Hi all,

In arm_init_frag(), when ARM or THUMB are recorded,  it will simply 
return without doing anything. This is not correct in a few cases.

For the following two cases:
case 1:
         nop        ----> MAP_ARM
         .long   0 ---> MAP_DATA
         .align  4 --->
                       ---->(1)
                       ----> start new frag
         .word   0x12345678

(1) paddings here are all treated as data, because the previous state is 
MAP_DATA.

case 2:
         nop       ----> MAP_ARM
         .long   0 ---> MAP_DATA
         .byte   1
         .align  2 ---> (1)
                      ----> (2)
                      ----> start new frag
         .long   0x12345678 ----> (3)

stage 1: while parsing the input file:
(1) arm_init_frag early returns, No state change, no new symbol. (3)No 
state change, no new symbol.
stage 2: Later, while writing the object file:
(2)MAP_DATA, MAP_ARM are inserted by insert_data_mapping_symol to handle 
alignment. (3) interpreted as instruction because of the MAP_ARM 
inserted at (2)


This change might generate redundant symbols, some of them can be 
removed later. In check_mapping_symbols(), the symbols at the end of a 
section, and overlapping symbols will be removed.

gas checked without any new issues. Okay to commit?

By the way, Nick, aarch64 has the same issue. The same can be done to 
aarch64_init_frag(). Another way is to remove 
fragP->tc_frag_data.recorded. It's only used in a few places. But I am 
sure what dose it intended to do, Could you please explain a little bit 
for me?

Regards,
Renlin Li


gas/ChangeLog:

2015-04-24  Renlin Li  <renlin.li@arm.com>

     * config/tc-arm.c (arm_init_frag): Always emit mapping symbols.

gas/testsuite/ChangeLog:

2015-04-24  Renlin Li  <renlin.li@arm.com>

     * gas/arm/thumb2_vpool_be.d: Adjust the desired output.
     * gas/arm/vldconst_be.d: Ditto.