www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - [Issue 2741] New: debug symbols on OSX

reply d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741

           Summary: debug symbols on OSX
           Product: D
           Version: unspecified
          Platform: PC
        OS/Version: Mac OS X
            Status: NEW
          Severity: normal
          Priority: P2
         Component: DMD
        AssignedTo: bugzilla digitalmars.com
        ReportedBy: bugzilla digitalmars.com


There is a problem with debug symbols, namely source file paths, generated by
DMD on OS X. It seems that gdb cannot find any source files when debugging a
program compiled with DMD's -g (or -gc) switch, with a "no source file named
[whatever]" error, so setting breakpoints is impossible. GDC on OS X does not
have this problem. Some googling around has led me to discover that GNU gcc on
linux (not Apple gcc) also had this problem a few releases ago until it was
patched, caused by something to do with the pathnames used in the debug symbols
it generated, though I'm not sure of the technical details.


-- 
Mar 17 2009
next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741





------- Comment #1 from captnjameskirk yahoo.com  2009-03-18 11:10 -------
A few more details if it'll help. It appears that no symbol table is loaded at
all on Leopard, whether using the gdb supplied with XCode 2.5 (gdb-696) or
XCode 3.1 (gdb-962). On Tiger (gdb-696 only), the situation is a little better,
but not much. Using the gdb command "info sources" on a simple Tango "hello"
program lists the following (comiled with "dmd -g") on Tiger, which lists no D
source files at all:
/System/Library/Frameworks/System.framework/PrivateHeaders/i386/cpu_capabilities.h,
/System/Library/Frameworks/System.framework/PrivateHeaders/machine/cpu_capabilities.h,
<command line>, <built-in>, /SourceCache/Libsystem/Libsystem-88.3.10//,
CommPageSymbols.st, {standard input}


Now, the same file compiled with GDC shows the following sources, with both
gdb-696 and gdb-962 on Leopard, and gdb-696 on Tiger (although the order is
different):
/System/Library/Frameworks/System.framework/PrivateHeaders/i386/cpu_capabilities.h,
/System/Library/Frameworks/System.framework/PrivateHeaders/machine/cpu_capabilities.h,
<command line>, <built-in>, /SourceCache/Libsystem/Libsystem-88.3.10//,
CommPageSymbols.st, {standard input}, ../shared/rt/typeinfo/ti_C.d, critical.c,
arraycast.d, ../shared/rt/util/string.d, ../shared/rt/util/hash.d,
../shared/rt/util/console.d, ../shared/rt/typeinfo/ti_wchar.d,
../shared/rt/typeinfo/ti_void.d, ../shared/rt/typeinfo/ti_ushort.d,
../shared/rt/typeinfo/ti_uint.d, ../shared/rt/typeinfo/ti_ubyte.d,
../shared/rt/typeinfo/ti_short.d, ../shared/rt/typeinfo/ti_int.d,
../shared/rt/typeinfo/ti_dchar.d, ../shared/rt/typeinfo/ti_char.d,
../shared/rt/typeinfo/ti_byte.d, ../shared/rt/typeinfo/ti_Ashort.d,
../shared/rt/typeinfo/ti_Aint.d, ../shared/rt/typeinfo/ti_Ag.d,
../shared/rt/typeinfo/ti_AC.d, monitor.c, memory_dyld.c, memory.d,
/Users/obijohn/projects/d/tango/std/c/stdarg.di, lifetime.d, genobj.d,
gcc/unwind.d, gcc/deh.d, dgccmain2.d, cmain.d, cast.d, aaA.d,
/Users/obijohn/projects/d/hello.d


-- 
Mar 18 2009
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741


Leandro Lucarella <llucax gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |llucax gmail.com
            Summary|debug symbols on OSX        |gdb: debug symbols on OSX
           Severity|normal                      |regression


-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Oct 12 2009
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741


Sean Kelly <sean invisibleduck.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sean invisibleduck.org


--- Comment #2 from Sean Kelly <sean invisibleduck.org> 2011-01-12 15:03:44 PST
---
I've been digging into this a bit, and the problem appears to be Mach-O
related.  From the ABI docs:

"When using the Stabs debugging format, the symbol table also holds debugging
information. When using DWARF, debugging information is stored in the image’s
corresponding dSYM file, specified by the uuid_command structure."

The standard location for the debug symbol file is:

"appname.dSYM/Contents/Resources/DWARF/appname"

And I've verified that at least some apps (like Valgrind) won't even look for
this file if a uuid_command structure isn't present in the executable, and they
won't use it if the UUIDs don't match between the two files.  I'm still sorting
out how the debug info file is structured though.  All I know so far is that it
isn't simply a copy of the executable.  Perhaps the file only contains the
debug sections and no code?

Using nm to compare the symbols in a D vs. a C app, the only notable difference
is that the C app contains entries pointing to the executable path and the
object file:

00000000 - 00 0000    SO /Users/sean/test/
4d2e279c - 00 0001   OSO /var/folders/blah/-Tmp-//blah.o

At the moment, I'm hoping that the only issue is that DMD current'y doesn't
generate the uuid_command block, and that adding a uuid_command and duplicating
the executable into the proper dSYM location will be enough to make debug info
work.  More experimentation to follow.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Jan 12 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741


Jacob Carlborg <doob me.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |doob me.com


--- Comment #3 from Jacob Carlborg <doob me.com> 2011-01-13 04:31:33 PST ---
(In reply to comment #2)
 At the moment, I'm hoping that the only issue is that DMD current'y doesn't
 generate the uuid_command block, and that adding a uuid_command and duplicating
 the executable into the proper dSYM location will be enough to make debug info
 work.  More experimentation to follow.

Probably not due to: http://d.puremagic.com/issues/show_bug.cgi?id=4154 -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email ------- You are receiving this mail because: -------
Jan 13 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741



--- Comment #4 from Sean Kelly <sean invisibleduck.org> 2011-01-13 08:18:14 PST
---
I added the uuid_command and changed the section names yesterday and it still
didn't work.  Not sure yet what else is needed.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Jan 13 2011
prev sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=2741



--- Comment #5 from Sean Kelly <sean invisibleduck.org> 2011-01-13 08:20:39 PST
---
Ah, I didn't notice the offset problem after changing the section names. 
That's next I suppose.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Jan 13 2011