"-g --relax" , debugging information with binutils

[Sumanth Gundapaneni <Sumanth dot Gundapaneni at kpitcummins dot com>]

Hi Swami,
     We have been working on CR16 gcc backend port for some time.
We have posted the gcc patches at following link:-
http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01676.html

We have validated the tool chain against various test suites and 
cornered some bugs . One among them is related to compiling the 
code with "-g "  and linking with  "--relax "  flag. The debug 
information for the above mentioned case was generated using 
readelf and found out the issue was similar to the one discussed 
by you in binutils mailing list at following link:-
http://sourceware.org/ml/binutils/2009-12/msg00099.html

Can you please let us know if there are any improvements in this 
issue?

Please find the explanation below for the particular test case, 
if our understanding was correct .

Case1: (without relaxation)  

?22354:? 00 18 e6 00 ???????? beq??????? *+0x2243a <.L37>:m
?22358:? 12 00 30 b2 ???????? loadd??? 0x20020 <_al+0x4>:l,(r4,r3)?????????? 
?2235c:? 20 00

And the debug information was apt as per the code generated

Case2: (with relaxation)

222b0:? 0c 16?????? ?????????????? beq??????? *+0x22388 <.L37>:s
222b2:? 12 00 30 b2 ???????? loadd??? 0x20020 <_al+0x4>:l,(r4,r3)
222b6: 20 00

It can be observed in the debugging information dump that, the pc 
was incremented from 0x222b0 to 0x222b4 instead of 0x222b2. Though 
the PC values and alignments were taken care, the debugging 
information was broken with relaxation. The debugging information 
generated still assumes the concerned branch was not relaxed. 

Thanks and Regards,
Sumanth G