.  Set the function pointer correctly in
dwarf_producer_init() and the rest of the code will just call
thru the function pointer.  Tedious work to find and change the
memcpy calls to be dbg->de_memcpy(), but once done the code is
no longer endian dependent (right now there is no way to ask
for cross-endian: a new flag needed or ?).

leb128 numbers are endian-independent, so nothing need be
done with those for cross-endian support (the storage
of leb128 on disk is always little-endian).

The .ps files are postscript. So those who cannot deal with mm
format files but do have a postscript printer (or have
ghostscript) can print the documents.  
This form was chosen before pdf format existed...

libdwarf2.1.pdf documents a way for a debugger to read dwarf information.
libdwarf2p.1.pdf documents a way for a compiler to generate dwarf information.
dwarf.v2.pdf documents Dwarf Version 2.
index.v2.pdf is an index to dwarf.v2.ps.
indexDW.v2  is a plain text index of dwarf #defines to dwarf.v2.ps
mips_extensions.ps documents the mips/sgi extensions to dwarf.

See the Makefile for the commands used to build pdf files
libdwarf.2.1.pdf and libdwarf1p.1.pdf.

pic is a picture processing tool (ATT command).
tbl is a table-processing tool.
(part of Documentor's Work Bench on ATT-like systems).
tbl and pic are available on linux.

psroff is a name for a troff-like processor, part of
Documentor's Work Bench on IRIX. Substitute a
troff-like or nroff-like processor (GNU groff works fine).

The index.v2.mm was generated by the dwarf-document writer
using some local ATT/USL tools (which neither SGI 
nor the open-source community generally has, so
there is no way I know of to regenerate this).

To use dwarf or libdwarf, you may want to install dwarf.h and
libdwarf.h somewhere convenient.

You will also need libelf (libelf.a and/or libelf.so) and
libelf.h installed. These are available from GNU repositories
and from the normal Linux repositories for Linux releases.
On Ubuntu Linux for example:
	sudo apt-get install libelf-dev libelf1 

Compiler warnings:
A few Warnings like:
dwarf_frame.c:715:29: warning: cast from pointer to integer of different size [-
Wpointer-to-int-cast]
dwarf_arange.c:113:13: warning: variable ‘local_extension_size’ set but not used
 [-Wunused-but-set-variable]
are considered normal.


As of January 2013 the code compiles with gcc without
problems with -Wall and -Wsign-compare
aside from the warnings hinted at above.

The following gcc/clang options 
have not all been tried as of January 2013,
but will be as time permits.
-Wsystem-headers 
-Wall  -Wsign-compare
-Wno-format-y2k -W 
-Wno-unused-parameter -Wstrict-prototypes 
-Wmissing-prototypes -Wpointer-arith 
-Wreturn-type -Wcast-qual -Wwrite-strings 
-Wswitch -Wshadow -Wunused-parameter 
-Wcast-align -Wchar-subscripts -Winline 
-Wnested-externs -Wredundant-decls 
-Wold-style-definition -Wno-pointer-sign


$Source: /home/davea/dwarf/dwarf-working/trunk/libdwarf/README,v $
$Revision: 1.1 $
$Date: 2009/11/23 17:15:37 $